class Aws::Redshift::Client
An API client for Redshift
. To construct a client, you need to configure a `:region` and `:credentials`.
client = Aws::Redshift::Client.new( region: region_name, credentials: credentials, # ... )
For details on configuring region and credentials see the [developer guide](/sdk-for-ruby/v3/developer-guide/setup-config.html).
See {#initialize} for a full list of supported configuration options.
Attributes
@api private
Public Class Methods
@overload initialize(options)
@param [Hash] options @option options [required, Aws::CredentialProvider] :credentials Your AWS credentials. This can be an instance of any one of the following classes: * `Aws::Credentials` - Used for configuring static, non-refreshing credentials. * `Aws::SharedCredentials` - Used for loading static credentials from a shared file, such as `~/.aws/config`. * `Aws::AssumeRoleCredentials` - Used when you need to assume a role. * `Aws::AssumeRoleWebIdentityCredentials` - Used when you need to assume a role after providing credentials via the web. * `Aws::SSOCredentials` - Used for loading credentials from AWS SSO using an access token generated from `aws login`. * `Aws::ProcessCredentials` - Used for loading credentials from a process that outputs to stdout. * `Aws::InstanceProfileCredentials` - Used for loading credentials from an EC2 IMDS on an EC2 instance. * `Aws::ECSCredentials` - Used for loading credentials from instances running in ECS. * `Aws::CognitoIdentityCredentials` - Used for loading credentials from the Cognito Identity service. When `:credentials` are not configured directly, the following locations will be searched for credentials: * `Aws.config[:credentials]` * The `:access_key_id`, `:secret_access_key`, and `:session_token` options. * ENV['AWS_ACCESS_KEY_ID'], ENV['AWS_SECRET_ACCESS_KEY'] * `~/.aws/credentials` * `~/.aws/config` * EC2/ECS IMDS instance profile - When used by default, the timeouts are very aggressive. Construct and pass an instance of `Aws::InstanceProfileCredentails` or `Aws::ECSCredentials` to enable retries and extended timeouts. @option options [required, String] :region The AWS region to connect to. The configured `:region` is used to determine the service `:endpoint`. When not passed, a default `:region` is searched for in the following locations: * `Aws.config[:region]` * `ENV['AWS_REGION']` * `ENV['AMAZON_REGION']` * `ENV['AWS_DEFAULT_REGION']` * `~/.aws/credentials` * `~/.aws/config` @option options [String] :access_key_id @option options [Boolean] :active_endpoint_cache (false) When set to `true`, a thread polling for endpoints will be running in the background every 60 secs (default). Defaults to `false`. @option options [Boolean] :adaptive_retry_wait_to_fill (true) Used only in `adaptive` retry mode. When true, the request will sleep until there is sufficent client side capacity to retry the request. When false, the request will raise a `RetryCapacityNotAvailableError` and will not retry instead of sleeping. @option options [Boolean] :client_side_monitoring (false) When `true`, client-side metrics will be collected for all API requests from this client. @option options [String] :client_side_monitoring_client_id ("") Allows you to provide an identifier for this client which will be attached to all generated client side metrics. Defaults to an empty string. @option options [String] :client_side_monitoring_host ("127.0.0.1") Allows you to specify the DNS hostname or IPv4 or IPv6 address that the client side monitoring agent is running on, where client metrics will be published via UDP. @option options [Integer] :client_side_monitoring_port (31000) Required for publishing client metrics. The port that the client side monitoring agent is running on, where client metrics will be published via UDP. @option options [Aws::ClientSideMonitoring::Publisher] :client_side_monitoring_publisher (Aws::ClientSideMonitoring::Publisher) Allows you to provide a custom client-side monitoring publisher class. By default, will use the Client Side Monitoring Agent Publisher. @option options [Boolean] :convert_params (true) When `true`, an attempt is made to coerce request parameters into the required types. @option options [Boolean] :correct_clock_skew (true) Used only in `standard` and adaptive retry modes. Specifies whether to apply a clock skew correction and retry requests with skewed client clocks. @option options [Boolean] :disable_host_prefix_injection (false) Set to true to disable SDK automatically adding host prefix to default service endpoint when available. @option options [String] :endpoint The client endpoint is normally constructed from the `:region` option. You should only configure an `:endpoint` when connecting to test or custom endpoints. This should be a valid HTTP(S) URI. @option options [Integer] :endpoint_cache_max_entries (1000) Used for the maximum size limit of the LRU cache storing endpoints data for endpoint discovery enabled operations. Defaults to 1000. @option options [Integer] :endpoint_cache_max_threads (10) Used for the maximum threads in use for polling endpoints to be cached, defaults to 10. @option options [Integer] :endpoint_cache_poll_interval (60) When :endpoint_discovery and :active_endpoint_cache is enabled, Use this option to config the time interval in seconds for making requests fetching endpoints information. Defaults to 60 sec. @option options [Boolean] :endpoint_discovery (false) When set to `true`, endpoint discovery will be enabled for operations when available. @option options [Aws::Log::Formatter] :log_formatter (Aws::Log::Formatter.default) The log formatter. @option options [Symbol] :log_level (:info) The log level to send messages to the `:logger` at. @option options [Logger] :logger The Logger instance to send log messages to. If this option is not set, logging will be disabled. @option options [Integer] :max_attempts (3) An integer representing the maximum number attempts that will be made for a single request, including the initial attempt. For example, setting this value to 5 will result in a request being retried up to 4 times. Used in `standard` and `adaptive` retry modes. @option options [String] :profile ("default") Used when loading credentials from the shared credentials file at HOME/.aws/credentials. When not specified, 'default' is used. @option options [Proc] :retry_backoff A proc or lambda used for backoff. Defaults to 2**retries * retry_base_delay. This option is only used in the `legacy` retry mode. @option options [Float] :retry_base_delay (0.3) The base delay in seconds used by the default backoff function. This option is only used in the `legacy` retry mode. @option options [Symbol] :retry_jitter (:none) A delay randomiser function used by the default backoff function. Some predefined functions can be referenced by name - :none, :equal, :full, otherwise a Proc that takes and returns a number. This option is only used in the `legacy` retry mode. @see https://www.awsarchitectureblog.com/2015/03/backoff.html @option options [Integer] :retry_limit (3) The maximum number of times to retry failed requests. Only ~ 500 level server errors and certain ~ 400 level client errors are retried. Generally, these are throttling errors, data checksum errors, networking errors, timeout errors, auth errors, endpoint discovery, and errors from expired credentials. This option is only used in the `legacy` retry mode. @option options [Integer] :retry_max_delay (0) The maximum number of seconds to delay between retries (0 for no limit) used by the default backoff function. This option is only used in the `legacy` retry mode. @option options [String] :retry_mode ("legacy") Specifies which retry algorithm to use. Values are: * `legacy` - The pre-existing retry behavior. This is default value if no retry mode is provided. * `standard` - A standardized set of retry rules across the AWS SDKs. This includes support for retry quotas, which limit the number of unsuccessful retries a client can make. * `adaptive` - An experimental retry mode that includes all the functionality of `standard` mode along with automatic client side throttling. This is a provisional mode that may change behavior in the future. @option options [String] :secret_access_key @option options [String] :session_token @option options [Boolean] :stub_responses (false) Causes the client to return stubbed responses. By default fake responses are generated and returned. You can specify the response data to return or errors to raise by calling {ClientStubs#stub_responses}. See {ClientStubs} for more information. ** Please note ** When response stubbing is enabled, no HTTP requests are made, and retries are disabled. @option options [Boolean] :validate_params (true) When `true`, request parameters are validated before sending the request. @option options [URI::HTTP,String] :http_proxy A proxy to send requests through. Formatted like 'http://proxy.com:123'. @option options [Float] :http_open_timeout (15) The number of seconds to wait when opening a HTTP session before raising a `Timeout::Error`. @option options [Integer] :http_read_timeout (60) The default number of seconds to wait for response data. This value can safely be set per-request on the session. @option options [Float] :http_idle_timeout (5) The number of seconds a connection is allowed to sit idle before it is considered stale. Stale connections are closed and removed from the pool before making a request. @option options [Float] :http_continue_timeout (1) The number of seconds to wait for a 100-continue response before sending the request body. This option has no effect unless the request has "Expect" header set to "100-continue". Defaults to `nil` which disables this behaviour. This value can safely be set per request on the session. @option options [Boolean] :http_wire_trace (false) When `true`, HTTP debug output will be sent to the `:logger`. @option options [Boolean] :ssl_verify_peer (true) When `true`, SSL peer certificates are verified when establishing a connection. @option options [String] :ssl_ca_bundle Full path to the SSL certificate authority bundle file that should be used when verifying peer certificates. If you do not pass `:ssl_ca_bundle` or `:ssl_ca_directory` the the system default will be used if available. @option options [String] :ssl_ca_directory Full path of the directory that contains the unbundled SSL certificate authority files for verifying peer certificates. If you do not pass `:ssl_ca_bundle` or `:ssl_ca_directory` the the system default will be used if available.
# File lib/aws-sdk-redshift/client.rb, line 324 def initialize(*args) super end
Private Class Methods
@api private
# File lib/aws-sdk-redshift/client.rb, line 10732 def errors_module Errors end
Public Instance Methods
Exchanges a DC1 Reserved Node for a DC2 Reserved Node with no changes to the configuration (term, payment type, or number of nodes) and no additional costs.
@option params [required, String] :reserved_node_id
A string representing the node identifier of the DC1 Reserved Node to be exchanged.
@option params [required, String] :target_reserved_node_offering_id
The unique identifier of the DC2 Reserved Node offering to be used for the exchange. You can obtain the value for the parameter by calling GetReservedNodeExchangeOfferings
@return [Types::AcceptReservedNodeExchangeOutputMessage] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::AcceptReservedNodeExchangeOutputMessage#exchanged_reserved_node #exchanged_reserved_node} => Types::ReservedNode
@example Request syntax with placeholder values
resp = client.accept_reserved_node_exchange({ reserved_node_id: "String", # required target_reserved_node_offering_id: "String", # required })
@example Response structure
resp.exchanged_reserved_node.reserved_node_id #=> String resp.exchanged_reserved_node.reserved_node_offering_id #=> String resp.exchanged_reserved_node.node_type #=> String resp.exchanged_reserved_node.start_time #=> Time resp.exchanged_reserved_node.duration #=> Integer resp.exchanged_reserved_node.fixed_price #=> Float resp.exchanged_reserved_node.usage_price #=> Float resp.exchanged_reserved_node.currency_code #=> String resp.exchanged_reserved_node.node_count #=> Integer resp.exchanged_reserved_node.state #=> String resp.exchanged_reserved_node.offering_type #=> String resp.exchanged_reserved_node.recurring_charges #=> Array resp.exchanged_reserved_node.recurring_charges[0].recurring_charge_amount #=> Float resp.exchanged_reserved_node.recurring_charges[0].recurring_charge_frequency #=> String resp.exchanged_reserved_node.reserved_node_offering_type #=> String, one of "Regular", "Upgradable"
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/AcceptReservedNodeExchange AWS API Documentation
@overload accept_reserved_node_exchange
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 376 def accept_reserved_node_exchange(params = {}, options = {}) req = build_request(:accept_reserved_node_exchange, params) req.send_request(options) end
Adds a partner integration to a cluster. This operation authorizes a partner to push status updates for the specified database. To complete the integration, you also set up the integration on the partner website.
@option params [required, String] :account_id
The Amazon Web Services account ID that owns the cluster.
@option params [required, String] :cluster_identifier
The cluster identifier of the cluster that receives data from the partner.
@option params [required, String] :database_name
The name of the database that receives data from the partner.
@option params [required, String] :partner_name
The name of the partner that is authorized to send data.
@return [Types::PartnerIntegrationOutputMessage] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::PartnerIntegrationOutputMessage#database_name #database_name} => String * {Types::PartnerIntegrationOutputMessage#partner_name #partner_name} => String
@example Request syntax with placeholder values
resp = client.add_partner({ account_id: "PartnerIntegrationAccountId", # required cluster_identifier: "PartnerIntegrationClusterIdentifier", # required database_name: "PartnerIntegrationDatabaseName", # required partner_name: "PartnerIntegrationPartnerName", # required })
@example Response structure
resp.database_name #=> String resp.partner_name #=> String
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/AddPartner AWS API Documentation
@overload add_partner
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 422 def add_partner(params = {}, options = {}) req = build_request(:add_partner, params) req.send_request(options) end
Deletes a set of cluster snapshots.
@option params [required, Array<Types::DeleteClusterSnapshotMessage>] :identifiers
A list of identifiers for the snapshots that you want to delete.
@return [Types::BatchDeleteClusterSnapshotsResult] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::BatchDeleteClusterSnapshotsResult#resources #resources} => Array<String> * {Types::BatchDeleteClusterSnapshotsResult#errors #errors} => Array<Types::SnapshotErrorMessage>
@example Request syntax with placeholder values
resp = client.batch_delete_cluster_snapshots({ identifiers: [ # required { snapshot_identifier: "String", # required snapshot_cluster_identifier: "String", }, ], })
@example Response structure
resp.resources #=> Array resp.resources[0] #=> String resp.errors #=> Array resp.errors[0].snapshot_identifier #=> String resp.errors[0].snapshot_cluster_identifier #=> String resp.errors[0].failure_code #=> String resp.errors[0].failure_reason #=> String
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/BatchDeleteClusterSnapshots AWS API Documentation
@overload batch_delete_cluster_snapshots
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 790 def batch_delete_cluster_snapshots(params = {}, options = {}) req = build_request(:batch_delete_cluster_snapshots, params) req.send_request(options) end
Modifies the settings for a set of cluster snapshots.
@option params [required, Array<String>] :snapshot_identifier_list
A list of snapshot identifiers you want to modify.
@option params [Integer] :manual_snapshot_retention_period
The number of days that a manual snapshot is retained. If you specify the value -1, the manual snapshot is retained indefinitely. The number must be either -1 or an integer between 1 and 3,653. If you decrease the manual snapshot retention period from its current value, existing manual snapshots that fall outside of the new retention period will return an error. If you want to suppress the errors and delete the snapshots, use the force option.
@option params [Boolean] :force
A boolean value indicating whether to override an exception if the retention period has passed.
@return [Types::BatchModifyClusterSnapshotsOutputMessage] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::BatchModifyClusterSnapshotsOutputMessage#resources #resources} => Array<String> * {Types::BatchModifyClusterSnapshotsOutputMessage#errors #errors} => Array<Types::SnapshotErrorMessage>
@example Request syntax with placeholder values
resp = client.batch_modify_cluster_snapshots({ snapshot_identifier_list: ["String"], # required manual_snapshot_retention_period: 1, force: false, })
@example Response structure
resp.resources #=> Array resp.resources[0] #=> String resp.errors #=> Array resp.errors[0].snapshot_identifier #=> String resp.errors[0].snapshot_cluster_identifier #=> String resp.errors[0].failure_code #=> String resp.errors[0].failure_reason #=> String
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/BatchModifyClusterSnapshots AWS API Documentation
@overload batch_modify_cluster_snapshots
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 842 def batch_modify_cluster_snapshots(params = {}, options = {}) req = build_request(:batch_modify_cluster_snapshots, params) req.send_request(options) end
@param params ({}) @api private
# File lib/aws-sdk-redshift/client.rb, line 10588 def build_request(operation_name, params = {}) handlers = @handlers.for(operation_name) context = Seahorse::Client::RequestContext.new( operation_name: operation_name, operation: config.api.operation(operation_name), client: self, params: params, config: config) context[:gem_name] = 'aws-sdk-redshift' context[:gem_version] = '1.69.0' Seahorse::Client::Request.new(handlers, context) end
Cancels a resize operation for a cluster.
@option params [required, String] :cluster_identifier
The unique identifier for the cluster that you want to cancel a resize operation for.
@return [Types::ResizeProgressMessage] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::ResizeProgressMessage#target_node_type #target_node_type} => String * {Types::ResizeProgressMessage#target_number_of_nodes #target_number_of_nodes} => Integer * {Types::ResizeProgressMessage#target_cluster_type #target_cluster_type} => String * {Types::ResizeProgressMessage#status #status} => String * {Types::ResizeProgressMessage#import_tables_completed #import_tables_completed} => Array<String> * {Types::ResizeProgressMessage#import_tables_in_progress #import_tables_in_progress} => Array<String> * {Types::ResizeProgressMessage#import_tables_not_started #import_tables_not_started} => Array<String> * {Types::ResizeProgressMessage#avg_resize_rate_in_mega_bytes_per_second #avg_resize_rate_in_mega_bytes_per_second} => Float * {Types::ResizeProgressMessage#total_resize_data_in_mega_bytes #total_resize_data_in_mega_bytes} => Integer * {Types::ResizeProgressMessage#progress_in_mega_bytes #progress_in_mega_bytes} => Integer * {Types::ResizeProgressMessage#elapsed_time_in_seconds #elapsed_time_in_seconds} => Integer * {Types::ResizeProgressMessage#estimated_time_to_completion_in_seconds #estimated_time_to_completion_in_seconds} => Integer * {Types::ResizeProgressMessage#resize_type #resize_type} => String * {Types::ResizeProgressMessage#message #message} => String * {Types::ResizeProgressMessage#target_encryption_type #target_encryption_type} => String * {Types::ResizeProgressMessage#data_transfer_progress_percent #data_transfer_progress_percent} => Float
@example Request syntax with placeholder values
resp = client.cancel_resize({ cluster_identifier: "String", # required })
@example Response structure
resp.target_node_type #=> String resp.target_number_of_nodes #=> Integer resp.target_cluster_type #=> String resp.status #=> String resp.import_tables_completed #=> Array resp.import_tables_completed[0] #=> String resp.import_tables_in_progress #=> Array resp.import_tables_in_progress[0] #=> String resp.import_tables_not_started #=> Array resp.import_tables_not_started[0] #=> String resp.avg_resize_rate_in_mega_bytes_per_second #=> Float resp.total_resize_data_in_mega_bytes #=> Integer resp.progress_in_mega_bytes #=> Integer resp.elapsed_time_in_seconds #=> Integer resp.estimated_time_to_completion_in_seconds #=> Integer resp.resize_type #=> String resp.message #=> String resp.target_encryption_type #=> String resp.data_transfer_progress_percent #=> Float
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/CancelResize AWS API Documentation
@overload cancel_resize
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 904 def cancel_resize(params = {}, options = {}) req = build_request(:cancel_resize, params) req.send_request(options) end
Copies the specified automated cluster snapshot to a new manual cluster snapshot. The source must be an automated snapshot and it must be in the available state.
When you delete a cluster, Amazon Redshift
deletes any automated snapshots of the cluster. Also, when the retention period of the snapshot expires, Amazon Redshift
automatically deletes it. If you want to keep an automated snapshot for a longer period, you can make a manual copy of the snapshot. Manual snapshots are retained until you delete them.
For more information about working with snapshots, go to [Amazon Redshift
Snapshots] in the *Amazon Redshift
Cluster Management Guide*.
[1]: docs.aws.amazon.com/redshift/latest/mgmt/working-with-snapshots.html
@option params [required, String] :source_snapshot_identifier
The identifier for the source snapshot. Constraints: * Must be the identifier for a valid automated snapshot whose state is `available`. ^
@option params [String] :source_snapshot_cluster_identifier
The identifier of the cluster the source snapshot was created from. This parameter is required if your IAM user has a policy containing a snapshot resource element that specifies anything other than * for the cluster name. Constraints: * Must be the identifier for a valid cluster. ^
@option params [required, String] :target_snapshot_identifier
The identifier given to the new manual snapshot. Constraints: * Cannot be null, empty, or blank. * Must contain from 1 to 255 alphanumeric characters or hyphens. * First character must be a letter. * Cannot end with a hyphen or contain two consecutive hyphens. * Must be unique for the Amazon Web Services account that is making the request.
@option params [Integer] :manual_snapshot_retention_period
The number of days that a manual snapshot is retained. If the value is -1, the manual snapshot is retained indefinitely. The value must be either -1 or an integer between 1 and 3,653. The default value is -1.
@return [Types::CopyClusterSnapshotResult] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::CopyClusterSnapshotResult#snapshot #snapshot} => Types::Snapshot
@example Request syntax with placeholder values
resp = client.copy_cluster_snapshot({ source_snapshot_identifier: "String", # required source_snapshot_cluster_identifier: "String", target_snapshot_identifier: "String", # required manual_snapshot_retention_period: 1, })
@example Response structure
resp.snapshot.snapshot_identifier #=> String resp.snapshot.cluster_identifier #=> String resp.snapshot.snapshot_create_time #=> Time resp.snapshot.status #=> String resp.snapshot.port #=> Integer resp.snapshot.availability_zone #=> String resp.snapshot.cluster_create_time #=> Time resp.snapshot.master_username #=> String resp.snapshot.cluster_version #=> String resp.snapshot.engine_full_version #=> String resp.snapshot.snapshot_type #=> String resp.snapshot.node_type #=> String resp.snapshot.number_of_nodes #=> Integer resp.snapshot.db_name #=> String resp.snapshot.vpc_id #=> String resp.snapshot.encrypted #=> Boolean resp.snapshot.kms_key_id #=> String resp.snapshot.encrypted_with_hsm #=> Boolean resp.snapshot.accounts_with_restore_access #=> Array resp.snapshot.accounts_with_restore_access[0].account_id #=> String resp.snapshot.accounts_with_restore_access[0].account_alias #=> String resp.snapshot.owner_account #=> String resp.snapshot.total_backup_size_in_mega_bytes #=> Float resp.snapshot.actual_incremental_backup_size_in_mega_bytes #=> Float resp.snapshot.backup_progress_in_mega_bytes #=> Float resp.snapshot.current_backup_rate_in_mega_bytes_per_second #=> Float resp.snapshot.estimated_seconds_to_completion #=> Integer resp.snapshot.elapsed_time_in_seconds #=> Integer resp.snapshot.source_region #=> String resp.snapshot.tags #=> Array resp.snapshot.tags[0].key #=> String resp.snapshot.tags[0].value #=> String resp.snapshot.restorable_node_types #=> Array resp.snapshot.restorable_node_types[0] #=> String resp.snapshot.enhanced_vpc_routing #=> Boolean resp.snapshot.maintenance_track_name #=> String resp.snapshot.manual_snapshot_retention_period #=> Integer resp.snapshot.manual_snapshot_remaining_days #=> Integer resp.snapshot.snapshot_retention_start_time #=> Time
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/CopyClusterSnapshot AWS API Documentation
@overload copy_cluster_snapshot
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 1033 def copy_cluster_snapshot(params = {}, options = {}) req = build_request(:copy_cluster_snapshot, params) req.send_request(options) end
Creates an authentication profile with the specified parameters.
@option params [required, String] :authentication_profile_name
The name of the authentication profile to be created.
@option params [required, String] :authentication_profile_content
The content of the authentication profile in JSON format. The maximum length of the JSON string is determined by a quota for your account.
@return [Types::CreateAuthenticationProfileResult] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::CreateAuthenticationProfileResult#authentication_profile_name #authentication_profile_name} => String * {Types::CreateAuthenticationProfileResult#authentication_profile_content #authentication_profile_content} => String
@example Request syntax with placeholder values
resp = client.create_authentication_profile({ authentication_profile_name: "AuthenticationProfileNameString", # required authentication_profile_content: "String", # required })
@example Response structure
resp.authentication_profile_name #=> String resp.authentication_profile_content #=> String
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/CreateAuthenticationProfile AWS API Documentation
@overload create_authentication_profile
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 1068 def create_authentication_profile(params = {}, options = {}) req = build_request(:create_authentication_profile, params) req.send_request(options) end
Creates a new cluster with the specified parameters.
To create a cluster in Virtual Private Cloud (VPC), you must provide a cluster subnet group name. The cluster subnet group identifies the subnets of your VPC that Amazon Redshift
uses when creating the cluster. For more information about managing clusters, go to [Amazon Redshift
Clusters] in the *Amazon Redshift
Cluster Management Guide*.
[1]: docs.aws.amazon.com/redshift/latest/mgmt/working-with-clusters.html
@option params [String] :db_name
The name of the first database to be created when the cluster is created. To create additional databases after the cluster is created, connect to the cluster with a SQL client and use SQL commands to create a database. For more information, go to [Create a Database][1] in the Amazon Redshift Database Developer Guide. Default: `dev` Constraints: * Must contain 1 to 64 alphanumeric characters. * Must contain only lowercase letters. * Cannot be a word that is reserved by the service. A list of reserved words can be found in [Reserved Words][2] in the Amazon Redshift Database Developer Guide. [1]: https://docs.aws.amazon.com/redshift/latest/dg/t_creating_database.html [2]: https://docs.aws.amazon.com/redshift/latest/dg/r_pg_keywords.html
@option params [required, String] :cluster_identifier
A unique identifier for the cluster. You use this identifier to refer to the cluster for any subsequent cluster operations such as deleting or modifying. The identifier also appears in the Amazon Redshift console. Constraints: * Must contain from 1 to 63 alphanumeric characters or hyphens. * Alphabetic characters must be lowercase. * First character must be a letter. * Cannot end with a hyphen or contain two consecutive hyphens. * Must be unique for all clusters within an Amazon Web Services account. Example: `myexamplecluster`
@option params [String] :cluster_type
The type of the cluster. When cluster type is specified as * `single-node`, the **NumberOfNodes** parameter is not required. * `multi-node`, the **NumberOfNodes** parameter is required. Valid Values: `multi-node` \| `single-node` Default: `multi-node`
@option params [required, String] :node_type
The node type to be provisioned for the cluster. For information about node types, go to [ Working with Clusters][1] in the *Amazon Redshift Cluster Management Guide*. Valid Values: `ds2.xlarge` \| `ds2.8xlarge` \| `dc1.large` \| `dc1.8xlarge` \| `dc2.large` \| `dc2.8xlarge` \| `ra3.xlplus` \| `ra3.4xlarge` \| `ra3.16xlarge` [1]: https://docs.aws.amazon.com/redshift/latest/mgmt/working-with-clusters.html#how-many-nodes
@option params [required, String] :master_username
The user name associated with the admin user account for the cluster that is being created. Constraints: * Must be 1 - 128 alphanumeric characters. The user name can't be `PUBLIC`. * First character must be a letter. * Cannot be a reserved word. A list of reserved words can be found in [Reserved Words][1] in the Amazon Redshift Database Developer Guide. [1]: https://docs.aws.amazon.com/redshift/latest/dg/r_pg_keywords.html
@option params [required, String] :master_user_password
The password associated with the admin user account for the cluster that is being created. Constraints: * Must be between 8 and 64 characters in length. * Must contain at least one uppercase letter. * Must contain at least one lowercase letter. * Must contain one number. * Can be any printable ASCII character (ASCII code 33 to 126) except ' (single quote), " (double quote), \\, /, @, or space.
@option params [Array<String>] :cluster_security_groups
A list of security groups to be associated with this cluster. Default: The default cluster security group for Amazon Redshift.
@option params [Array<String>] :vpc_security_group_ids
A list of Virtual Private Cloud (VPC) security groups to be associated with the cluster. Default: The default VPC security group is associated with the cluster.
@option params [String] :cluster_subnet_group_name
The name of a cluster subnet group to be associated with this cluster. If this parameter is not provided the resulting cluster will be deployed outside virtual private cloud (VPC).
@option params [String] :availability_zone
The EC2 Availability Zone (AZ) in which you want Amazon Redshift to provision the cluster. For example, if you have several EC2 instances running in a specific Availability Zone, then you might want the cluster to be provisioned in the same zone in order to decrease network latency. Default: A random, system-chosen Availability Zone in the region that is specified by the endpoint. Example: `us-east-2d` Constraint: The specified Availability Zone must be in the same region as the current endpoint.
@option params [String] :preferred_maintenance_window
The weekly time range (in UTC) during which automated cluster maintenance can occur. Format: `ddd:hh24:mi-ddd:hh24:mi` Default: A 30-minute window selected at random from an 8-hour block of time per region, occurring on a random day of the week. For more information about the time blocks for each region, see [Maintenance Windows][1] in Amazon Redshift Cluster Management Guide. Valid Days: Mon \| Tue \| Wed \| Thu \| Fri \| Sat \| Sun Constraints: Minimum 30-minute window. [1]: https://docs.aws.amazon.com/redshift/latest/mgmt/working-with-clusters.html#rs-maintenance-windows
@option params [String] :cluster_parameter_group_name
The name of the parameter group to be associated with this cluster. Default: The default Amazon Redshift cluster parameter group. For information about the default parameter group, go to [Working with Amazon Redshift Parameter Groups][1] Constraints: * Must be 1 to 255 alphanumeric characters or hyphens. * First character must be a letter. * Cannot end with a hyphen or contain two consecutive hyphens. [1]: https://docs.aws.amazon.com/redshift/latest/mgmt/working-with-parameter-groups.html
@option params [Integer] :automated_snapshot_retention_period
The number of days that automated snapshots are retained. If the value is 0, automated snapshots are disabled. Even if automated snapshots are disabled, you can still create manual snapshots when you want with CreateClusterSnapshot. You can't disable automated snapshots for RA3 node types. Set the automated retention period from 1-35 days. Default: `1` Constraints: Must be a value from 0 to 35.
@option params [Integer] :manual_snapshot_retention_period
The default number of days to retain a manual snapshot. If the value is -1, the snapshot is retained indefinitely. This setting doesn't change the retention period of existing snapshots. The value must be either -1 or an integer between 1 and 3,653.
@option params [Integer] :port
The port number on which the cluster accepts incoming connections. The cluster is accessible only via the JDBC and ODBC connection strings. Part of the connection string requires the port on which the cluster will listen for incoming connections. Default: `5439` Valid Values: `1150-65535`
@option params [String] :cluster_version
The version of the Amazon Redshift engine software that you want to deploy on the cluster. The version selected runs on all the nodes in the cluster. Constraints: Only version 1.0 is currently available. Example: `1.0`
@option params [Boolean] :allow_version_upgrade
If `true`, major version upgrades can be applied during the maintenance window to the Amazon Redshift engine that is running on the cluster. When a new major version of the Amazon Redshift engine is released, you can request that the service automatically apply upgrades during the maintenance window to the Amazon Redshift engine that is running on your cluster. Default: `true`
@option params [Integer] :number_of_nodes
The number of compute nodes in the cluster. This parameter is required when the **ClusterType** parameter is specified as `multi-node`. For information about determining how many nodes you need, go to [ Working with Clusters][1] in the *Amazon Redshift Cluster Management Guide*. If you don't specify this parameter, you get a single-node cluster. When requesting a multi-node cluster, you must specify the number of nodes that you want in the cluster. Default: `1` Constraints: Value must be at least 1 and no more than 100. [1]: https://docs.aws.amazon.com/redshift/latest/mgmt/working-with-clusters.html#how-many-nodes
@option params [Boolean] :publicly_accessible
If `true`, the cluster can be accessed from a public network.
@option params [Boolean] :encrypted
If `true`, the data in the cluster is encrypted at rest. Default: false
@option params [String] :hsm_client_certificate_identifier
Specifies the name of the HSM client certificate the Amazon Redshift cluster uses to retrieve the data encryption keys stored in an HSM.
@option params [String] :hsm_configuration_identifier
Specifies the name of the HSM configuration that contains the information the Amazon Redshift cluster can use to retrieve and store keys in an HSM.
@option params [String] :elastic_ip
The Elastic IP (EIP) address for the cluster. Constraints: The cluster must be provisioned in EC2-VPC and publicly-accessible through an Internet gateway. For more information about provisioning clusters in EC2-VPC, go to [Supported Platforms to Launch Your Cluster][1] in the Amazon Redshift Cluster Management Guide. [1]: https://docs.aws.amazon.com/redshift/latest/mgmt/working-with-clusters.html#cluster-platforms
@option params [Array<Types::Tag>] :tags
A list of tag instances.
@option params [String] :kms_key_id
The Key Management Service (KMS) key ID of the encryption key that you want to use to encrypt data in the cluster.
@option params [Boolean] :enhanced_vpc_routing
An option that specifies whether to create the cluster with enhanced VPC routing enabled. To create a cluster that uses enhanced VPC routing, the cluster must be in a VPC. For more information, see [Enhanced VPC Routing][1] in the Amazon Redshift Cluster Management Guide. If this option is `true`, enhanced VPC routing is enabled. Default: false [1]: https://docs.aws.amazon.com/redshift/latest/mgmt/enhanced-vpc-routing.html
@option params [String] :additional_info
Reserved.
@option params [Array<String>] :iam_roles
A list of Identity and Access Management (IAM) roles that can be used by the cluster to access other Amazon Web Services services. You must supply the IAM roles in their Amazon Resource Name (ARN) format. You can supply up to 10 IAM roles in a single request. A cluster can have up to 10 IAM roles associated with it at any time.
@option params [String] :maintenance_track_name
An optional parameter for the name of the maintenance track for the cluster. If you don't provide a maintenance track name, the cluster is assigned to the `current` track.
@option params [String] :snapshot_schedule_identifier
A unique identifier for the snapshot schedule.
@option params [Boolean] :availability_zone_relocation
The option to enable relocation for an Amazon Redshift cluster between Availability Zones after the cluster is created.
@option params [String] :aqua_configuration_status
The value represents how the cluster is configured to use AQUA (Advanced Query Accelerator) when it is created. Possible values include the following. * enabled - Use AQUA if it is available for the current Amazon Web Services Region and Amazon Redshift node type. * disabled - Don't use AQUA. * auto - Amazon Redshift determines whether to use AQUA.
@return [Types::CreateClusterResult] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::CreateClusterResult#cluster #cluster} => Types::Cluster
@example Request syntax with placeholder values
resp = client.create_cluster({ db_name: "String", cluster_identifier: "String", # required cluster_type: "String", node_type: "String", # required master_username: "String", # required master_user_password: "String", # required cluster_security_groups: ["String"], vpc_security_group_ids: ["String"], cluster_subnet_group_name: "String", availability_zone: "String", preferred_maintenance_window: "String", cluster_parameter_group_name: "String", automated_snapshot_retention_period: 1, manual_snapshot_retention_period: 1, port: 1, cluster_version: "String", allow_version_upgrade: false, number_of_nodes: 1, publicly_accessible: false, encrypted: false, hsm_client_certificate_identifier: "String", hsm_configuration_identifier: "String", elastic_ip: "String", tags: [ { key: "String", value: "String", }, ], kms_key_id: "String", enhanced_vpc_routing: false, additional_info: "String", iam_roles: ["String"], maintenance_track_name: "String", snapshot_schedule_identifier: "String", availability_zone_relocation: false, aqua_configuration_status: "enabled", # accepts enabled, disabled, auto })
@example Response structure
resp.cluster.cluster_identifier #=> String resp.cluster.node_type #=> String resp.cluster.cluster_status #=> String resp.cluster.cluster_availability_status #=> String resp.cluster.modify_status #=> String resp.cluster.master_username #=> String resp.cluster.db_name #=> String resp.cluster.endpoint.address #=> String resp.cluster.endpoint.port #=> Integer resp.cluster.endpoint.vpc_endpoints #=> Array resp.cluster.endpoint.vpc_endpoints[0].vpc_endpoint_id #=> String resp.cluster.endpoint.vpc_endpoints[0].vpc_id #=> String resp.cluster.endpoint.vpc_endpoints[0].network_interfaces #=> Array resp.cluster.endpoint.vpc_endpoints[0].network_interfaces[0].network_interface_id #=> String resp.cluster.endpoint.vpc_endpoints[0].network_interfaces[0].subnet_id #=> String resp.cluster.endpoint.vpc_endpoints[0].network_interfaces[0].private_ip_address #=> String resp.cluster.endpoint.vpc_endpoints[0].network_interfaces[0].availability_zone #=> String resp.cluster.cluster_create_time #=> Time resp.cluster.automated_snapshot_retention_period #=> Integer resp.cluster.manual_snapshot_retention_period #=> Integer resp.cluster.cluster_security_groups #=> Array resp.cluster.cluster_security_groups[0].cluster_security_group_name #=> String resp.cluster.cluster_security_groups[0].status #=> String resp.cluster.vpc_security_groups #=> Array resp.cluster.vpc_security_groups[0].vpc_security_group_id #=> String resp.cluster.vpc_security_groups[0].status #=> String resp.cluster.cluster_parameter_groups #=> Array resp.cluster.cluster_parameter_groups[0].parameter_group_name #=> String resp.cluster.cluster_parameter_groups[0].parameter_apply_status #=> String resp.cluster.cluster_parameter_groups[0].cluster_parameter_status_list #=> Array resp.cluster.cluster_parameter_groups[0].cluster_parameter_status_list[0].parameter_name #=> String resp.cluster.cluster_parameter_groups[0].cluster_parameter_status_list[0].parameter_apply_status #=> String resp.cluster.cluster_parameter_groups[0].cluster_parameter_status_list[0].parameter_apply_error_description #=> String resp.cluster.cluster_subnet_group_name #=> String resp.cluster.vpc_id #=> String resp.cluster.availability_zone #=> String resp.cluster.preferred_maintenance_window #=> String resp.cluster.pending_modified_values.master_user_password #=> String resp.cluster.pending_modified_values.node_type #=> String resp.cluster.pending_modified_values.number_of_nodes #=> Integer resp.cluster.pending_modified_values.cluster_type #=> String resp.cluster.pending_modified_values.cluster_version #=> String resp.cluster.pending_modified_values.automated_snapshot_retention_period #=> Integer resp.cluster.pending_modified_values.cluster_identifier #=> String resp.cluster.pending_modified_values.publicly_accessible #=> Boolean resp.cluster.pending_modified_values.enhanced_vpc_routing #=> Boolean resp.cluster.pending_modified_values.maintenance_track_name #=> String resp.cluster.pending_modified_values.encryption_type #=> String resp.cluster.cluster_version #=> String resp.cluster.allow_version_upgrade #=> Boolean resp.cluster.number_of_nodes #=> Integer resp.cluster.publicly_accessible #=> Boolean resp.cluster.encrypted #=> Boolean resp.cluster.restore_status.status #=> String resp.cluster.restore_status.current_restore_rate_in_mega_bytes_per_second #=> Float resp.cluster.restore_status.snapshot_size_in_mega_bytes #=> Integer resp.cluster.restore_status.progress_in_mega_bytes #=> Integer resp.cluster.restore_status.elapsed_time_in_seconds #=> Integer resp.cluster.restore_status.estimated_time_to_completion_in_seconds #=> Integer resp.cluster.data_transfer_progress.status #=> String resp.cluster.data_transfer_progress.current_rate_in_mega_bytes_per_second #=> Float resp.cluster.data_transfer_progress.total_data_in_mega_bytes #=> Integer resp.cluster.data_transfer_progress.data_transferred_in_mega_bytes #=> Integer resp.cluster.data_transfer_progress.estimated_time_to_completion_in_seconds #=> Integer resp.cluster.data_transfer_progress.elapsed_time_in_seconds #=> Integer resp.cluster.hsm_status.hsm_client_certificate_identifier #=> String resp.cluster.hsm_status.hsm_configuration_identifier #=> String resp.cluster.hsm_status.status #=> String resp.cluster.cluster_snapshot_copy_status.destination_region #=> String resp.cluster.cluster_snapshot_copy_status.retention_period #=> Integer resp.cluster.cluster_snapshot_copy_status.manual_snapshot_retention_period #=> Integer resp.cluster.cluster_snapshot_copy_status.snapshot_copy_grant_name #=> String resp.cluster.cluster_public_key #=> String resp.cluster.cluster_nodes #=> Array resp.cluster.cluster_nodes[0].node_role #=> String resp.cluster.cluster_nodes[0].private_ip_address #=> String resp.cluster.cluster_nodes[0].public_ip_address #=> String resp.cluster.elastic_ip_status.elastic_ip #=> String resp.cluster.elastic_ip_status.status #=> String resp.cluster.cluster_revision_number #=> String resp.cluster.tags #=> Array resp.cluster.tags[0].key #=> String resp.cluster.tags[0].value #=> String resp.cluster.kms_key_id #=> String resp.cluster.enhanced_vpc_routing #=> Boolean resp.cluster.iam_roles #=> Array resp.cluster.iam_roles[0].iam_role_arn #=> String resp.cluster.iam_roles[0].apply_status #=> String resp.cluster.pending_actions #=> Array resp.cluster.pending_actions[0] #=> String resp.cluster.maintenance_track_name #=> String resp.cluster.elastic_resize_number_of_node_options #=> String resp.cluster.deferred_maintenance_windows #=> Array resp.cluster.deferred_maintenance_windows[0].defer_maintenance_identifier #=> String resp.cluster.deferred_maintenance_windows[0].defer_maintenance_start_time #=> Time resp.cluster.deferred_maintenance_windows[0].defer_maintenance_end_time #=> Time resp.cluster.snapshot_schedule_identifier #=> String resp.cluster.snapshot_schedule_state #=> String, one of "MODIFYING", "ACTIVE", "FAILED" resp.cluster.expected_next_snapshot_schedule_time #=> Time resp.cluster.expected_next_snapshot_schedule_time_status #=> String resp.cluster.next_maintenance_window_start_time #=> Time resp.cluster.resize_info.resize_type #=> String resp.cluster.resize_info.allow_cancel_resize #=> Boolean resp.cluster.availability_zone_relocation_status #=> String resp.cluster.cluster_namespace_arn #=> String resp.cluster.total_storage_capacity_in_mega_bytes #=> Integer resp.cluster.aqua_configuration.aqua_status #=> String, one of "enabled", "disabled", "applying" resp.cluster.aqua_configuration.aqua_configuration_status #=> String, one of "enabled", "disabled", "auto"
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/CreateCluster AWS API Documentation
@overload create_cluster
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 1584 def create_cluster(params = {}, options = {}) req = build_request(:create_cluster, params) req.send_request(options) end
Creates an Amazon Redshift
parameter group.
Creating parameter groups is independent of creating clusters. You can associate a cluster with a parameter group when you create the cluster. You can also associate an existing cluster with a parameter group after the cluster is created by using ModifyCluster.
Parameters in the parameter group define specific behavior that applies to the databases you create on the cluster. For more information about parameters and parameter groups, go to [Amazon Redshift
Parameter Groups] in the *Amazon Redshift
Cluster Management Guide*.
[1]: docs.aws.amazon.com/redshift/latest/mgmt/working-with-parameter-groups.html
@option params [required, String] :parameter_group_name
The name of the cluster parameter group. Constraints: * Must be 1 to 255 alphanumeric characters or hyphens * First character must be a letter. * Cannot end with a hyphen or contain two consecutive hyphens. * Must be unique withing your Amazon Web Services account. <note markdown="1"> This value is stored as a lower-case string. </note>
@option params [required, String] :parameter_group_family
The Amazon Redshift engine version to which the cluster parameter group applies. The cluster engine version determines the set of parameters. To get a list of valid parameter group family names, you can call DescribeClusterParameterGroups. By default, Amazon Redshift returns a list of all the parameter groups that are owned by your Amazon Web Services account, including the default parameter groups for each Amazon Redshift engine version. The parameter group family names associated with the default parameter groups provide you the valid values. For example, a valid family name is "redshift-1.0".
@option params [required, String] :description
A description of the parameter group.
@option params [Array<Types::Tag>] :tags
A list of tag instances.
@return [Types::CreateClusterParameterGroupResult] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::CreateClusterParameterGroupResult#cluster_parameter_group #cluster_parameter_group} => Types::ClusterParameterGroup
@example Request syntax with placeholder values
resp = client.create_cluster_parameter_group({ parameter_group_name: "String", # required parameter_group_family: "String", # required description: "String", # required tags: [ { key: "String", value: "String", }, ], })
@example Response structure
resp.cluster_parameter_group.parameter_group_name #=> String resp.cluster_parameter_group.parameter_group_family #=> String resp.cluster_parameter_group.description #=> String resp.cluster_parameter_group.tags #=> Array resp.cluster_parameter_group.tags[0].key #=> String resp.cluster_parameter_group.tags[0].value #=> String
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/CreateClusterParameterGroup AWS API Documentation
@overload create_cluster_parameter_group
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 1673 def create_cluster_parameter_group(params = {}, options = {}) req = build_request(:create_cluster_parameter_group, params) req.send_request(options) end
Creates a new Amazon Redshift
security group. You use security groups to control access to non-VPC clusters.
For information about managing security groups, go to [Amazon Redshift
Cluster Security Groups] in the *Amazon Redshift
Cluster Management Guide*.
[1]: docs.aws.amazon.com/redshift/latest/mgmt/working-with-security-groups.html
@option params [required, String] :cluster_security_group_name
The name for the security group. Amazon Redshift stores the value as a lowercase string. Constraints: * Must contain no more than 255 alphanumeric characters or hyphens. * Must not be "Default". * Must be unique for all security groups that are created by your Amazon Web Services account. Example: `examplesecuritygroup`
@option params [required, String] :description
A description for the security group.
@option params [Array<Types::Tag>] :tags
A list of tag instances.
@return [Types::CreateClusterSecurityGroupResult] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::CreateClusterSecurityGroupResult#cluster_security_group #cluster_security_group} => Types::ClusterSecurityGroup
@example Request syntax with placeholder values
resp = client.create_cluster_security_group({ cluster_security_group_name: "String", # required description: "String", # required tags: [ { key: "String", value: "String", }, ], })
@example Response structure
resp.cluster_security_group.cluster_security_group_name #=> String resp.cluster_security_group.description #=> String resp.cluster_security_group.ec2_security_groups #=> Array resp.cluster_security_group.ec2_security_groups[0].status #=> String resp.cluster_security_group.ec2_security_groups[0].ec2_security_group_name #=> String resp.cluster_security_group.ec2_security_groups[0].ec2_security_group_owner_id #=> String resp.cluster_security_group.ec2_security_groups[0].tags #=> Array resp.cluster_security_group.ec2_security_groups[0].tags[0].key #=> String resp.cluster_security_group.ec2_security_groups[0].tags[0].value #=> String resp.cluster_security_group.ip_ranges #=> Array resp.cluster_security_group.ip_ranges[0].status #=> String resp.cluster_security_group.ip_ranges[0].cidrip #=> String resp.cluster_security_group.ip_ranges[0].tags #=> Array resp.cluster_security_group.ip_ranges[0].tags[0].key #=> String resp.cluster_security_group.ip_ranges[0].tags[0].value #=> String resp.cluster_security_group.tags #=> Array resp.cluster_security_group.tags[0].key #=> String resp.cluster_security_group.tags[0].value #=> String
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/CreateClusterSecurityGroup AWS API Documentation
@overload create_cluster_security_group
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 1752 def create_cluster_security_group(params = {}, options = {}) req = build_request(:create_cluster_security_group, params) req.send_request(options) end
Creates a manual snapshot of the specified cluster. The cluster must be in the `available` state.
For more information about working with snapshots, go to [Amazon Redshift
Snapshots] in the *Amazon Redshift
Cluster Management Guide*.
[1]: docs.aws.amazon.com/redshift/latest/mgmt/working-with-snapshots.html
@option params [required, String] :snapshot_identifier
A unique identifier for the snapshot that you are requesting. This identifier must be unique for all snapshots within the Amazon Web Services account. Constraints: * Cannot be null, empty, or blank * Must contain from 1 to 255 alphanumeric characters or hyphens * First character must be a letter * Cannot end with a hyphen or contain two consecutive hyphens Example: `my-snapshot-id`
@option params [required, String] :cluster_identifier
The cluster identifier for which you want a snapshot.
@option params [Integer] :manual_snapshot_retention_period
The number of days that a manual snapshot is retained. If the value is -1, the manual snapshot is retained indefinitely. The value must be either -1 or an integer between 1 and 3,653. The default value is -1.
@option params [Array<Types::Tag>] :tags
A list of tag instances.
@return [Types::CreateClusterSnapshotResult] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::CreateClusterSnapshotResult#snapshot #snapshot} => Types::Snapshot
@example Request syntax with placeholder values
resp = client.create_cluster_snapshot({ snapshot_identifier: "String", # required cluster_identifier: "String", # required manual_snapshot_retention_period: 1, tags: [ { key: "String", value: "String", }, ], })
@example Response structure
resp.snapshot.snapshot_identifier #=> String resp.snapshot.cluster_identifier #=> String resp.snapshot.snapshot_create_time #=> Time resp.snapshot.status #=> String resp.snapshot.port #=> Integer resp.snapshot.availability_zone #=> String resp.snapshot.cluster_create_time #=> Time resp.snapshot.master_username #=> String resp.snapshot.cluster_version #=> String resp.snapshot.engine_full_version #=> String resp.snapshot.snapshot_type #=> String resp.snapshot.node_type #=> String resp.snapshot.number_of_nodes #=> Integer resp.snapshot.db_name #=> String resp.snapshot.vpc_id #=> String resp.snapshot.encrypted #=> Boolean resp.snapshot.kms_key_id #=> String resp.snapshot.encrypted_with_hsm #=> Boolean resp.snapshot.accounts_with_restore_access #=> Array resp.snapshot.accounts_with_restore_access[0].account_id #=> String resp.snapshot.accounts_with_restore_access[0].account_alias #=> String resp.snapshot.owner_account #=> String resp.snapshot.total_backup_size_in_mega_bytes #=> Float resp.snapshot.actual_incremental_backup_size_in_mega_bytes #=> Float resp.snapshot.backup_progress_in_mega_bytes #=> Float resp.snapshot.current_backup_rate_in_mega_bytes_per_second #=> Float resp.snapshot.estimated_seconds_to_completion #=> Integer resp.snapshot.elapsed_time_in_seconds #=> Integer resp.snapshot.source_region #=> String resp.snapshot.tags #=> Array resp.snapshot.tags[0].key #=> String resp.snapshot.tags[0].value #=> String resp.snapshot.restorable_node_types #=> Array resp.snapshot.restorable_node_types[0] #=> String resp.snapshot.enhanced_vpc_routing #=> Boolean resp.snapshot.maintenance_track_name #=> String resp.snapshot.manual_snapshot_retention_period #=> Integer resp.snapshot.manual_snapshot_remaining_days #=> Integer resp.snapshot.snapshot_retention_start_time #=> Time
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/CreateClusterSnapshot AWS API Documentation
@overload create_cluster_snapshot
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 1863 def create_cluster_snapshot(params = {}, options = {}) req = build_request(:create_cluster_snapshot, params) req.send_request(options) end
Creates a new Amazon Redshift
subnet group. You must provide a list of one or more subnets in your existing Amazon Virtual Private Cloud (Amazon VPC) when creating Amazon Redshift
subnet group.
For information about subnet groups, go to [Amazon Redshift
Cluster Subnet Groups] in the *Amazon Redshift
Cluster Management Guide*.
[1]: docs.aws.amazon.com/redshift/latest/mgmt/working-with-cluster-subnet-groups.html
@option params [required, String] :cluster_subnet_group_name
The name for the subnet group. Amazon Redshift stores the value as a lowercase string. Constraints: * Must contain no more than 255 alphanumeric characters or hyphens. * Must not be "Default". * Must be unique for all subnet groups that are created by your Amazon Web Services account. Example: `examplesubnetgroup`
@option params [required, String] :description
A description for the subnet group.
@option params [required, Array<String>] :subnet_ids
An array of VPC subnet IDs. A maximum of 20 subnets can be modified in a single request.
@option params [Array<Types::Tag>] :tags
A list of tag instances.
@return [Types::CreateClusterSubnetGroupResult] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::CreateClusterSubnetGroupResult#cluster_subnet_group #cluster_subnet_group} => Types::ClusterSubnetGroup
@example Request syntax with placeholder values
resp = client.create_cluster_subnet_group({ cluster_subnet_group_name: "String", # required description: "String", # required subnet_ids: ["String"], # required tags: [ { key: "String", value: "String", }, ], })
@example Response structure
resp.cluster_subnet_group.cluster_subnet_group_name #=> String resp.cluster_subnet_group.description #=> String resp.cluster_subnet_group.vpc_id #=> String resp.cluster_subnet_group.subnet_group_status #=> String resp.cluster_subnet_group.subnets #=> Array resp.cluster_subnet_group.subnets[0].subnet_identifier #=> String resp.cluster_subnet_group.subnets[0].subnet_availability_zone.name #=> String resp.cluster_subnet_group.subnets[0].subnet_availability_zone.supported_platforms #=> Array resp.cluster_subnet_group.subnets[0].subnet_availability_zone.supported_platforms[0].name #=> String resp.cluster_subnet_group.subnets[0].subnet_status #=> String resp.cluster_subnet_group.tags #=> Array resp.cluster_subnet_group.tags[0].key #=> String resp.cluster_subnet_group.tags[0].value #=> String
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/CreateClusterSubnetGroup AWS API Documentation
@overload create_cluster_subnet_group
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 1942 def create_cluster_subnet_group(params = {}, options = {}) req = build_request(:create_cluster_subnet_group, params) req.send_request(options) end
Creates a Redshift-managed VPC endpoint.
@option params [String] :cluster_identifier
The cluster identifier of the cluster to access.
@option params [String] :resource_owner
The Amazon Web Services account ID of the owner of the cluster. This is only required if the cluster is in another Amazon Web Services account.
@option params [required, String] :endpoint_name
The Redshift-managed VPC endpoint name. An endpoint name must contain 1-30 characters. Valid characters are A-Z, a-z, 0-9, and hyphen(-). The first character must be a letter. The name can't contain two consecutive hyphens or end with a hyphen.
@option params [required, String] :subnet_group_name
The subnet group from which Amazon Redshift chooses the subnet to deploy the endpoint.
@option params [Array<String>] :vpc_security_group_ids
The security group that defines the ports, protocols, and sources for inbound traffic that you are authorizing into your endpoint.
@return [Types::EndpointAccess] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::EndpointAccess#cluster_identifier #cluster_identifier} => String * {Types::EndpointAccess#resource_owner #resource_owner} => String * {Types::EndpointAccess#subnet_group_name #subnet_group_name} => String * {Types::EndpointAccess#endpoint_status #endpoint_status} => String * {Types::EndpointAccess#endpoint_name #endpoint_name} => String * {Types::EndpointAccess#endpoint_create_time #endpoint_create_time} => Time * {Types::EndpointAccess#port #port} => Integer * {Types::EndpointAccess#address #address} => String * {Types::EndpointAccess#vpc_security_groups #vpc_security_groups} => Array<Types::VpcSecurityGroupMembership> * {Types::EndpointAccess#vpc_endpoint #vpc_endpoint} => Types::VpcEndpoint
@example Request syntax with placeholder values
resp = client.create_endpoint_access({ cluster_identifier: "String", resource_owner: "String", endpoint_name: "String", # required subnet_group_name: "String", # required vpc_security_group_ids: ["String"], })
@example Response structure
resp.cluster_identifier #=> String resp.resource_owner #=> String resp.subnet_group_name #=> String resp.endpoint_status #=> String resp.endpoint_name #=> String resp.endpoint_create_time #=> Time resp.port #=> Integer resp.address #=> String resp.vpc_security_groups #=> Array resp.vpc_security_groups[0].vpc_security_group_id #=> String resp.vpc_security_groups[0].status #=> String resp.vpc_endpoint.vpc_endpoint_id #=> String resp.vpc_endpoint.vpc_id #=> String resp.vpc_endpoint.network_interfaces #=> Array resp.vpc_endpoint.network_interfaces[0].network_interface_id #=> String resp.vpc_endpoint.network_interfaces[0].subnet_id #=> String resp.vpc_endpoint.network_interfaces[0].private_ip_address #=> String resp.vpc_endpoint.network_interfaces[0].availability_zone #=> String
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/CreateEndpointAccess AWS API Documentation
@overload create_endpoint_access
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 2020 def create_endpoint_access(params = {}, options = {}) req = build_request(:create_endpoint_access, params) req.send_request(options) end
Creates an Amazon Redshift
event notification subscription. This action requires an ARN (Amazon Resource
Name) of an Amazon SNS topic created by either the Amazon Redshift
console, the Amazon SNS console, or the Amazon SNS API. To obtain an ARN with Amazon SNS, you must create a topic in Amazon SNS and subscribe to the topic. The ARN is displayed in the SNS console.
You can specify the source type, and lists of Amazon Redshift
source IDs, event categories, and event severities. Notifications will be sent for all events you want that match those criteria. For example, you can specify source type = cluster, source ID = my-cluster-1 and mycluster2, event categories = Availability, Backup, and severity = ERROR. The subscription will only send notifications for those ERROR events in the Availability and Backup categories for the specified clusters.
If you specify both the source type and source IDs, such as source type = cluster and source identifier = my-cluster-1, notifications will be sent for all the cluster events for my-cluster-1. If you specify a source type but do not specify a source identifier, you will receive notice of the events for the objects of that type in your Amazon Web Services account. If you do not specify either the SourceType nor the SourceIdentifier, you will be notified of events generated from all Amazon Redshift
sources belonging to your Amazon Web Services account. You must specify a source type if you specify a source ID.
@option params [required, String] :subscription_name
The name of the event subscription to be created. Constraints: * Cannot be null, empty, or blank. * Must contain from 1 to 255 alphanumeric characters or hyphens. * First character must be a letter. * Cannot end with a hyphen or contain two consecutive hyphens.
@option params [required, String] :sns_topic_arn
The Amazon Resource Name (ARN) of the Amazon SNS topic used to transmit the event notifications. The ARN is created by Amazon SNS when you create a topic and subscribe to it.
@option params [String] :source_type
The type of source that will be generating the events. For example, if you want to be notified of events generated by a cluster, you would set this parameter to cluster. If this value is not specified, events are returned for all Amazon Redshift objects in your Amazon Web Services account. You must specify a source type in order to specify source IDs. Valid values: cluster, cluster-parameter-group, cluster-security-group, cluster-snapshot, and scheduled-action.
@option params [Array<String>] :source_ids
A list of one or more identifiers of Amazon Redshift source objects. All of the objects must be of the same type as was specified in the source type parameter. The event subscription will return only events generated by the specified objects. If not specified, then events are returned for all objects within the source type specified. Example: my-cluster-1, my-cluster-2 Example: my-snapshot-20131010
@option params [Array<String>] :event_categories
Specifies the Amazon Redshift event categories to be published by the event notification subscription. Values: configuration, management, monitoring, security, pending
@option params [String] :severity
Specifies the Amazon Redshift event severity to be published by the event notification subscription. Values: ERROR, INFO
@option params [Boolean] :enabled
A boolean value; set to `true` to activate the subscription, and set to `false` to create the subscription but not activate it.
@option params [Array<Types::Tag>] :tags
A list of tag instances.
@return [Types::CreateEventSubscriptionResult] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::CreateEventSubscriptionResult#event_subscription #event_subscription} => Types::EventSubscription
@example Request syntax with placeholder values
resp = client.create_event_subscription({ subscription_name: "String", # required sns_topic_arn: "String", # required source_type: "String", source_ids: ["String"], event_categories: ["String"], severity: "String", enabled: false, tags: [ { key: "String", value: "String", }, ], })
@example Response structure
resp.event_subscription.customer_aws_id #=> String resp.event_subscription.cust_subscription_id #=> String resp.event_subscription.sns_topic_arn #=> String resp.event_subscription.status #=> String resp.event_subscription.subscription_creation_time #=> Time resp.event_subscription.source_type #=> String resp.event_subscription.source_ids_list #=> Array resp.event_subscription.source_ids_list[0] #=> String resp.event_subscription.event_categories_list #=> Array resp.event_subscription.event_categories_list[0] #=> String resp.event_subscription.severity #=> String resp.event_subscription.enabled #=> Boolean resp.event_subscription.tags #=> Array resp.event_subscription.tags[0].key #=> String resp.event_subscription.tags[0].value #=> String
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/CreateEventSubscription AWS API Documentation
@overload create_event_subscription
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 2155 def create_event_subscription(params = {}, options = {}) req = build_request(:create_event_subscription, params) req.send_request(options) end
Creates an HSM client certificate that an Amazon Redshift
cluster will use to connect to the client's HSM in order to store and retrieve the keys used to encrypt the cluster databases.
The command returns a public key, which you must store in the HSM. In addition to creating the HSM certificate, you must create an Amazon Redshift
HSM configuration that provides a cluster the information needed to store and use encryption keys in the HSM. For more information, go to [Hardware Security Modules] in the *Amazon Redshift
Cluster Management Guide*.
[1]: docs.aws.amazon.com/redshift/latest/mgmt/working-with-db-encryption.html#working-with-HSM
@option params [required, String] :hsm_client_certificate_identifier
The identifier to be assigned to the new HSM client certificate that the cluster will use to connect to the HSM to use the database encryption keys.
@option params [Array<Types::Tag>] :tags
A list of tag instances.
@return [Types::CreateHsmClientCertificateResult] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::CreateHsmClientCertificateResult#hsm_client_certificate #hsm_client_certificate} => Types::HsmClientCertificate
@example Request syntax with placeholder values
resp = client.create_hsm_client_certificate({ hsm_client_certificate_identifier: "String", # required tags: [ { key: "String", value: "String", }, ], })
@example Response structure
resp.hsm_client_certificate.hsm_client_certificate_identifier #=> String resp.hsm_client_certificate.hsm_client_certificate_public_key #=> String resp.hsm_client_certificate.tags #=> Array resp.hsm_client_certificate.tags[0].key #=> String resp.hsm_client_certificate.tags[0].value #=> String
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/CreateHsmClientCertificate AWS API Documentation
@overload create_hsm_client_certificate
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 2211 def create_hsm_client_certificate(params = {}, options = {}) req = build_request(:create_hsm_client_certificate, params) req.send_request(options) end
Creates an HSM configuration that contains the information required by an Amazon Redshift
cluster to store and use database encryption keys in a Hardware Security Module (HSM). After creating the HSM configuration, you can specify it as a parameter when creating a cluster. The cluster will then store its encryption keys in the HSM.
In addition to creating an HSM configuration, you must also create an HSM client certificate. For more information, go to [Hardware Security Modules] in the Amazon Redshift
Cluster Management Guide.
[1]: docs.aws.amazon.com/redshift/latest/mgmt/working-with-HSM.html
@option params [required, String] :hsm_configuration_identifier
The identifier to be assigned to the new Amazon Redshift HSM configuration.
@option params [required, String] :description
A text description of the HSM configuration to be created.
@option params [required, String] :hsm_ip_address
The IP address that the Amazon Redshift cluster must use to access the HSM.
@option params [required, String] :hsm_partition_name
The name of the partition in the HSM where the Amazon Redshift clusters will store their database encryption keys.
@option params [required, String] :hsm_partition_password
The password required to access the HSM partition.
@option params [required, String] :hsm_server_public_certificate
The HSMs public certificate file. When using Cloud HSM, the file name is server.pem.
@option params [Array<Types::Tag>] :tags
A list of tag instances.
@return [Types::CreateHsmConfigurationResult] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::CreateHsmConfigurationResult#hsm_configuration #hsm_configuration} => Types::HsmConfiguration
@example Request syntax with placeholder values
resp = client.create_hsm_configuration({ hsm_configuration_identifier: "String", # required description: "String", # required hsm_ip_address: "String", # required hsm_partition_name: "String", # required hsm_partition_password: "String", # required hsm_server_public_certificate: "String", # required tags: [ { key: "String", value: "String", }, ], })
@example Response structure
resp.hsm_configuration.hsm_configuration_identifier #=> String resp.hsm_configuration.description #=> String resp.hsm_configuration.hsm_ip_address #=> String resp.hsm_configuration.hsm_partition_name #=> String resp.hsm_configuration.tags #=> Array resp.hsm_configuration.tags[0].key #=> String resp.hsm_configuration.tags[0].value #=> String
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/CreateHsmConfiguration AWS API Documentation
@overload create_hsm_configuration
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 2290 def create_hsm_configuration(params = {}, options = {}) req = build_request(:create_hsm_configuration, params) req.send_request(options) end
Creates a scheduled action. A scheduled action contains a schedule and an Amazon Redshift
API action. For example, you can create a schedule of when to run the `ResizeCluster` API operation.
@option params [required, String] :scheduled_action_name
The name of the scheduled action. The name must be unique within an account. For more information about this parameter, see ScheduledAction.
@option params [required, Types::ScheduledActionType] :target_action
A JSON format string of the Amazon Redshift API operation with input parameters. For more information about this parameter, see ScheduledAction.
@option params [required, String] :schedule
The schedule in `at( )` or `cron( )` format. For more information about this parameter, see ScheduledAction.
@option params [required, String] :iam_role
The IAM role to assume to run the target action. For more information about this parameter, see ScheduledAction.
@option params [String] :scheduled_action_description
The description of the scheduled action.
@option params [Time,DateTime,Date,Integer,String] :start_time
The start time in UTC of the scheduled action. Before this time, the scheduled action does not trigger. For more information about this parameter, see ScheduledAction.
@option params [Time,DateTime,Date,Integer,String] :end_time
The end time in UTC of the scheduled action. After this time, the scheduled action does not trigger. For more information about this parameter, see ScheduledAction.
@option params [Boolean] :enable
If true, the schedule is enabled. If false, the scheduled action does not trigger. For more information about `state` of the scheduled action, see ScheduledAction.
@return [Types::ScheduledAction] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::ScheduledAction#scheduled_action_name #scheduled_action_name} => String * {Types::ScheduledAction#target_action #target_action} => Types::ScheduledActionType * {Types::ScheduledAction#schedule #schedule} => String * {Types::ScheduledAction#iam_role #iam_role} => String * {Types::ScheduledAction#scheduled_action_description #scheduled_action_description} => String * {Types::ScheduledAction#state #state} => String * {Types::ScheduledAction#next_invocations #next_invocations} => Array<Time> * {Types::ScheduledAction#start_time #start_time} => Time * {Types::ScheduledAction#end_time #end_time} => Time
@example Request syntax with placeholder values
resp = client.create_scheduled_action({ scheduled_action_name: "String", # required target_action: { # required resize_cluster: { cluster_identifier: "String", # required cluster_type: "String", node_type: "String", number_of_nodes: 1, classic: false, }, pause_cluster: { cluster_identifier: "String", # required }, resume_cluster: { cluster_identifier: "String", # required }, }, schedule: "String", # required iam_role: "String", # required scheduled_action_description: "String", start_time: Time.now, end_time: Time.now, enable: false, })
@example Response structure
resp.scheduled_action_name #=> String resp.target_action.resize_cluster.cluster_identifier #=> String resp.target_action.resize_cluster.cluster_type #=> String resp.target_action.resize_cluster.node_type #=> String resp.target_action.resize_cluster.number_of_nodes #=> Integer resp.target_action.resize_cluster.classic #=> Boolean resp.target_action.pause_cluster.cluster_identifier #=> String resp.target_action.resume_cluster.cluster_identifier #=> String resp.schedule #=> String resp.iam_role #=> String resp.scheduled_action_description #=> String resp.state #=> String, one of "ACTIVE", "DISABLED" resp.next_invocations #=> Array resp.next_invocations[0] #=> Time resp.start_time #=> Time resp.end_time #=> Time
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/CreateScheduledAction AWS API Documentation
@overload create_scheduled_action
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 2397 def create_scheduled_action(params = {}, options = {}) req = build_request(:create_scheduled_action, params) req.send_request(options) end
Creates a snapshot copy grant that permits Amazon Redshift
to use a customer master key (CMK) from Key Management Service (KMS) to encrypt copied snapshots in a destination region.
For more information about managing snapshot copy grants, go to
Cluster Management Guide*.
[1]: docs.aws.amazon.com/redshift/latest/mgmt/working-with-db-encryption.html
@option params [required, String] :snapshot_copy_grant_name
The name of the snapshot copy grant. This name must be unique in the region for the Amazon Web Services account. Constraints: * Must contain from 1 to 63 alphanumeric characters or hyphens. * Alphabetic characters must be lowercase. * First character must be a letter. * Cannot end with a hyphen or contain two consecutive hyphens. * Must be unique for all clusters within an Amazon Web Services account.
@option params [String] :kms_key_id
The unique identifier of the customer master key (CMK) to which to grant Amazon Redshift permission. If no key is specified, the default key is used.
@option params [Array<Types::Tag>] :tags
A list of tag instances.
@return [Types::CreateSnapshotCopyGrantResult] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::CreateSnapshotCopyGrantResult#snapshot_copy_grant #snapshot_copy_grant} => Types::SnapshotCopyGrant
@example Request syntax with placeholder values
resp = client.create_snapshot_copy_grant({ snapshot_copy_grant_name: "String", # required kms_key_id: "String", tags: [ { key: "String", value: "String", }, ], })
@example Response structure
resp.snapshot_copy_grant.snapshot_copy_grant_name #=> String resp.snapshot_copy_grant.kms_key_id #=> String resp.snapshot_copy_grant.tags #=> Array resp.snapshot_copy_grant.tags[0].key #=> String resp.snapshot_copy_grant.tags[0].value #=> String
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/CreateSnapshotCopyGrant AWS API Documentation
@overload create_snapshot_copy_grant
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 2468 def create_snapshot_copy_grant(params = {}, options = {}) req = build_request(:create_snapshot_copy_grant, params) req.send_request(options) end
Create a snapshot schedule that can be associated to a cluster and which overrides the default system backup schedule.
@option params [Array<String>] :schedule_definitions
The definition of the snapshot schedule. The definition is made up of schedule expressions, for example "cron(30 12 *)" or "rate(12 hours)".
@option params [String] :schedule_identifier
A unique identifier for a snapshot schedule. Only alphanumeric characters are allowed for the identifier.
@option params [String] :schedule_description
The description of the snapshot schedule.
@option params [Array<Types::Tag>] :tags
An optional set of tags you can use to search for the schedule.
@option params [Boolean] :dry_run
@option params [Integer] :next_invocations
@return [Types::SnapshotSchedule] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::SnapshotSchedule#schedule_definitions #schedule_definitions} => Array<String> * {Types::SnapshotSchedule#schedule_identifier #schedule_identifier} => String * {Types::SnapshotSchedule#schedule_description #schedule_description} => String * {Types::SnapshotSchedule#tags #tags} => Array<Types::Tag> * {Types::SnapshotSchedule#next_invocations #next_invocations} => Array<Time> * {Types::SnapshotSchedule#associated_cluster_count #associated_cluster_count} => Integer * {Types::SnapshotSchedule#associated_clusters #associated_clusters} => Array<Types::ClusterAssociatedToSchedule>
@example Request syntax with placeholder values
resp = client.create_snapshot_schedule({ schedule_definitions: ["String"], schedule_identifier: "String", schedule_description: "String", tags: [ { key: "String", value: "String", }, ], dry_run: false, next_invocations: 1, })
@example Response structure
resp.schedule_definitions #=> Array resp.schedule_definitions[0] #=> String resp.schedule_identifier #=> String resp.schedule_description #=> String resp.tags #=> Array resp.tags[0].key #=> String resp.tags[0].value #=> String resp.next_invocations #=> Array resp.next_invocations[0] #=> Time resp.associated_cluster_count #=> Integer resp.associated_clusters #=> Array resp.associated_clusters[0].cluster_identifier #=> String resp.associated_clusters[0].schedule_association_state #=> String, one of "MODIFYING", "ACTIVE", "FAILED"
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/CreateSnapshotSchedule AWS API Documentation
@overload create_snapshot_schedule
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 2541 def create_snapshot_schedule(params = {}, options = {}) req = build_request(:create_snapshot_schedule, params) req.send_request(options) end
Creates a usage limit for a specified Amazon Redshift
feature on a cluster. The usage limit is identified by the returned usage limit identifier.
@option params [required, String] :cluster_identifier
The identifier of the cluster that you want to limit usage.
@option params [required, String] :feature_type
The Amazon Redshift feature that you want to limit.
@option params [required, String] :limit_type
The type of limit. Depending on the feature type, this can be based on a time duration or data size. If `FeatureType` is `spectrum`, then `LimitType` must be `data-scanned`. If `FeatureType` is `concurrency-scaling`, then `LimitType` must be `time`.
@option params [required, Integer] :amount
The limit amount. If time-based, this amount is in minutes. If data-based, this amount is in terabytes (TB). The value must be a positive number.
@option params [String] :period
The time period that the amount applies to. A `weekly` period begins on Sunday. The default is `monthly`.
@option params [String] :breach_action
The action that Amazon Redshift takes when the limit is reached. The default is log. For more information about this parameter, see UsageLimit.
@option params [Array<Types::Tag>] :tags
A list of tag instances.
@return [Types::UsageLimit] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::UsageLimit#usage_limit_id #usage_limit_id} => String * {Types::UsageLimit#cluster_identifier #cluster_identifier} => String * {Types::UsageLimit#feature_type #feature_type} => String * {Types::UsageLimit#limit_type #limit_type} => String * {Types::UsageLimit#amount #amount} => Integer * {Types::UsageLimit#period #period} => String * {Types::UsageLimit#breach_action #breach_action} => String * {Types::UsageLimit#tags #tags} => Array<Types::Tag>
@example Request syntax with placeholder values
resp = client.create_usage_limit({ cluster_identifier: "String", # required feature_type: "spectrum", # required, accepts spectrum, concurrency-scaling limit_type: "time", # required, accepts time, data-scanned amount: 1, # required period: "daily", # accepts daily, weekly, monthly breach_action: "log", # accepts log, emit-metric, disable tags: [ { key: "String", value: "String", }, ], })
@example Response structure
resp.usage_limit_id #=> String resp.cluster_identifier #=> String resp.feature_type #=> String, one of "spectrum", "concurrency-scaling" resp.limit_type #=> String, one of "time", "data-scanned" resp.amount #=> Integer resp.period #=> String, one of "daily", "weekly", "monthly" resp.breach_action #=> String, one of "log", "emit-metric", "disable" resp.tags #=> Array resp.tags[0].key #=> String resp.tags[0].value #=> String
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/CreateUsageLimit AWS API Documentation
@overload create_usage_limit
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 2668 def create_usage_limit(params = {}, options = {}) req = build_request(:create_usage_limit, params) req.send_request(options) end
Deletes an authentication profile.
@option params [required, String] :authentication_profile_name
The name of the authentication profile to delete.
@return [Types::DeleteAuthenticationProfileResult] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::DeleteAuthenticationProfileResult#authentication_profile_name #authentication_profile_name} => String
@example Request syntax with placeholder values
resp = client.delete_authentication_profile({ authentication_profile_name: "AuthenticationProfileNameString", # required })
@example Response structure
resp.authentication_profile_name #=> String
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/DeleteAuthenticationProfile AWS API Documentation
@overload delete_authentication_profile
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 2741 def delete_authentication_profile(params = {}, options = {}) req = build_request(:delete_authentication_profile, params) req.send_request(options) end
Deletes a previously provisioned cluster without its final snapshot being created. A successful response from the web service indicates that the request was received correctly. Use DescribeClusters to monitor the status of the deletion. The delete operation cannot be canceled or reverted once submitted. For more information about managing clusters, go to [Amazon Redshift
Clusters] in the *Amazon Redshift
Cluster Management Guide*.
If you want to shut down the cluster and retain it for future use, set SkipFinalClusterSnapshot to `false` and specify a name for FinalClusterSnapshotIdentifier. You can later restore this snapshot to resume using the cluster. If a final cluster snapshot is requested, the status of the cluster will be “final-snapshot” while the snapshot is being taken, then it's “deleting” once Amazon Redshift
begins deleting the cluster.
For more information about managing clusters, go to [Amazon Redshift
Clusters] in the *Amazon Redshift
Cluster Management Guide*.
[1]: docs.aws.amazon.com/redshift/latest/mgmt/working-with-clusters.html
@option params [required, String] :cluster_identifier
The identifier of the cluster to be deleted. Constraints: * Must contain lowercase characters. * Must contain from 1 to 63 alphanumeric characters or hyphens. * First character must be a letter. * Cannot end with a hyphen or contain two consecutive hyphens.
@option params [Boolean] :skip_final_cluster_snapshot
Determines whether a final snapshot of the cluster is created before Amazon Redshift deletes the cluster. If `true`, a final cluster snapshot is not created. If `false`, a final cluster snapshot is created before the cluster is deleted. <note markdown="1"> The *FinalClusterSnapshotIdentifier* parameter must be specified if *SkipFinalClusterSnapshot* is `false`. </note> Default: `false`
@option params [String] :final_cluster_snapshot_identifier
The identifier of the final snapshot that is to be created immediately before deleting the cluster. If this parameter is provided, *SkipFinalClusterSnapshot* must be `false`. Constraints: * Must be 1 to 255 alphanumeric characters. * First character must be a letter. * Cannot end with a hyphen or contain two consecutive hyphens.
@option params [Integer] :final_cluster_snapshot_retention_period
The number of days that a manual snapshot is retained. If the value is -1, the manual snapshot is retained indefinitely. The value must be either -1 or an integer between 1 and 3,653. The default value is -1.
@return [Types::DeleteClusterResult] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::DeleteClusterResult#cluster #cluster} => Types::Cluster
@example Request syntax with placeholder values
resp = client.delete_cluster({ cluster_identifier: "String", # required skip_final_cluster_snapshot: false, final_cluster_snapshot_identifier: "String", final_cluster_snapshot_retention_period: 1, })
@example Response structure
resp.cluster.cluster_identifier #=> String resp.cluster.node_type #=> String resp.cluster.cluster_status #=> String resp.cluster.cluster_availability_status #=> String resp.cluster.modify_status #=> String resp.cluster.master_username #=> String resp.cluster.db_name #=> String resp.cluster.endpoint.address #=> String resp.cluster.endpoint.port #=> Integer resp.cluster.endpoint.vpc_endpoints #=> Array resp.cluster.endpoint.vpc_endpoints[0].vpc_endpoint_id #=> String resp.cluster.endpoint.vpc_endpoints[0].vpc_id #=> String resp.cluster.endpoint.vpc_endpoints[0].network_interfaces #=> Array resp.cluster.endpoint.vpc_endpoints[0].network_interfaces[0].network_interface_id #=> String resp.cluster.endpoint.vpc_endpoints[0].network_interfaces[0].subnet_id #=> String resp.cluster.endpoint.vpc_endpoints[0].network_interfaces[0].private_ip_address #=> String resp.cluster.endpoint.vpc_endpoints[0].network_interfaces[0].availability_zone #=> String resp.cluster.cluster_create_time #=> Time resp.cluster.automated_snapshot_retention_period #=> Integer resp.cluster.manual_snapshot_retention_period #=> Integer resp.cluster.cluster_security_groups #=> Array resp.cluster.cluster_security_groups[0].cluster_security_group_name #=> String resp.cluster.cluster_security_groups[0].status #=> String resp.cluster.vpc_security_groups #=> Array resp.cluster.vpc_security_groups[0].vpc_security_group_id #=> String resp.cluster.vpc_security_groups[0].status #=> String resp.cluster.cluster_parameter_groups #=> Array resp.cluster.cluster_parameter_groups[0].parameter_group_name #=> String resp.cluster.cluster_parameter_groups[0].parameter_apply_status #=> String resp.cluster.cluster_parameter_groups[0].cluster_parameter_status_list #=> Array resp.cluster.cluster_parameter_groups[0].cluster_parameter_status_list[0].parameter_name #=> String resp.cluster.cluster_parameter_groups[0].cluster_parameter_status_list[0].parameter_apply_status #=> String resp.cluster.cluster_parameter_groups[0].cluster_parameter_status_list[0].parameter_apply_error_description #=> String resp.cluster.cluster_subnet_group_name #=> String resp.cluster.vpc_id #=> String resp.cluster.availability_zone #=> String resp.cluster.preferred_maintenance_window #=> String resp.cluster.pending_modified_values.master_user_password #=> String resp.cluster.pending_modified_values.node_type #=> String resp.cluster.pending_modified_values.number_of_nodes #=> Integer resp.cluster.pending_modified_values.cluster_type #=> String resp.cluster.pending_modified_values.cluster_version #=> String resp.cluster.pending_modified_values.automated_snapshot_retention_period #=> Integer resp.cluster.pending_modified_values.cluster_identifier #=> String resp.cluster.pending_modified_values.publicly_accessible #=> Boolean resp.cluster.pending_modified_values.enhanced_vpc_routing #=> Boolean resp.cluster.pending_modified_values.maintenance_track_name #=> String resp.cluster.pending_modified_values.encryption_type #=> String resp.cluster.cluster_version #=> String resp.cluster.allow_version_upgrade #=> Boolean resp.cluster.number_of_nodes #=> Integer resp.cluster.publicly_accessible #=> Boolean resp.cluster.encrypted #=> Boolean resp.cluster.restore_status.status #=> String resp.cluster.restore_status.current_restore_rate_in_mega_bytes_per_second #=> Float resp.cluster.restore_status.snapshot_size_in_mega_bytes #=> Integer resp.cluster.restore_status.progress_in_mega_bytes #=> Integer resp.cluster.restore_status.elapsed_time_in_seconds #=> Integer resp.cluster.restore_status.estimated_time_to_completion_in_seconds #=> Integer resp.cluster.data_transfer_progress.status #=> String resp.cluster.data_transfer_progress.current_rate_in_mega_bytes_per_second #=> Float resp.cluster.data_transfer_progress.total_data_in_mega_bytes #=> Integer resp.cluster.data_transfer_progress.data_transferred_in_mega_bytes #=> Integer resp.cluster.data_transfer_progress.estimated_time_to_completion_in_seconds #=> Integer resp.cluster.data_transfer_progress.elapsed_time_in_seconds #=> Integer resp.cluster.hsm_status.hsm_client_certificate_identifier #=> String resp.cluster.hsm_status.hsm_configuration_identifier #=> String resp.cluster.hsm_status.status #=> String resp.cluster.cluster_snapshot_copy_status.destination_region #=> String resp.cluster.cluster_snapshot_copy_status.retention_period #=> Integer resp.cluster.cluster_snapshot_copy_status.manual_snapshot_retention_period #=> Integer resp.cluster.cluster_snapshot_copy_status.snapshot_copy_grant_name #=> String resp.cluster.cluster_public_key #=> String resp.cluster.cluster_nodes #=> Array resp.cluster.cluster_nodes[0].node_role #=> String resp.cluster.cluster_nodes[0].private_ip_address #=> String resp.cluster.cluster_nodes[0].public_ip_address #=> String resp.cluster.elastic_ip_status.elastic_ip #=> String resp.cluster.elastic_ip_status.status #=> String resp.cluster.cluster_revision_number #=> String resp.cluster.tags #=> Array resp.cluster.tags[0].key #=> String resp.cluster.tags[0].value #=> String resp.cluster.kms_key_id #=> String resp.cluster.enhanced_vpc_routing #=> Boolean resp.cluster.iam_roles #=> Array resp.cluster.iam_roles[0].iam_role_arn #=> String resp.cluster.iam_roles[0].apply_status #=> String resp.cluster.pending_actions #=> Array resp.cluster.pending_actions[0] #=> String resp.cluster.maintenance_track_name #=> String resp.cluster.elastic_resize_number_of_node_options #=> String resp.cluster.deferred_maintenance_windows #=> Array resp.cluster.deferred_maintenance_windows[0].defer_maintenance_identifier #=> String resp.cluster.deferred_maintenance_windows[0].defer_maintenance_start_time #=> Time resp.cluster.deferred_maintenance_windows[0].defer_maintenance_end_time #=> Time resp.cluster.snapshot_schedule_identifier #=> String resp.cluster.snapshot_schedule_state #=> String, one of "MODIFYING", "ACTIVE", "FAILED" resp.cluster.expected_next_snapshot_schedule_time #=> Time resp.cluster.expected_next_snapshot_schedule_time_status #=> String resp.cluster.next_maintenance_window_start_time #=> Time resp.cluster.resize_info.resize_type #=> String resp.cluster.resize_info.allow_cancel_resize #=> Boolean resp.cluster.availability_zone_relocation_status #=> String resp.cluster.cluster_namespace_arn #=> String resp.cluster.total_storage_capacity_in_mega_bytes #=> Integer resp.cluster.aqua_configuration.aqua_status #=> String, one of "enabled", "disabled", "applying" resp.cluster.aqua_configuration.aqua_configuration_status #=> String, one of "enabled", "disabled", "auto"
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/DeleteCluster AWS API Documentation
@overload delete_cluster
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 2944 def delete_cluster(params = {}, options = {}) req = build_request(:delete_cluster, params) req.send_request(options) end
Deletes a specified Amazon Redshift
parameter group.
<note markdown=“1”> You cannot delete a parameter group if it is associated with a cluster.
</note>
@option params [required, String] :parameter_group_name
The name of the parameter group to be deleted. Constraints: * Must be the name of an existing cluster parameter group. * Cannot delete a default cluster parameter group.
@return [Struct] Returns an empty {Seahorse::Client::Response response}.
@example Request syntax with placeholder values
resp = client.delete_cluster_parameter_group({ parameter_group_name: "String", # required })
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/DeleteClusterParameterGroup AWS API Documentation
@overload delete_cluster_parameter_group
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 2977 def delete_cluster_parameter_group(params = {}, options = {}) req = build_request(:delete_cluster_parameter_group, params) req.send_request(options) end
Deletes an Amazon Redshift
security group.
<note markdown=“1”> You cannot delete a security group that is associated with any clusters. You cannot delete the default security group.
</note>
For information about managing security groups, go to [Amazon Redshift
Cluster Security Groups] in the *Amazon Redshift
Cluster Management Guide*.
[1]: docs.aws.amazon.com/redshift/latest/mgmt/working-with-security-groups.html
@option params [required, String] :cluster_security_group_name
The name of the cluster security group to be deleted.
@return [Struct] Returns an empty {Seahorse::Client::Response response}.
@example Request syntax with placeholder values
resp = client.delete_cluster_security_group({ cluster_security_group_name: "String", # required })
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/DeleteClusterSecurityGroup AWS API Documentation
@overload delete_cluster_security_group
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 3012 def delete_cluster_security_group(params = {}, options = {}) req = build_request(:delete_cluster_security_group, params) req.send_request(options) end
Deletes the specified manual snapshot. The snapshot must be in the `available` state, with no other users authorized to access the snapshot.
Unlike automated snapshots, manual snapshots are retained even after you delete your cluster. Amazon Redshift
does not delete your manual snapshots. You must delete manual snapshot explicitly to avoid getting charged. If other accounts are authorized to access the snapshot, you must revoke all of the authorizations before you can delete the snapshot.
@option params [required, String] :snapshot_identifier
The unique identifier of the manual snapshot to be deleted. Constraints: Must be the name of an existing snapshot that is in the `available`, `failed`, or `cancelled` state.
@option params [String] :snapshot_cluster_identifier
The unique identifier of the cluster the snapshot was created from. This parameter is required if your IAM user has a policy containing a snapshot resource element that specifies anything other than * for the cluster name. Constraints: Must be the name of valid cluster.
@return [Types::DeleteClusterSnapshotResult] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::DeleteClusterSnapshotResult#snapshot #snapshot} => Types::Snapshot
@example Request syntax with placeholder values
resp = client.delete_cluster_snapshot({ snapshot_identifier: "String", # required snapshot_cluster_identifier: "String", })
@example Response structure
resp.snapshot.snapshot_identifier #=> String resp.snapshot.cluster_identifier #=> String resp.snapshot.snapshot_create_time #=> Time resp.snapshot.status #=> String resp.snapshot.port #=> Integer resp.snapshot.availability_zone #=> String resp.snapshot.cluster_create_time #=> Time resp.snapshot.master_username #=> String resp.snapshot.cluster_version #=> String resp.snapshot.engine_full_version #=> String resp.snapshot.snapshot_type #=> String resp.snapshot.node_type #=> String resp.snapshot.number_of_nodes #=> Integer resp.snapshot.db_name #=> String resp.snapshot.vpc_id #=> String resp.snapshot.encrypted #=> Boolean resp.snapshot.kms_key_id #=> String resp.snapshot.encrypted_with_hsm #=> Boolean resp.snapshot.accounts_with_restore_access #=> Array resp.snapshot.accounts_with_restore_access[0].account_id #=> String resp.snapshot.accounts_with_restore_access[0].account_alias #=> String resp.snapshot.owner_account #=> String resp.snapshot.total_backup_size_in_mega_bytes #=> Float resp.snapshot.actual_incremental_backup_size_in_mega_bytes #=> Float resp.snapshot.backup_progress_in_mega_bytes #=> Float resp.snapshot.current_backup_rate_in_mega_bytes_per_second #=> Float resp.snapshot.estimated_seconds_to_completion #=> Integer resp.snapshot.elapsed_time_in_seconds #=> Integer resp.snapshot.source_region #=> String resp.snapshot.tags #=> Array resp.snapshot.tags[0].key #=> String resp.snapshot.tags[0].value #=> String resp.snapshot.restorable_node_types #=> Array resp.snapshot.restorable_node_types[0] #=> String resp.snapshot.enhanced_vpc_routing #=> Boolean resp.snapshot.maintenance_track_name #=> String resp.snapshot.manual_snapshot_retention_period #=> Integer resp.snapshot.manual_snapshot_remaining_days #=> Integer resp.snapshot.snapshot_retention_start_time #=> Time
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/DeleteClusterSnapshot AWS API Documentation
@overload delete_cluster_snapshot
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 3099 def delete_cluster_snapshot(params = {}, options = {}) req = build_request(:delete_cluster_snapshot, params) req.send_request(options) end
Deletes the specified cluster subnet group.
@option params [required, String] :cluster_subnet_group_name
The name of the cluster subnet group name to be deleted.
@return [Struct] Returns an empty {Seahorse::Client::Response response}.
@example Request syntax with placeholder values
resp = client.delete_cluster_subnet_group({ cluster_subnet_group_name: "String", # required })
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/DeleteClusterSubnetGroup AWS API Documentation
@overload delete_cluster_subnet_group
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 3121 def delete_cluster_subnet_group(params = {}, options = {}) req = build_request(:delete_cluster_subnet_group, params) req.send_request(options) end
Deletes a Redshift-managed VPC endpoint.
@option params [required, String] :endpoint_name
The Redshift-managed VPC endpoint to delete.
@return [Types::EndpointAccess] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::EndpointAccess#cluster_identifier #cluster_identifier} => String * {Types::EndpointAccess#resource_owner #resource_owner} => String * {Types::EndpointAccess#subnet_group_name #subnet_group_name} => String * {Types::EndpointAccess#endpoint_status #endpoint_status} => String * {Types::EndpointAccess#endpoint_name #endpoint_name} => String * {Types::EndpointAccess#endpoint_create_time #endpoint_create_time} => Time * {Types::EndpointAccess#port #port} => Integer * {Types::EndpointAccess#address #address} => String * {Types::EndpointAccess#vpc_security_groups #vpc_security_groups} => Array<Types::VpcSecurityGroupMembership> * {Types::EndpointAccess#vpc_endpoint #vpc_endpoint} => Types::VpcEndpoint
@example Request syntax with placeholder values
resp = client.delete_endpoint_access({ endpoint_name: "String", # required })
@example Response structure
resp.cluster_identifier #=> String resp.resource_owner #=> String resp.subnet_group_name #=> String resp.endpoint_status #=> String resp.endpoint_name #=> String resp.endpoint_create_time #=> Time resp.port #=> Integer resp.address #=> String resp.vpc_security_groups #=> Array resp.vpc_security_groups[0].vpc_security_group_id #=> String resp.vpc_security_groups[0].status #=> String resp.vpc_endpoint.vpc_endpoint_id #=> String resp.vpc_endpoint.vpc_id #=> String resp.vpc_endpoint.network_interfaces #=> Array resp.vpc_endpoint.network_interfaces[0].network_interface_id #=> String resp.vpc_endpoint.network_interfaces[0].subnet_id #=> String resp.vpc_endpoint.network_interfaces[0].private_ip_address #=> String resp.vpc_endpoint.network_interfaces[0].availability_zone #=> String
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/DeleteEndpointAccess AWS API Documentation
@overload delete_endpoint_access
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 3175 def delete_endpoint_access(params = {}, options = {}) req = build_request(:delete_endpoint_access, params) req.send_request(options) end
Deletes an Amazon Redshift
event notification subscription.
@option params [required, String] :subscription_name
The name of the Amazon Redshift event notification subscription to be deleted.
@return [Struct] Returns an empty {Seahorse::Client::Response response}.
@example Request syntax with placeholder values
resp = client.delete_event_subscription({ subscription_name: "String", # required })
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/DeleteEventSubscription AWS API Documentation
@overload delete_event_subscription
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 3198 def delete_event_subscription(params = {}, options = {}) req = build_request(:delete_event_subscription, params) req.send_request(options) end
Deletes the specified HSM client certificate.
@option params [required, String] :hsm_client_certificate_identifier
The identifier of the HSM client certificate to be deleted.
@return [Struct] Returns an empty {Seahorse::Client::Response response}.
@example Request syntax with placeholder values
resp = client.delete_hsm_client_certificate({ hsm_client_certificate_identifier: "String", # required })
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/DeleteHsmClientCertificate AWS API Documentation
@overload delete_hsm_client_certificate
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 3220 def delete_hsm_client_certificate(params = {}, options = {}) req = build_request(:delete_hsm_client_certificate, params) req.send_request(options) end
Deletes the specified Amazon Redshift
HSM configuration.
@option params [required, String] :hsm_configuration_identifier
The identifier of the Amazon Redshift HSM configuration to be deleted.
@return [Struct] Returns an empty {Seahorse::Client::Response response}.
@example Request syntax with placeholder values
resp = client.delete_hsm_configuration({ hsm_configuration_identifier: "String", # required })
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/DeleteHsmConfiguration AWS API Documentation
@overload delete_hsm_configuration
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 3242 def delete_hsm_configuration(params = {}, options = {}) req = build_request(:delete_hsm_configuration, params) req.send_request(options) end
Deletes a partner integration from a cluster. Data can still flow to the cluster until the integration is deleted at the partner's website.
@option params [required, String] :account_id
The Amazon Web Services account ID that owns the cluster.
@option params [required, String] :cluster_identifier
The cluster identifier of the cluster that receives data from the partner.
@option params [required, String] :database_name
The name of the database that receives data from the partner.
@option params [required, String] :partner_name
The name of the partner that is authorized to send data.
@return [Types::PartnerIntegrationOutputMessage] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::PartnerIntegrationOutputMessage#database_name #database_name} => String * {Types::PartnerIntegrationOutputMessage#partner_name #partner_name} => String
@example Request syntax with placeholder values
resp = client.delete_partner({ account_id: "PartnerIntegrationAccountId", # required cluster_identifier: "PartnerIntegrationClusterIdentifier", # required database_name: "PartnerIntegrationDatabaseName", # required partner_name: "PartnerIntegrationPartnerName", # required })
@example Response structure
resp.database_name #=> String resp.partner_name #=> String
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/DeletePartner AWS API Documentation
@overload delete_partner
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 3287 def delete_partner(params = {}, options = {}) req = build_request(:delete_partner, params) req.send_request(options) end
Deletes a scheduled action.
@option params [required, String] :scheduled_action_name
The name of the scheduled action to delete.
@return [Struct] Returns an empty {Seahorse::Client::Response response}.
@example Request syntax with placeholder values
resp = client.delete_scheduled_action({ scheduled_action_name: "String", # required })
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/DeleteScheduledAction AWS API Documentation
@overload delete_scheduled_action
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 3309 def delete_scheduled_action(params = {}, options = {}) req = build_request(:delete_scheduled_action, params) req.send_request(options) end
Deletes the specified snapshot copy grant.
@option params [required, String] :snapshot_copy_grant_name
The name of the snapshot copy grant to delete.
@return [Struct] Returns an empty {Seahorse::Client::Response response}.
@example Request syntax with placeholder values
resp = client.delete_snapshot_copy_grant({ snapshot_copy_grant_name: "String", # required })
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/DeleteSnapshotCopyGrant AWS API Documentation
@overload delete_snapshot_copy_grant
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 3331 def delete_snapshot_copy_grant(params = {}, options = {}) req = build_request(:delete_snapshot_copy_grant, params) req.send_request(options) end
Deletes a snapshot schedule.
@option params [required, String] :schedule_identifier
A unique identifier of the snapshot schedule to delete.
@return [Struct] Returns an empty {Seahorse::Client::Response response}.
@example Request syntax with placeholder values
resp = client.delete_snapshot_schedule({ schedule_identifier: "String", # required })
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/DeleteSnapshotSchedule AWS API Documentation
@overload delete_snapshot_schedule
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 3353 def delete_snapshot_schedule(params = {}, options = {}) req = build_request(:delete_snapshot_schedule, params) req.send_request(options) end
Deletes a usage limit from a cluster.
@option params [required, String] :usage_limit_id
The identifier of the usage limit to delete.
@return [Struct] Returns an empty {Seahorse::Client::Response response}.
@example Request syntax with placeholder values
resp = client.delete_usage_limit({ usage_limit_id: "String", # required })
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/DeleteUsageLimit AWS API Documentation
@overload delete_usage_limit
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 3404 def delete_usage_limit(params = {}, options = {}) req = build_request(:delete_usage_limit, params) req.send_request(options) end
Returns a list of attributes attached to an account
@option params [Array<String>] :attribute_names
A list of attribute names.
@return [Types::AccountAttributeList] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::AccountAttributeList#account_attributes #account_attributes} => Array<Types::AccountAttribute>
@example Request syntax with placeholder values
resp = client.describe_account_attributes({ attribute_names: ["String"], })
@example Response structure
resp.account_attributes #=> Array resp.account_attributes[0].attribute_name #=> String resp.account_attributes[0].attribute_values #=> Array resp.account_attributes[0].attribute_values[0].attribute_value #=> String
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/DescribeAccountAttributes AWS API Documentation
@overload describe_account_attributes
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 3435 def describe_account_attributes(params = {}, options = {}) req = build_request(:describe_account_attributes, params) req.send_request(options) end
Describes an authentication profile.
@option params [String] :authentication_profile_name
The name of the authentication profile to describe. If not specified then all authentication profiles owned by the account are listed.
@return [Types::DescribeAuthenticationProfilesResult] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::DescribeAuthenticationProfilesResult#authentication_profiles #authentication_profiles} => Array<Types::AuthenticationProfile>
@example Request syntax with placeholder values
resp = client.describe_authentication_profiles({ authentication_profile_name: "AuthenticationProfileNameString", })
@example Response structure
resp.authentication_profiles #=> Array resp.authentication_profiles[0].authentication_profile_name #=> String resp.authentication_profiles[0].authentication_profile_content #=> String
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/DescribeAuthenticationProfiles AWS API Documentation
@overload describe_authentication_profiles
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 3466 def describe_authentication_profiles(params = {}, options = {}) req = build_request(:describe_authentication_profiles, params) req.send_request(options) end
Returns an array of `ClusterDbRevision` objects.
@option params [String] :cluster_identifier
A unique identifier for a cluster whose `ClusterDbRevisions` you are requesting. This parameter is case sensitive. All clusters defined for an account are returned by default.
@option params [Integer] :max_records
The maximum number of response records to return in each call. If the number of remaining response records exceeds the specified MaxRecords value, a value is returned in the `marker` field of the response. You can retrieve the next set of response records by providing the returned `marker` value in the `marker` parameter and retrying the request. Default: 100 Constraints: minimum 20, maximum 100.
@option params [String] :marker
An optional parameter that specifies the starting point for returning a set of response records. When the results of a `DescribeClusterDbRevisions` request exceed the value specified in `MaxRecords`, Amazon Redshift returns a value in the `marker` field of the response. You can retrieve the next set of response records by providing the returned `marker` value in the `marker` parameter and retrying the request. Constraints: You can specify either the `ClusterIdentifier` parameter, or the `marker` parameter, but not both.
@return [Types::ClusterDbRevisionsMessage] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::ClusterDbRevisionsMessage#marker #marker} => String * {Types::ClusterDbRevisionsMessage#cluster_db_revisions #cluster_db_revisions} => Array<Types::ClusterDbRevision>
The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
@example Request syntax with placeholder values
resp = client.describe_cluster_db_revisions({ cluster_identifier: "String", max_records: 1, marker: "String", })
@example Response structure
resp.marker #=> String resp.cluster_db_revisions #=> Array resp.cluster_db_revisions[0].cluster_identifier #=> String resp.cluster_db_revisions[0].current_database_revision #=> String resp.cluster_db_revisions[0].database_revision_release_date #=> Time resp.cluster_db_revisions[0].revision_targets #=> Array resp.cluster_db_revisions[0].revision_targets[0].database_revision #=> String resp.cluster_db_revisions[0].revision_targets[0].description #=> String resp.cluster_db_revisions[0].revision_targets[0].database_revision_release_date #=> Time
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/DescribeClusterDbRevisions AWS API Documentation
@overload describe_cluster_db_revisions
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 3533 def describe_cluster_db_revisions(params = {}, options = {}) req = build_request(:describe_cluster_db_revisions, params) req.send_request(options) end
Returns a list of Amazon Redshift
parameter groups, including parameter groups you created and the default parameter group. For each parameter group, the response includes the parameter group name, description, and parameter group family name. You can optionally specify a name to retrieve the description of a specific parameter group.
For more information about parameters and parameter groups, go to
Management Guide*.
If you specify both tag keys and tag values in the same request, Amazon Redshift
returns all parameter groups that match any combination of the specified keys and values. For example, if you have `owner` and `environment` for tag keys, and `admin` and `test` for tag values, all parameter groups that have any combination of those values are returned.
If both tag keys and values are omitted from the request, parameter groups are returned regardless of whether they have tag keys or values associated with them.
[1]: docs.aws.amazon.com/redshift/latest/mgmt/working-with-parameter-groups.html
@option params [String] :parameter_group_name
The name of a specific parameter group for which to return details. By default, details about all parameter groups and the default parameter group are returned.
@option params [Integer] :max_records
The maximum number of response records to return in each call. If the number of remaining response records exceeds the specified `MaxRecords` value, a value is returned in a `marker` field of the response. You can retrieve the next set of records by retrying the command with the returned marker value. Default: `100` Constraints: minimum 20, maximum 100.
@option params [String] :marker
An optional parameter that specifies the starting point to return a set of response records. When the results of a DescribeClusterParameterGroups request exceed the value specified in `MaxRecords`, Amazon Web Services returns a value in the `Marker` field of the response. You can retrieve the next set of response records by providing the returned marker value in the `Marker` parameter and retrying the request.
@option params [Array<String>] :tag_keys
A tag key or keys for which you want to return all matching cluster parameter groups that are associated with the specified key or keys. For example, suppose that you have parameter groups that are tagged with keys called `owner` and `environment`. If you specify both of these tag keys in the request, Amazon Redshift returns a response with the parameter groups that have either or both of these tag keys associated with them.
@option params [Array<String>] :tag_values
A tag value or values for which you want to return all matching cluster parameter groups that are associated with the specified tag value or values. For example, suppose that you have parameter groups that are tagged with values called `admin` and `test`. If you specify both of these tag values in the request, Amazon Redshift returns a response with the parameter groups that have either or both of these tag values associated with them.
@return [Types::ClusterParameterGroupsMessage] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::ClusterParameterGroupsMessage#marker #marker} => String * {Types::ClusterParameterGroupsMessage#parameter_groups #parameter_groups} => Array<Types::ClusterParameterGroup>
The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
@example Request syntax with placeholder values
resp = client.describe_cluster_parameter_groups({ parameter_group_name: "String", max_records: 1, marker: "String", tag_keys: ["String"], tag_values: ["String"], })
@example Response structure
resp.marker #=> String resp.parameter_groups #=> Array resp.parameter_groups[0].parameter_group_name #=> String resp.parameter_groups[0].parameter_group_family #=> String resp.parameter_groups[0].description #=> String resp.parameter_groups[0].tags #=> Array resp.parameter_groups[0].tags[0].key #=> String resp.parameter_groups[0].tags[0].value #=> String
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/DescribeClusterParameterGroups AWS API Documentation
@overload describe_cluster_parameter_groups
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 3639 def describe_cluster_parameter_groups(params = {}, options = {}) req = build_request(:describe_cluster_parameter_groups, params) req.send_request(options) end
Returns a detailed list of parameters contained within the specified Amazon Redshift
parameter group. For each parameter the response includes information such as parameter name, description, data type, value, whether the parameter value is modifiable, and so on.
You can specify source filter to retrieve parameters of only specific type. For example, to retrieve parameters that were modified by a user action such as from ModifyClusterParameterGroup, you can specify source equal to user.
For more information about parameters and parameter groups, go to
Management Guide*.
[1]: docs.aws.amazon.com/redshift/latest/mgmt/working-with-parameter-groups.html
@option params [required, String] :parameter_group_name
The name of a cluster parameter group for which to return details.
@option params [String] :source
The parameter types to return. Specify `user` to show parameters that are different form the default. Similarly, specify `engine-default` to show parameters that are the same as the default parameter group. Default: All parameter types returned. Valid Values: `user` \| `engine-default`
@option params [Integer] :max_records
The maximum number of response records to return in each call. If the number of remaining response records exceeds the specified `MaxRecords` value, a value is returned in a `marker` field of the response. You can retrieve the next set of records by retrying the command with the returned marker value. Default: `100` Constraints: minimum 20, maximum 100.
@option params [String] :marker
An optional parameter that specifies the starting point to return a set of response records. When the results of a DescribeClusterParameters request exceed the value specified in `MaxRecords`, Amazon Web Services returns a value in the `Marker` field of the response. You can retrieve the next set of response records by providing the returned marker value in the `Marker` parameter and retrying the request.
@return [Types::ClusterParameterGroupDetails] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::ClusterParameterGroupDetails#parameters #parameters} => Array<Types::Parameter> * {Types::ClusterParameterGroupDetails#marker #marker} => String
The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
@example Request syntax with placeholder values
resp = client.describe_cluster_parameters({ parameter_group_name: "String", # required source: "String", max_records: 1, marker: "String", })
@example Response structure
resp.parameters #=> Array resp.parameters[0].parameter_name #=> String resp.parameters[0].parameter_value #=> String resp.parameters[0].description #=> String resp.parameters[0].source #=> String resp.parameters[0].data_type #=> String resp.parameters[0].allowed_values #=> String resp.parameters[0].apply_type #=> String, one of "static", "dynamic" resp.parameters[0].is_modifiable #=> Boolean resp.parameters[0].minimum_engine_version #=> String resp.marker #=> String
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/DescribeClusterParameters AWS API Documentation
@overload describe_cluster_parameters
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 3728 def describe_cluster_parameters(params = {}, options = {}) req = build_request(:describe_cluster_parameters, params) req.send_request(options) end
Returns information about Amazon Redshift
security groups. If the name of a security group is specified, the response will contain only information about only that security group.
For information about managing security groups, go to [Amazon Redshift
Cluster Security Groups] in the *Amazon Redshift
Cluster Management Guide*.
If you specify both tag keys and tag values in the same request, Amazon Redshift
returns all security groups that match any combination of the specified keys and values. For example, if you have `owner` and `environment` for tag keys, and `admin` and `test` for tag values, all security groups that have any combination of those values are returned.
If both tag keys and values are omitted from the request, security groups are returned regardless of whether they have tag keys or values associated with them.
[1]: docs.aws.amazon.com/redshift/latest/mgmt/working-with-security-groups.html
@option params [String] :cluster_security_group_name
The name of a cluster security group for which you are requesting details. You can specify either the **Marker** parameter or a **ClusterSecurityGroupName** parameter, but not both. Example: `securitygroup1`
@option params [Integer] :max_records
The maximum number of response records to return in each call. If the number of remaining response records exceeds the specified `MaxRecords` value, a value is returned in a `marker` field of the response. You can retrieve the next set of records by retrying the command with the returned marker value. Default: `100` Constraints: minimum 20, maximum 100.
@option params [String] :marker
An optional parameter that specifies the starting point to return a set of response records. When the results of a DescribeClusterSecurityGroups request exceed the value specified in `MaxRecords`, Amazon Web Services returns a value in the `Marker` field of the response. You can retrieve the next set of response records by providing the returned marker value in the `Marker` parameter and retrying the request. Constraints: You can specify either the **ClusterSecurityGroupName** parameter or the **Marker** parameter, but not both.
@option params [Array<String>] :tag_keys
A tag key or keys for which you want to return all matching cluster security groups that are associated with the specified key or keys. For example, suppose that you have security groups that are tagged with keys called `owner` and `environment`. If you specify both of these tag keys in the request, Amazon Redshift returns a response with the security groups that have either or both of these tag keys associated with them.
@option params [Array<String>] :tag_values
A tag value or values for which you want to return all matching cluster security groups that are associated with the specified tag value or values. For example, suppose that you have security groups that are tagged with values called `admin` and `test`. If you specify both of these tag values in the request, Amazon Redshift returns a response with the security groups that have either or both of these tag values associated with them.
@return [Types::ClusterSecurityGroupMessage] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::ClusterSecurityGroupMessage#marker #marker} => String * {Types::ClusterSecurityGroupMessage#cluster_security_groups #cluster_security_groups} => Array<Types::ClusterSecurityGroup>
The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
@example Request syntax with placeholder values
resp = client.describe_cluster_security_groups({ cluster_security_group_name: "String", max_records: 1, marker: "String", tag_keys: ["String"], tag_values: ["String"], })
@example Response structure
resp.marker #=> String resp.cluster_security_groups #=> Array resp.cluster_security_groups[0].cluster_security_group_name #=> String resp.cluster_security_groups[0].description #=> String resp.cluster_security_groups[0].ec2_security_groups #=> Array resp.cluster_security_groups[0].ec2_security_groups[0].status #=> String resp.cluster_security_groups[0].ec2_security_groups[0].ec2_security_group_name #=> String resp.cluster_security_groups[0].ec2_security_groups[0].ec2_security_group_owner_id #=> String resp.cluster_security_groups[0].ec2_security_groups[0].tags #=> Array resp.cluster_security_groups[0].ec2_security_groups[0].tags[0].key #=> String resp.cluster_security_groups[0].ec2_security_groups[0].tags[0].value #=> String resp.cluster_security_groups[0].ip_ranges #=> Array resp.cluster_security_groups[0].ip_ranges[0].status #=> String resp.cluster_security_groups[0].ip_ranges[0].cidrip #=> String resp.cluster_security_groups[0].ip_ranges[0].tags #=> Array resp.cluster_security_groups[0].ip_ranges[0].tags[0].key #=> String resp.cluster_security_groups[0].ip_ranges[0].tags[0].value #=> String resp.cluster_security_groups[0].tags #=> Array resp.cluster_security_groups[0].tags[0].key #=> String resp.cluster_security_groups[0].tags[0].value #=> String
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/DescribeClusterSecurityGroups AWS API Documentation
@overload describe_cluster_security_groups
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 3848 def describe_cluster_security_groups(params = {}, options = {}) req = build_request(:describe_cluster_security_groups, params) req.send_request(options) end
Returns one or more snapshot objects, which contain metadata about your cluster snapshots. By default, this operation returns information about all snapshots of all clusters that are owned by your Amazon Web Services account. No information is returned for snapshots owned by inactive Amazon Web Services accounts.
If you specify both tag keys and tag values in the same request, Amazon Redshift
returns all snapshots that match any combination of the specified keys and values. For example, if you have `owner` and `environment` for tag keys, and `admin` and `test` for tag values, all snapshots that have any combination of those values are returned. Only snapshots that you own are returned in the response; shared snapshots are not returned with the tag key and tag value request parameters.
If both tag keys and values are omitted from the request, snapshots are returned regardless of whether they have tag keys or values associated with them.
@option params [String] :cluster_identifier
The identifier of the cluster which generated the requested snapshots.
@option params [String] :snapshot_identifier
The snapshot identifier of the snapshot about which to return information.
@option params [String] :snapshot_type
The type of snapshots for which you are requesting information. By default, snapshots of all types are returned. Valid Values: `automated` \| `manual`
@option params [Time,DateTime,Date,Integer,String] :start_time
A value that requests only snapshots created at or after the specified time. The time value is specified in ISO 8601 format. For more information about ISO 8601, go to the [ISO8601 Wikipedia page.][1] Example: `2012-07-16T18:00:00Z` [1]: http://en.wikipedia.org/wiki/ISO_8601
@option params [Time,DateTime,Date,Integer,String] :end_time
A time value that requests only snapshots created at or before the specified time. The time value is specified in ISO 8601 format. For more information about ISO 8601, go to the [ISO8601 Wikipedia page.][1] Example: `2012-07-16T18:00:00Z` [1]: http://en.wikipedia.org/wiki/ISO_8601
@option params [Integer] :max_records
The maximum number of response records to return in each call. If the number of remaining response records exceeds the specified `MaxRecords` value, a value is returned in a `marker` field of the response. You can retrieve the next set of records by retrying the command with the returned marker value. Default: `100` Constraints: minimum 20, maximum 100.
@option params [String] :marker
An optional parameter that specifies the starting point to return a set of response records. When the results of a DescribeClusterSnapshots request exceed the value specified in `MaxRecords`, Amazon Web Services returns a value in the `Marker` field of the response. You can retrieve the next set of response records by providing the returned marker value in the `Marker` parameter and retrying the request.
@option params [String] :owner_account
The Amazon Web Services account used to create or copy the snapshot. Use this field to filter the results to snapshots owned by a particular account. To describe snapshots you own, either specify your Amazon Web Services account, or do not specify the parameter.
@option params [Array<String>] :tag_keys
A tag key or keys for which you want to return all matching cluster snapshots that are associated with the specified key or keys. For example, suppose that you have snapshots that are tagged with keys called `owner` and `environment`. If you specify both of these tag keys in the request, Amazon Redshift returns a response with the snapshots that have either or both of these tag keys associated with them.
@option params [Array<String>] :tag_values
A tag value or values for which you want to return all matching cluster snapshots that are associated with the specified tag value or values. For example, suppose that you have snapshots that are tagged with values called `admin` and `test`. If you specify both of these tag values in the request, Amazon Redshift returns a response with the snapshots that have either or both of these tag values associated with them.
@option params [Boolean] :cluster_exists
A value that indicates whether to return snapshots only for an existing cluster. You can perform table-level restore only by using a snapshot of an existing cluster, that is, a cluster that has not been deleted. Values for this parameter work as follows: * If `ClusterExists` is set to `true`, `ClusterIdentifier` is required. * If `ClusterExists` is set to `false` and `ClusterIdentifier` isn't specified, all snapshots associated with deleted clusters (orphaned snapshots) are returned. * If `ClusterExists` is set to `false` and `ClusterIdentifier` is specified for a deleted cluster, snapshots associated with that cluster are returned. * If `ClusterExists` is set to `false` and `ClusterIdentifier` is specified for an existing cluster, no snapshots are returned.
@option params [Array<Types::SnapshotSortingEntity>] :sorting_entities
@return [Types::SnapshotMessage] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::SnapshotMessage#marker #marker} => String * {Types::SnapshotMessage#snapshots #snapshots} => Array<Types::Snapshot>
The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
@example Request syntax with placeholder values
resp = client.describe_cluster_snapshots({ cluster_identifier: "String", snapshot_identifier: "String", snapshot_type: "String", start_time: Time.now, end_time: Time.now, max_records: 1, marker: "String", owner_account: "String", tag_keys: ["String"], tag_values: ["String"], cluster_exists: false, sorting_entities: [ { attribute: "SOURCE_TYPE", # required, accepts SOURCE_TYPE, TOTAL_SIZE, CREATE_TIME sort_order: "ASC", # accepts ASC, DESC }, ], })
@example Response structure
resp.marker #=> String resp.snapshots #=> Array resp.snapshots[0].snapshot_identifier #=> String resp.snapshots[0].cluster_identifier #=> String resp.snapshots[0].snapshot_create_time #=> Time resp.snapshots[0].status #=> String resp.snapshots[0].port #=> Integer resp.snapshots[0].availability_zone #=> String resp.snapshots[0].cluster_create_time #=> Time resp.snapshots[0].master_username #=> String resp.snapshots[0].cluster_version #=> String resp.snapshots[0].engine_full_version #=> String resp.snapshots[0].snapshot_type #=> String resp.snapshots[0].node_type #=> String resp.snapshots[0].number_of_nodes #=> Integer resp.snapshots[0].db_name #=> String resp.snapshots[0].vpc_id #=> String resp.snapshots[0].encrypted #=> Boolean resp.snapshots[0].kms_key_id #=> String resp.snapshots[0].encrypted_with_hsm #=> Boolean resp.snapshots[0].accounts_with_restore_access #=> Array resp.snapshots[0].accounts_with_restore_access[0].account_id #=> String resp.snapshots[0].accounts_with_restore_access[0].account_alias #=> String resp.snapshots[0].owner_account #=> String resp.snapshots[0].total_backup_size_in_mega_bytes #=> Float resp.snapshots[0].actual_incremental_backup_size_in_mega_bytes #=> Float resp.snapshots[0].backup_progress_in_mega_bytes #=> Float resp.snapshots[0].current_backup_rate_in_mega_bytes_per_second #=> Float resp.snapshots[0].estimated_seconds_to_completion #=> Integer resp.snapshots[0].elapsed_time_in_seconds #=> Integer resp.snapshots[0].source_region #=> String resp.snapshots[0].tags #=> Array resp.snapshots[0].tags[0].key #=> String resp.snapshots[0].tags[0].value #=> String resp.snapshots[0].restorable_node_types #=> Array resp.snapshots[0].restorable_node_types[0] #=> String resp.snapshots[0].enhanced_vpc_routing #=> Boolean resp.snapshots[0].maintenance_track_name #=> String resp.snapshots[0].manual_snapshot_retention_period #=> Integer resp.snapshots[0].manual_snapshot_remaining_days #=> Integer resp.snapshots[0].snapshot_retention_start_time #=> Time
The following waiters are defined for this operation (see {Client#wait_until} for detailed usage):
* snapshot_available
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/DescribeClusterSnapshots AWS API Documentation
@overload describe_cluster_snapshots
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 4055 def describe_cluster_snapshots(params = {}, options = {}) req = build_request(:describe_cluster_snapshots, params) req.send_request(options) end
Returns one or more cluster subnet group objects, which contain metadata about your cluster subnet groups. By default, this operation returns information about all cluster subnet groups that are defined in your Amazon Web Services account.
If you specify both tag keys and tag values in the same request, Amazon Redshift
returns all subnet groups that match any combination of the specified keys and values. For example, if you have `owner` and `environment` for tag keys, and `admin` and `test` for tag values, all subnet groups that have any combination of those values are returned.
If both tag keys and values are omitted from the request, subnet groups are returned regardless of whether they have tag keys or values associated with them.
@option params [String] :cluster_subnet_group_name
The name of the cluster subnet group for which information is requested.
@option params [Integer] :max_records
The maximum number of response records to return in each call. If the number of remaining response records exceeds the specified `MaxRecords` value, a value is returned in a `marker` field of the response. You can retrieve the next set of records by retrying the command with the returned marker value. Default: `100` Constraints: minimum 20, maximum 100.
@option params [String] :marker
An optional parameter that specifies the starting point to return a set of response records. When the results of a DescribeClusterSubnetGroups request exceed the value specified in `MaxRecords`, Amazon Web Services returns a value in the `Marker` field of the response. You can retrieve the next set of response records by providing the returned marker value in the `Marker` parameter and retrying the request.
@option params [Array<String>] :tag_keys
A tag key or keys for which you want to return all matching cluster subnet groups that are associated with the specified key or keys. For example, suppose that you have subnet groups that are tagged with keys called `owner` and `environment`. If you specify both of these tag keys in the request, Amazon Redshift returns a response with the subnet groups that have either or both of these tag keys associated with them.
@option params [Array<String>] :tag_values
A tag value or values for which you want to return all matching cluster subnet groups that are associated with the specified tag value or values. For example, suppose that you have subnet groups that are tagged with values called `admin` and `test`. If you specify both of these tag values in the request, Amazon Redshift returns a response with the subnet groups that have either or both of these tag values associated with them.
@return [Types::ClusterSubnetGroupMessage] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::ClusterSubnetGroupMessage#marker #marker} => String * {Types::ClusterSubnetGroupMessage#cluster_subnet_groups #cluster_subnet_groups} => Array<Types::ClusterSubnetGroup>
The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
@example Request syntax with placeholder values
resp = client.describe_cluster_subnet_groups({ cluster_subnet_group_name: "String", max_records: 1, marker: "String", tag_keys: ["String"], tag_values: ["String"], })
@example Response structure
resp.marker #=> String resp.cluster_subnet_groups #=> Array resp.cluster_subnet_groups[0].cluster_subnet_group_name #=> String resp.cluster_subnet_groups[0].description #=> String resp.cluster_subnet_groups[0].vpc_id #=> String resp.cluster_subnet_groups[0].subnet_group_status #=> String resp.cluster_subnet_groups[0].subnets #=> Array resp.cluster_subnet_groups[0].subnets[0].subnet_identifier #=> String resp.cluster_subnet_groups[0].subnets[0].subnet_availability_zone.name #=> String resp.cluster_subnet_groups[0].subnets[0].subnet_availability_zone.supported_platforms #=> Array resp.cluster_subnet_groups[0].subnets[0].subnet_availability_zone.supported_platforms[0].name #=> String resp.cluster_subnet_groups[0].subnets[0].subnet_status #=> String resp.cluster_subnet_groups[0].tags #=> Array resp.cluster_subnet_groups[0].tags[0].key #=> String resp.cluster_subnet_groups[0].tags[0].value #=> String
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/DescribeClusterSubnetGroups AWS API Documentation
@overload describe_cluster_subnet_groups
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 4156 def describe_cluster_subnet_groups(params = {}, options = {}) req = build_request(:describe_cluster_subnet_groups, params) req.send_request(options) end
Returns a list of all the available maintenance tracks.
@option params [String] :maintenance_track_name
The name of the maintenance track.
@option params [Integer] :max_records
An integer value for the maximum number of maintenance tracks to return.
@option params [String] :marker
An optional parameter that specifies the starting point to return a set of response records. When the results of a `DescribeClusterTracks` request exceed the value specified in `MaxRecords`, Amazon Redshift returns a value in the `Marker` field of the response. You can retrieve the next set of response records by providing the returned marker value in the `Marker` parameter and retrying the request.
@return [Types::TrackListMessage] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::TrackListMessage#maintenance_tracks #maintenance_tracks} => Array<Types::MaintenanceTrack> * {Types::TrackListMessage#marker #marker} => String
The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
@example Request syntax with placeholder values
resp = client.describe_cluster_tracks({ maintenance_track_name: "String", max_records: 1, marker: "String", })
@example Response structure
resp.maintenance_tracks #=> Array resp.maintenance_tracks[0].maintenance_track_name #=> String resp.maintenance_tracks[0].database_version #=> String resp.maintenance_tracks[0].update_targets #=> Array resp.maintenance_tracks[0].update_targets[0].maintenance_track_name #=> String resp.maintenance_tracks[0].update_targets[0].database_version #=> String resp.maintenance_tracks[0].update_targets[0].supported_operations #=> Array resp.maintenance_tracks[0].update_targets[0].supported_operations[0].operation_name #=> String resp.marker #=> String
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/DescribeClusterTracks AWS API Documentation
@overload describe_cluster_tracks
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 4209 def describe_cluster_tracks(params = {}, options = {}) req = build_request(:describe_cluster_tracks, params) req.send_request(options) end
Returns descriptions of the available Amazon Redshift
cluster versions. You can call this operation even before creating any clusters to learn more about the Amazon Redshift
versions. For more information about managing clusters, go to [Amazon Redshift
Clusters] in the *Amazon Redshift
Cluster Management Guide*.
[1]: docs.aws.amazon.com/redshift/latest/mgmt/working-with-clusters.html
@option params [String] :cluster_version
The specific cluster version to return. Example: `1.0`
@option params [String] :cluster_parameter_group_family
The name of a specific cluster parameter group family to return details for. Constraints: * Must be 1 to 255 alphanumeric characters * First character must be a letter * Cannot end with a hyphen or contain two consecutive hyphens
@option params [Integer] :max_records
The maximum number of response records to return in each call. If the number of remaining response records exceeds the specified `MaxRecords` value, a value is returned in a `marker` field of the response. You can retrieve the next set of records by retrying the command with the returned marker value. Default: `100` Constraints: minimum 20, maximum 100.
@option params [String] :marker
An optional parameter that specifies the starting point to return a set of response records. When the results of a DescribeClusterVersions request exceed the value specified in `MaxRecords`, Amazon Web Services returns a value in the `Marker` field of the response. You can retrieve the next set of response records by providing the returned marker value in the `Marker` parameter and retrying the request.
@return [Types::ClusterVersionsMessage] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::ClusterVersionsMessage#marker #marker} => String * {Types::ClusterVersionsMessage#cluster_versions #cluster_versions} => Array<Types::ClusterVersion>
The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
@example Request syntax with placeholder values
resp = client.describe_cluster_versions({ cluster_version: "String", cluster_parameter_group_family: "String", max_records: 1, marker: "String", })
@example Response structure
resp.marker #=> String resp.cluster_versions #=> Array resp.cluster_versions[0].cluster_version #=> String resp.cluster_versions[0].cluster_parameter_group_family #=> String resp.cluster_versions[0].description #=> String
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/DescribeClusterVersions AWS API Documentation
@overload describe_cluster_versions
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 4289 def describe_cluster_versions(params = {}, options = {}) req = build_request(:describe_cluster_versions, params) req.send_request(options) end
Returns properties of provisioned clusters including general cluster properties, cluster database properties, maintenance and backup properties, and security and access properties. This operation supports pagination. For more information about managing clusters, go to [Amazon Redshift
Clusters] in the *Amazon Redshift
Cluster Management Guide*.
If you specify both tag keys and tag values in the same request, Amazon Redshift
returns all clusters that match any combination of the specified keys and values. For example, if you have `owner` and `environment` for tag keys, and `admin` and `test` for tag values, all clusters that have any combination of those values are returned.
If both tag keys and values are omitted from the request, clusters are returned regardless of whether they have tag keys or values associated with them.
[1]: docs.aws.amazon.com/redshift/latest/mgmt/working-with-clusters.html
@option params [String] :cluster_identifier
The unique identifier of a cluster whose properties you are requesting. This parameter is case sensitive. The default is that all clusters defined for an account are returned.
@option params [Integer] :max_records
The maximum number of response records to return in each call. If the number of remaining response records exceeds the specified `MaxRecords` value, a value is returned in a `marker` field of the response. You can retrieve the next set of records by retrying the command with the returned marker value. Default: `100` Constraints: minimum 20, maximum 100.
@option params [String] :marker
An optional parameter that specifies the starting point to return a set of response records. When the results of a DescribeClusters request exceed the value specified in `MaxRecords`, Amazon Web Services returns a value in the `Marker` field of the response. You can retrieve the next set of response records by providing the returned marker value in the `Marker` parameter and retrying the request. Constraints: You can specify either the **ClusterIdentifier** parameter or the **Marker** parameter, but not both.
@option params [Array<String>] :tag_keys
A tag key or keys for which you want to return all matching clusters that are associated with the specified key or keys. For example, suppose that you have clusters that are tagged with keys called `owner` and `environment`. If you specify both of these tag keys in the request, Amazon Redshift returns a response with the clusters that have either or both of these tag keys associated with them.
@option params [Array<String>] :tag_values
A tag value or values for which you want to return all matching clusters that are associated with the specified tag value or values. For example, suppose that you have clusters that are tagged with values called `admin` and `test`. If you specify both of these tag values in the request, Amazon Redshift returns a response with the clusters that have either or both of these tag values associated with them.
@return [Types::ClustersMessage] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::ClustersMessage#marker #marker} => String * {Types::ClustersMessage#clusters #clusters} => Array<Types::Cluster>
The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
@example Request syntax with placeholder values
resp = client.describe_clusters({ cluster_identifier: "String", max_records: 1, marker: "String", tag_keys: ["String"], tag_values: ["String"], })
@example Response structure
resp.marker #=> String resp.clusters #=> Array resp.clusters[0].cluster_identifier #=> String resp.clusters[0].node_type #=> String resp.clusters[0].cluster_status #=> String resp.clusters[0].cluster_availability_status #=> String resp.clusters[0].modify_status #=> String resp.clusters[0].master_username #=> String resp.clusters[0].db_name #=> String resp.clusters[0].endpoint.address #=> String resp.clusters[0].endpoint.port #=> Integer resp.clusters[0].endpoint.vpc_endpoints #=> Array resp.clusters[0].endpoint.vpc_endpoints[0].vpc_endpoint_id #=> String resp.clusters[0].endpoint.vpc_endpoints[0].vpc_id #=> String resp.clusters[0].endpoint.vpc_endpoints[0].network_interfaces #=> Array resp.clusters[0].endpoint.vpc_endpoints[0].network_interfaces[0].network_interface_id #=> String resp.clusters[0].endpoint.vpc_endpoints[0].network_interfaces[0].subnet_id #=> String resp.clusters[0].endpoint.vpc_endpoints[0].network_interfaces[0].private_ip_address #=> String resp.clusters[0].endpoint.vpc_endpoints[0].network_interfaces[0].availability_zone #=> String resp.clusters[0].cluster_create_time #=> Time resp.clusters[0].automated_snapshot_retention_period #=> Integer resp.clusters[0].manual_snapshot_retention_period #=> Integer resp.clusters[0].cluster_security_groups #=> Array resp.clusters[0].cluster_security_groups[0].cluster_security_group_name #=> String resp.clusters[0].cluster_security_groups[0].status #=> String resp.clusters[0].vpc_security_groups #=> Array resp.clusters[0].vpc_security_groups[0].vpc_security_group_id #=> String resp.clusters[0].vpc_security_groups[0].status #=> String resp.clusters[0].cluster_parameter_groups #=> Array resp.clusters[0].cluster_parameter_groups[0].parameter_group_name #=> String resp.clusters[0].cluster_parameter_groups[0].parameter_apply_status #=> String resp.clusters[0].cluster_parameter_groups[0].cluster_parameter_status_list #=> Array resp.clusters[0].cluster_parameter_groups[0].cluster_parameter_status_list[0].parameter_name #=> String resp.clusters[0].cluster_parameter_groups[0].cluster_parameter_status_list[0].parameter_apply_status #=> String resp.clusters[0].cluster_parameter_groups[0].cluster_parameter_status_list[0].parameter_apply_error_description #=> String resp.clusters[0].cluster_subnet_group_name #=> String resp.clusters[0].vpc_id #=> String resp.clusters[0].availability_zone #=> String resp.clusters[0].preferred_maintenance_window #=> String resp.clusters[0].pending_modified_values.master_user_password #=> String resp.clusters[0].pending_modified_values.node_type #=> String resp.clusters[0].pending_modified_values.number_of_nodes #=> Integer resp.clusters[0].pending_modified_values.cluster_type #=> String resp.clusters[0].pending_modified_values.cluster_version #=> String resp.clusters[0].pending_modified_values.automated_snapshot_retention_period #=> Integer resp.clusters[0].pending_modified_values.cluster_identifier #=> String resp.clusters[0].pending_modified_values.publicly_accessible #=> Boolean resp.clusters[0].pending_modified_values.enhanced_vpc_routing #=> Boolean resp.clusters[0].pending_modified_values.maintenance_track_name #=> String resp.clusters[0].pending_modified_values.encryption_type #=> String resp.clusters[0].cluster_version #=> String resp.clusters[0].allow_version_upgrade #=> Boolean resp.clusters[0].number_of_nodes #=> Integer resp.clusters[0].publicly_accessible #=> Boolean resp.clusters[0].encrypted #=> Boolean resp.clusters[0].restore_status.status #=> String resp.clusters[0].restore_status.current_restore_rate_in_mega_bytes_per_second #=> Float resp.clusters[0].restore_status.snapshot_size_in_mega_bytes #=> Integer resp.clusters[0].restore_status.progress_in_mega_bytes #=> Integer resp.clusters[0].restore_status.elapsed_time_in_seconds #=> Integer resp.clusters[0].restore_status.estimated_time_to_completion_in_seconds #=> Integer resp.clusters[0].data_transfer_progress.status #=> String resp.clusters[0].data_transfer_progress.current_rate_in_mega_bytes_per_second #=> Float resp.clusters[0].data_transfer_progress.total_data_in_mega_bytes #=> Integer resp.clusters[0].data_transfer_progress.data_transferred_in_mega_bytes #=> Integer resp.clusters[0].data_transfer_progress.estimated_time_to_completion_in_seconds #=> Integer resp.clusters[0].data_transfer_progress.elapsed_time_in_seconds #=> Integer resp.clusters[0].hsm_status.hsm_client_certificate_identifier #=> String resp.clusters[0].hsm_status.hsm_configuration_identifier #=> String resp.clusters[0].hsm_status.status #=> String resp.clusters[0].cluster_snapshot_copy_status.destination_region #=> String resp.clusters[0].cluster_snapshot_copy_status.retention_period #=> Integer resp.clusters[0].cluster_snapshot_copy_status.manual_snapshot_retention_period #=> Integer resp.clusters[0].cluster_snapshot_copy_status.snapshot_copy_grant_name #=> String resp.clusters[0].cluster_public_key #=> String resp.clusters[0].cluster_nodes #=> Array resp.clusters[0].cluster_nodes[0].node_role #=> String resp.clusters[0].cluster_nodes[0].private_ip_address #=> String resp.clusters[0].cluster_nodes[0].public_ip_address #=> String resp.clusters[0].elastic_ip_status.elastic_ip #=> String resp.clusters[0].elastic_ip_status.status #=> String resp.clusters[0].cluster_revision_number #=> String resp.clusters[0].tags #=> Array resp.clusters[0].tags[0].key #=> String resp.clusters[0].tags[0].value #=> String resp.clusters[0].kms_key_id #=> String resp.clusters[0].enhanced_vpc_routing #=> Boolean resp.clusters[0].iam_roles #=> Array resp.clusters[0].iam_roles[0].iam_role_arn #=> String resp.clusters[0].iam_roles[0].apply_status #=> String resp.clusters[0].pending_actions #=> Array resp.clusters[0].pending_actions[0] #=> String resp.clusters[0].maintenance_track_name #=> String resp.clusters[0].elastic_resize_number_of_node_options #=> String resp.clusters[0].deferred_maintenance_windows #=> Array resp.clusters[0].deferred_maintenance_windows[0].defer_maintenance_identifier #=> String resp.clusters[0].deferred_maintenance_windows[0].defer_maintenance_start_time #=> Time resp.clusters[0].deferred_maintenance_windows[0].defer_maintenance_end_time #=> Time resp.clusters[0].snapshot_schedule_identifier #=> String resp.clusters[0].snapshot_schedule_state #=> String, one of "MODIFYING", "ACTIVE", "FAILED" resp.clusters[0].expected_next_snapshot_schedule_time #=> Time resp.clusters[0].expected_next_snapshot_schedule_time_status #=> String resp.clusters[0].next_maintenance_window_start_time #=> Time resp.clusters[0].resize_info.resize_type #=> String resp.clusters[0].resize_info.allow_cancel_resize #=> Boolean resp.clusters[0].availability_zone_relocation_status #=> String resp.clusters[0].cluster_namespace_arn #=> String resp.clusters[0].total_storage_capacity_in_mega_bytes #=> Integer resp.clusters[0].aqua_configuration.aqua_status #=> String, one of "enabled", "disabled", "applying" resp.clusters[0].aqua_configuration.aqua_configuration_status #=> String, one of "enabled", "disabled", "auto"
The following waiters are defined for this operation (see {Client#wait_until} for detailed usage):
* cluster_available * cluster_deleted * cluster_restored
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/DescribeClusters AWS API Documentation
@overload describe_clusters
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 4502 def describe_clusters(params = {}, options = {}) req = build_request(:describe_clusters, params) req.send_request(options) end
Returns a list of parameter settings for the specified parameter group family.
For more information about parameters and parameter groups, go to
Management Guide*.
[1]: docs.aws.amazon.com/redshift/latest/mgmt/working-with-parameter-groups.html
@option params [required, String] :parameter_group_family
The name of the cluster parameter group family.
@option params [Integer] :max_records
The maximum number of response records to return in each call. If the number of remaining response records exceeds the specified `MaxRecords` value, a value is returned in a `marker` field of the response. You can retrieve the next set of records by retrying the command with the returned marker value. Default: `100` Constraints: minimum 20, maximum 100.
@option params [String] :marker
An optional parameter that specifies the starting point to return a set of response records. When the results of a DescribeDefaultClusterParameters request exceed the value specified in `MaxRecords`, Amazon Web Services returns a value in the `Marker` field of the response. You can retrieve the next set of response records by providing the returned marker value in the `Marker` parameter and retrying the request.
@return [Types::DescribeDefaultClusterParametersResult] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::DescribeDefaultClusterParametersResult#default_cluster_parameters #default_cluster_parameters} => Types::DefaultClusterParameters
The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
@example Request syntax with placeholder values
resp = client.describe_default_cluster_parameters({ parameter_group_family: "String", # required max_records: 1, marker: "String", })
@example Response structure
resp.default_cluster_parameters.parameter_group_family #=> String resp.default_cluster_parameters.marker #=> String resp.default_cluster_parameters.parameters #=> Array resp.default_cluster_parameters.parameters[0].parameter_name #=> String resp.default_cluster_parameters.parameters[0].parameter_value #=> String resp.default_cluster_parameters.parameters[0].description #=> String resp.default_cluster_parameters.parameters[0].source #=> String resp.default_cluster_parameters.parameters[0].data_type #=> String resp.default_cluster_parameters.parameters[0].allowed_values #=> String resp.default_cluster_parameters.parameters[0].apply_type #=> String, one of "static", "dynamic" resp.default_cluster_parameters.parameters[0].is_modifiable #=> Boolean resp.default_cluster_parameters.parameters[0].minimum_engine_version #=> String
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/DescribeDefaultClusterParameters AWS API Documentation
@overload describe_default_cluster_parameters
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 4758 def describe_default_cluster_parameters(params = {}, options = {}) req = build_request(:describe_default_cluster_parameters, params) req.send_request(options) end
Describes a Redshift-managed VPC endpoint.
@option params [String] :cluster_identifier
The cluster identifier associated with the described endpoint.
@option params [String] :resource_owner
The Amazon Web Services account ID of the owner of the cluster.
@option params [String] :endpoint_name
The name of the endpoint to be described.
@option params [String] :vpc_id
The virtual private cloud (VPC) identifier with access to the cluster.
@option params [Integer] :max_records
The maximum number of records to include in the response. If more records exist than the specified `MaxRecords` value, a pagination token called a `Marker` is included in the response so that the remaining results can be retrieved.
@option params [String] :marker
An optional pagination token provided by a previous `DescribeEndpointAccess` request. If this parameter is specified, the response includes only records beyond the marker, up to the value specified by the `MaxRecords` parameter.
@return [Types::EndpointAccessList] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::EndpointAccessList#endpoint_access_list #endpoint_access_list} => Array<Types::EndpointAccess> * {Types::EndpointAccessList#marker #marker} => String
The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
@example Request syntax with placeholder values
resp = client.describe_endpoint_access({ cluster_identifier: "String", resource_owner: "String", endpoint_name: "String", vpc_id: "String", max_records: 1, marker: "String", })
@example Response structure
resp.endpoint_access_list #=> Array resp.endpoint_access_list[0].cluster_identifier #=> String resp.endpoint_access_list[0].resource_owner #=> String resp.endpoint_access_list[0].subnet_group_name #=> String resp.endpoint_access_list[0].endpoint_status #=> String resp.endpoint_access_list[0].endpoint_name #=> String resp.endpoint_access_list[0].endpoint_create_time #=> Time resp.endpoint_access_list[0].port #=> Integer resp.endpoint_access_list[0].address #=> String resp.endpoint_access_list[0].vpc_security_groups #=> Array resp.endpoint_access_list[0].vpc_security_groups[0].vpc_security_group_id #=> String resp.endpoint_access_list[0].vpc_security_groups[0].status #=> String resp.endpoint_access_list[0].vpc_endpoint.vpc_endpoint_id #=> String resp.endpoint_access_list[0].vpc_endpoint.vpc_id #=> String resp.endpoint_access_list[0].vpc_endpoint.network_interfaces #=> Array resp.endpoint_access_list[0].vpc_endpoint.network_interfaces[0].network_interface_id #=> String resp.endpoint_access_list[0].vpc_endpoint.network_interfaces[0].subnet_id #=> String resp.endpoint_access_list[0].vpc_endpoint.network_interfaces[0].private_ip_address #=> String resp.endpoint_access_list[0].vpc_endpoint.network_interfaces[0].availability_zone #=> String resp.marker #=> String
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/DescribeEndpointAccess AWS API Documentation
@overload describe_endpoint_access
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 4834 def describe_endpoint_access(params = {}, options = {}) req = build_request(:describe_endpoint_access, params) req.send_request(options) end
Displays a list of event categories for all event source types, or for a specified source type. For a list of the event categories and source types, go to [Amazon Redshift
Event Notifications].
[1]: docs.aws.amazon.com/redshift/latest/mgmt/working-with-event-notifications.html
@option params [String] :source_type
The source type, such as cluster or parameter group, to which the described event categories apply. Valid values: cluster, cluster-snapshot, cluster-parameter-group, cluster-security-group, and scheduled-action.
@return [Types::EventCategoriesMessage] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::EventCategoriesMessage#event_categories_map_list #event_categories_map_list} => Array<Types::EventCategoriesMap>
@example Request syntax with placeholder values
resp = client.describe_event_categories({ source_type: "String", })
@example Response structure
resp.event_categories_map_list #=> Array resp.event_categories_map_list[0].source_type #=> String resp.event_categories_map_list[0].events #=> Array resp.event_categories_map_list[0].events[0].event_id #=> String resp.event_categories_map_list[0].events[0].event_categories #=> Array resp.event_categories_map_list[0].events[0].event_categories[0] #=> String resp.event_categories_map_list[0].events[0].event_description #=> String resp.event_categories_map_list[0].events[0].severity #=> String
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/DescribeEventCategories AWS API Documentation
@overload describe_event_categories
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 4948 def describe_event_categories(params = {}, options = {}) req = build_request(:describe_event_categories, params) req.send_request(options) end
Lists descriptions of all the Amazon Redshift
event notification subscriptions for a customer account. If you specify a subscription name, lists the description for that subscription.
If you specify both tag keys and tag values in the same request, Amazon Redshift
returns all event notification subscriptions that match any combination of the specified keys and values. For example, if you have `owner` and `environment` for tag keys, and `admin` and `test` for tag values, all subscriptions that have any combination of those values are returned.
If both tag keys and values are omitted from the request, subscriptions are returned regardless of whether they have tag keys or values associated with them.
@option params [String] :subscription_name
The name of the Amazon Redshift event notification subscription to be described.
@option params [Integer] :max_records
The maximum number of response records to return in each call. If the number of remaining response records exceeds the specified `MaxRecords` value, a value is returned in a `marker` field of the response. You can retrieve the next set of records by retrying the command with the returned marker value. Default: `100` Constraints: minimum 20, maximum 100.
@option params [String] :marker
An optional parameter that specifies the starting point to return a set of response records. When the results of a DescribeEventSubscriptions request exceed the value specified in `MaxRecords`, Amazon Web Services returns a value in the `Marker` field of the response. You can retrieve the next set of response records by providing the returned marker value in the `Marker` parameter and retrying the request.
@option params [Array<String>] :tag_keys
A tag key or keys for which you want to return all matching event notification subscriptions that are associated with the specified key or keys. For example, suppose that you have subscriptions that are tagged with keys called `owner` and `environment`. If you specify both of these tag keys in the request, Amazon Redshift returns a response with the subscriptions that have either or both of these tag keys associated with them.
@option params [Array<String>] :tag_values
A tag value or values for which you want to return all matching event notification subscriptions that are associated with the specified tag value or values. For example, suppose that you have subscriptions that are tagged with values called `admin` and `test`. If you specify both of these tag values in the request, Amazon Redshift returns a response with the subscriptions that have either or both of these tag values associated with them.
@return [Types::EventSubscriptionsMessage] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::EventSubscriptionsMessage#marker #marker} => String * {Types::EventSubscriptionsMessage#event_subscriptions_list #event_subscriptions_list} => Array<Types::EventSubscription>
The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
@example Request syntax with placeholder values
resp = client.describe_event_subscriptions({ subscription_name: "String", max_records: 1, marker: "String", tag_keys: ["String"], tag_values: ["String"], })
@example Response structure
resp.marker #=> String resp.event_subscriptions_list #=> Array resp.event_subscriptions_list[0].customer_aws_id #=> String resp.event_subscriptions_list[0].cust_subscription_id #=> String resp.event_subscriptions_list[0].sns_topic_arn #=> String resp.event_subscriptions_list[0].status #=> String resp.event_subscriptions_list[0].subscription_creation_time #=> Time resp.event_subscriptions_list[0].source_type #=> String resp.event_subscriptions_list[0].source_ids_list #=> Array resp.event_subscriptions_list[0].source_ids_list[0] #=> String resp.event_subscriptions_list[0].event_categories_list #=> Array resp.event_subscriptions_list[0].event_categories_list[0] #=> String resp.event_subscriptions_list[0].severity #=> String resp.event_subscriptions_list[0].enabled #=> Boolean resp.event_subscriptions_list[0].tags #=> Array resp.event_subscriptions_list[0].tags[0].key #=> String resp.event_subscriptions_list[0].tags[0].value #=> String
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/DescribeEventSubscriptions AWS API Documentation
@overload describe_event_subscriptions
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 5051 def describe_event_subscriptions(params = {}, options = {}) req = build_request(:describe_event_subscriptions, params) req.send_request(options) end
Returns events related to clusters, security groups, snapshots, and parameter groups for the past 14 days. Events specific to a particular cluster, security group, snapshot or parameter group can be obtained by providing the name as a parameter. By default, the past hour of events are returned.
@option params [String] :source_identifier
The identifier of the event source for which events will be returned. If this parameter is not specified, then all sources are included in the response. Constraints: If *SourceIdentifier* is supplied, *SourceType* must also be provided. * Specify a cluster identifier when *SourceType* is `cluster`. * Specify a cluster security group name when *SourceType* is `cluster-security-group`. * Specify a cluster parameter group name when *SourceType* is `cluster-parameter-group`. * Specify a cluster snapshot identifier when *SourceType* is `cluster-snapshot`.
@option params [String] :source_type
The event source to retrieve events for. If no value is specified, all events are returned. Constraints: If *SourceType* is supplied, *SourceIdentifier* must also be provided. * Specify `cluster` when *SourceIdentifier* is a cluster identifier. * Specify `cluster-security-group` when *SourceIdentifier* is a cluster security group name. * Specify `cluster-parameter-group` when *SourceIdentifier* is a cluster parameter group name. * Specify `cluster-snapshot` when *SourceIdentifier* is a cluster snapshot identifier.
@option params [Time,DateTime,Date,Integer,String] :start_time
The beginning of the time interval to retrieve events for, specified in ISO 8601 format. For more information about ISO 8601, go to the [ISO8601 Wikipedia page.][1] Example: `2009-07-08T18:00Z` [1]: http://en.wikipedia.org/wiki/ISO_8601
@option params [Time,DateTime,Date,Integer,String] :end_time
The end of the time interval for which to retrieve events, specified in ISO 8601 format. For more information about ISO 8601, go to the [ISO8601 Wikipedia page.][1] Example: `2009-07-08T18:00Z` [1]: http://en.wikipedia.org/wiki/ISO_8601
@option params [Integer] :duration
The number of minutes prior to the time of the request for which to retrieve events. For example, if the request is sent at 18:00 and you specify a duration of 60, then only events which have occurred after 17:00 will be returned. Default: `60`
@option params [Integer] :max_records
The maximum number of response records to return in each call. If the number of remaining response records exceeds the specified `MaxRecords` value, a value is returned in a `marker` field of the response. You can retrieve the next set of records by retrying the command with the returned marker value. Default: `100` Constraints: minimum 20, maximum 100.
@option params [String] :marker
An optional parameter that specifies the starting point to return a set of response records. When the results of a DescribeEvents request exceed the value specified in `MaxRecords`, Amazon Web Services returns a value in the `Marker` field of the response. You can retrieve the next set of response records by providing the returned marker value in the `Marker` parameter and retrying the request.
@return [Types::EventsMessage] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::EventsMessage#marker #marker} => String * {Types::EventsMessage#events #events} => Array<Types::Event>
The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
@example Request syntax with placeholder values
resp = client.describe_events({ source_identifier: "String", source_type: "cluster", # accepts cluster, cluster-parameter-group, cluster-security-group, cluster-snapshot, scheduled-action start_time: Time.now, end_time: Time.now, duration: 1, max_records: 1, marker: "String", })
@example Response structure
resp.marker #=> String resp.events #=> Array resp.events[0].source_identifier #=> String resp.events[0].source_type #=> String, one of "cluster", "cluster-parameter-group", "cluster-security-group", "cluster-snapshot", "scheduled-action" resp.events[0].message #=> String resp.events[0].event_categories #=> Array resp.events[0].event_categories[0] #=> String resp.events[0].severity #=> String resp.events[0].date #=> Time resp.events[0].event_id #=> String
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/DescribeEvents AWS API Documentation
@overload describe_events
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 5186 def describe_events(params = {}, options = {}) req = build_request(:describe_events, params) req.send_request(options) end
Returns information about the specified HSM client certificate. If no certificate ID is specified, returns information about all the HSM certificates owned by your Amazon Web Services account.
If you specify both tag keys and tag values in the same request, Amazon Redshift
returns all HSM client certificates that match any combination of the specified keys and values. For example, if you have `owner` and `environment` for tag keys, and `admin` and `test` for tag values, all HSM client certificates that have any combination of those values are returned.
If both tag keys and values are omitted from the request, HSM client certificates are returned regardless of whether they have tag keys or values associated with them.
@option params [String] :hsm_client_certificate_identifier
The identifier of a specific HSM client certificate for which you want information. If no identifier is specified, information is returned for all HSM client certificates owned by your Amazon Web Services account.
@option params [Integer] :max_records
The maximum number of response records to return in each call. If the number of remaining response records exceeds the specified `MaxRecords` value, a value is returned in a `marker` field of the response. You can retrieve the next set of records by retrying the command with the returned marker value. Default: `100` Constraints: minimum 20, maximum 100.
@option params [String] :marker
An optional parameter that specifies the starting point to return a set of response records. When the results of a DescribeHsmClientCertificates request exceed the value specified in `MaxRecords`, Amazon Web Services returns a value in the `Marker` field of the response. You can retrieve the next set of response records by providing the returned marker value in the `Marker` parameter and retrying the request.
@option params [Array<String>] :tag_keys
A tag key or keys for which you want to return all matching HSM client certificates that are associated with the specified key or keys. For example, suppose that you have HSM client certificates that are tagged with keys called `owner` and `environment`. If you specify both of these tag keys in the request, Amazon Redshift returns a response with the HSM client certificates that have either or both of these tag keys associated with them.
@option params [Array<String>] :tag_values
A tag value or values for which you want to return all matching HSM client certificates that are associated with the specified tag value or values. For example, suppose that you have HSM client certificates that are tagged with values called `admin` and `test`. If you specify both of these tag values in the request, Amazon Redshift returns a response with the HSM client certificates that have either or both of these tag values associated with them.
@return [Types::HsmClientCertificateMessage] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::HsmClientCertificateMessage#marker #marker} => String * {Types::HsmClientCertificateMessage#hsm_client_certificates #hsm_client_certificates} => Array<Types::HsmClientCertificate>
The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
@example Request syntax with placeholder values
resp = client.describe_hsm_client_certificates({ hsm_client_certificate_identifier: "String", max_records: 1, marker: "String", tag_keys: ["String"], tag_values: ["String"], })
@example Response structure
resp.marker #=> String resp.hsm_client_certificates #=> Array resp.hsm_client_certificates[0].hsm_client_certificate_identifier #=> String resp.hsm_client_certificates[0].hsm_client_certificate_public_key #=> String resp.hsm_client_certificates[0].tags #=> Array resp.hsm_client_certificates[0].tags[0].key #=> String resp.hsm_client_certificates[0].tags[0].value #=> String
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/DescribeHsmClientCertificates AWS API Documentation
@overload describe_hsm_client_certificates
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 5281 def describe_hsm_client_certificates(params = {}, options = {}) req = build_request(:describe_hsm_client_certificates, params) req.send_request(options) end
Returns information about the specified Amazon Redshift
HSM configuration. If no configuration ID is specified, returns information about all the HSM configurations owned by your Amazon Web Services account.
If you specify both tag keys and tag values in the same request, Amazon Redshift
returns all HSM connections that match any combination of the specified keys and values. For example, if you have `owner` and `environment` for tag keys, and `admin` and `test` for tag values, all HSM connections that have any combination of those values are returned.
If both tag keys and values are omitted from the request, HSM connections are returned regardless of whether they have tag keys or values associated with them.
@option params [String] :hsm_configuration_identifier
The identifier of a specific Amazon Redshift HSM configuration to be described. If no identifier is specified, information is returned for all HSM configurations owned by your Amazon Web Services account.
@option params [Integer] :max_records
The maximum number of response records to return in each call. If the number of remaining response records exceeds the specified `MaxRecords` value, a value is returned in a `marker` field of the response. You can retrieve the next set of records by retrying the command with the returned marker value. Default: `100` Constraints: minimum 20, maximum 100.
@option params [String] :marker
An optional parameter that specifies the starting point to return a set of response records. When the results of a DescribeHsmConfigurations request exceed the value specified in `MaxRecords`, Amazon Web Services returns a value in the `Marker` field of the response. You can retrieve the next set of response records by providing the returned marker value in the `Marker` parameter and retrying the request.
@option params [Array<String>] :tag_keys
A tag key or keys for which you want to return all matching HSM configurations that are associated with the specified key or keys. For example, suppose that you have HSM configurations that are tagged with keys called `owner` and `environment`. If you specify both of these tag keys in the request, Amazon Redshift returns a response with the HSM configurations that have either or both of these tag keys associated with them.
@option params [Array<String>] :tag_values
A tag value or values for which you want to return all matching HSM configurations that are associated with the specified tag value or values. For example, suppose that you have HSM configurations that are tagged with values called `admin` and `test`. If you specify both of these tag values in the request, Amazon Redshift returns a response with the HSM configurations that have either or both of these tag values associated with them.
@return [Types::HsmConfigurationMessage] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::HsmConfigurationMessage#marker #marker} => String * {Types::HsmConfigurationMessage#hsm_configurations #hsm_configurations} => Array<Types::HsmConfiguration>
The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
@example Request syntax with placeholder values
resp = client.describe_hsm_configurations({ hsm_configuration_identifier: "String", max_records: 1, marker: "String", tag_keys: ["String"], tag_values: ["String"], })
@example Response structure
resp.marker #=> String resp.hsm_configurations #=> Array resp.hsm_configurations[0].hsm_configuration_identifier #=> String resp.hsm_configurations[0].description #=> String resp.hsm_configurations[0].hsm_ip_address #=> String resp.hsm_configurations[0].hsm_partition_name #=> String resp.hsm_configurations[0].tags #=> Array resp.hsm_configurations[0].tags[0].key #=> String resp.hsm_configurations[0].tags[0].value #=> String
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/DescribeHsmConfigurations AWS API Documentation
@overload describe_hsm_configurations
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 5378 def describe_hsm_configurations(params = {}, options = {}) req = build_request(:describe_hsm_configurations, params) req.send_request(options) end
Describes whether information, such as queries and connection attempts, is being logged for the specified Amazon Redshift
cluster.
@option params [required, String] :cluster_identifier
The identifier of the cluster from which to get the logging status. Example: `examplecluster`
@return [Types::LoggingStatus] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::LoggingStatus#logging_enabled #logging_enabled} => Boolean * {Types::LoggingStatus#bucket_name #bucket_name} => String * {Types::LoggingStatus#s3_key_prefix #s3_key_prefix} => String * {Types::LoggingStatus#last_successful_delivery_time #last_successful_delivery_time} => Time * {Types::LoggingStatus#last_failure_time #last_failure_time} => Time * {Types::LoggingStatus#last_failure_message #last_failure_message} => String
@example Request syntax with placeholder values
resp = client.describe_logging_status({ cluster_identifier: "String", # required })
@example Response structure
resp.logging_enabled #=> Boolean resp.bucket_name #=> String resp.s3_key_prefix #=> String resp.last_successful_delivery_time #=> Time resp.last_failure_time #=> Time resp.last_failure_message #=> String
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/DescribeLoggingStatus AWS API Documentation
@overload describe_logging_status
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 5419 def describe_logging_status(params = {}, options = {}) req = build_request(:describe_logging_status, params) req.send_request(options) end
Returns properties of possible node configurations such as node type, number of nodes, and disk usage for the specified action type.
@option params [required, String] :action_type
The action type to evaluate for possible node configurations. Specify "restore-cluster" to get configuration combinations based on an existing snapshot. Specify "recommend-node-config" to get configuration recommendations based on an existing cluster or snapshot. Specify "resize-cluster" to get configuration combinations for elastic resize based on an existing cluster.
@option params [String] :cluster_identifier
The identifier of the cluster to evaluate for possible node configurations.
@option params [String] :snapshot_identifier
The identifier of the snapshot to evaluate for possible node configurations.
@option params [String] :owner_account
The Amazon Web Services account used to create or copy the snapshot. Required if you are restoring a snapshot you do not own, optional if you own the snapshot.
@option params [Array<Types::NodeConfigurationOptionsFilter>] :filters
A set of name, operator, and value items to filter the results.
@option params [String] :marker
An optional parameter that specifies the starting point to return a set of response records. When the results of a DescribeNodeConfigurationOptions request exceed the value specified in `MaxRecords`, Amazon Web Services returns a value in the `Marker` field of the response. You can retrieve the next set of response records by providing the returned marker value in the `Marker` parameter and retrying the request.
@option params [Integer] :max_records
The maximum number of response records to return in each call. If the number of remaining response records exceeds the specified `MaxRecords` value, a value is returned in a `marker` field of the response. You can retrieve the next set of records by retrying the command with the returned marker value. Default: `500` Constraints: minimum 100, maximum 500.
@return [Types::NodeConfigurationOptionsMessage] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::NodeConfigurationOptionsMessage#node_configuration_option_list #node_configuration_option_list} => Array<Types::NodeConfigurationOption> * {Types::NodeConfigurationOptionsMessage#marker #marker} => String
The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
@example Request syntax with placeholder values
resp = client.describe_node_configuration_options({ action_type: "restore-cluster", # required, accepts restore-cluster, recommend-node-config, resize-cluster cluster_identifier: "String", snapshot_identifier: "String", owner_account: "String", filters: [ { name: "NodeType", # accepts NodeType, NumberOfNodes, EstimatedDiskUtilizationPercent, Mode operator: "eq", # accepts eq, lt, gt, le, ge, in, between values: ["String"], }, ], marker: "String", max_records: 1, })
@example Response structure
resp.node_configuration_option_list #=> Array resp.node_configuration_option_list[0].node_type #=> String resp.node_configuration_option_list[0].number_of_nodes #=> Integer resp.node_configuration_option_list[0].estimated_disk_utilization_percent #=> Float resp.node_configuration_option_list[0].mode #=> String, one of "standard", "high-performance" resp.marker #=> String
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/DescribeNodeConfigurationOptions AWS API Documentation
@overload describe_node_configuration_options
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 5509 def describe_node_configuration_options(params = {}, options = {}) req = build_request(:describe_node_configuration_options, params) req.send_request(options) end
Returns a list of orderable cluster options. Before you create a new cluster you can use this operation to find what options are available, such as the EC2 Availability Zones (AZ) in the specific Amazon Web Services Region that you can specify, and the node types you can request. The node types differ by available storage, memory, CPU and price. With the cost involved you might want to obtain a list of cluster options in the specific region and specify values when creating a cluster. For more information about managing clusters, go to [Amazon Redshift
Clusters] in the *Amazon Redshift
Cluster Management Guide*.
[1]: docs.aws.amazon.com/redshift/latest/mgmt/working-with-clusters.html
@option params [String] :cluster_version
The version filter value. Specify this parameter to show only the available offerings matching the specified version. Default: All versions. Constraints: Must be one of the version returned from DescribeClusterVersions.
@option params [String] :node_type
The node type filter value. Specify this parameter to show only the available offerings matching the specified node type.
@option params [Integer] :max_records
The maximum number of response records to return in each call. If the number of remaining response records exceeds the specified `MaxRecords` value, a value is returned in a `marker` field of the response. You can retrieve the next set of records by retrying the command with the returned marker value. Default: `100` Constraints: minimum 20, maximum 100.
@option params [String] :marker
An optional parameter that specifies the starting point to return a set of response records. When the results of a DescribeOrderableClusterOptions request exceed the value specified in `MaxRecords`, Amazon Web Services returns a value in the `Marker` field of the response. You can retrieve the next set of response records by providing the returned marker value in the `Marker` parameter and retrying the request.
@return [Types::OrderableClusterOptionsMessage] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::OrderableClusterOptionsMessage#orderable_cluster_options #orderable_cluster_options} => Array<Types::OrderableClusterOption> * {Types::OrderableClusterOptionsMessage#marker #marker} => String
The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
@example Request syntax with placeholder values
resp = client.describe_orderable_cluster_options({ cluster_version: "String", node_type: "String", max_records: 1, marker: "String", })
@example Response structure
resp.orderable_cluster_options #=> Array resp.orderable_cluster_options[0].cluster_version #=> String resp.orderable_cluster_options[0].cluster_type #=> String resp.orderable_cluster_options[0].node_type #=> String resp.orderable_cluster_options[0].availability_zones #=> Array resp.orderable_cluster_options[0].availability_zones[0].name #=> String resp.orderable_cluster_options[0].availability_zones[0].supported_platforms #=> Array resp.orderable_cluster_options[0].availability_zones[0].supported_platforms[0].name #=> String resp.marker #=> String
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/DescribeOrderableClusterOptions AWS API Documentation
@overload describe_orderable_cluster_options
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 5594 def describe_orderable_cluster_options(params = {}, options = {}) req = build_request(:describe_orderable_cluster_options, params) req.send_request(options) end
Returns information about the partner integrations defined for a cluster.
@option params [required, String] :account_id
The Amazon Web Services account ID that owns the cluster.
@option params [required, String] :cluster_identifier
The cluster identifier of the cluster whose partner integration is being described.
@option params [String] :database_name
The name of the database whose partner integration is being described. If database name is not specified, then all databases in the cluster are described.
@option params [String] :partner_name
The name of the partner that is being described. If partner name is not specified, then all partner integrations are described.
@return [Types::DescribePartnersOutputMessage] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::DescribePartnersOutputMessage#partner_integration_info_list #partner_integration_info_list} => Array<Types::PartnerIntegrationInfo>
@example Request syntax with placeholder values
resp = client.describe_partners({ account_id: "PartnerIntegrationAccountId", # required cluster_identifier: "PartnerIntegrationClusterIdentifier", # required database_name: "PartnerIntegrationDatabaseName", partner_name: "PartnerIntegrationPartnerName", })
@example Response structure
resp.partner_integration_info_list #=> Array resp.partner_integration_info_list[0].database_name #=> String resp.partner_integration_info_list[0].partner_name #=> String resp.partner_integration_info_list[0].status #=> String, one of "Active", "Inactive", "RuntimeFailure", "ConnectionFailure" resp.partner_integration_info_list[0].status_message #=> String resp.partner_integration_info_list[0].created_at #=> Time resp.partner_integration_info_list[0].updated_at #=> Time
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/DescribePartners AWS API Documentation
@overload describe_partners
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 5645 def describe_partners(params = {}, options = {}) req = build_request(:describe_partners, params) req.send_request(options) end
Returns a list of the available reserved node offerings by Amazon Redshift
with their descriptions including the node type, the fixed and recurring costs of reserving the node and duration the node will be reserved for you. These descriptions help you determine which reserve node offering you want to purchase. You then use the unique offering ID in you call to PurchaseReservedNodeOffering to reserve one or more nodes for your Amazon Redshift
cluster.
For more information about reserved node offerings, go to [Purchasing Reserved Nodes] in the *Amazon Redshift
Cluster Management Guide*.
[1]: docs.aws.amazon.com/redshift/latest/mgmt/purchase-reserved-node-instance.html
@option params [String] :reserved_node_offering_id
The unique identifier for the offering.
@option params [Integer] :max_records
The maximum number of response records to return in each call. If the number of remaining response records exceeds the specified `MaxRecords` value, a value is returned in a `marker` field of the response. You can retrieve the next set of records by retrying the command with the returned marker value. Default: `100` Constraints: minimum 20, maximum 100.
@option params [String] :marker
An optional parameter that specifies the starting point to return a set of response records. When the results of a DescribeReservedNodeOfferings request exceed the value specified in `MaxRecords`, Amazon Web Services returns a value in the `Marker` field of the response. You can retrieve the next set of response records by providing the returned marker value in the `Marker` parameter and retrying the request.
@return [Types::ReservedNodeOfferingsMessage] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::ReservedNodeOfferingsMessage#marker #marker} => String * {Types::ReservedNodeOfferingsMessage#reserved_node_offerings #reserved_node_offerings} => Array<Types::ReservedNodeOffering>
The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
@example Request syntax with placeholder values
resp = client.describe_reserved_node_offerings({ reserved_node_offering_id: "String", max_records: 1, marker: "String", })
@example Response structure
resp.marker #=> String resp.reserved_node_offerings #=> Array resp.reserved_node_offerings[0].reserved_node_offering_id #=> String resp.reserved_node_offerings[0].node_type #=> String resp.reserved_node_offerings[0].duration #=> Integer resp.reserved_node_offerings[0].fixed_price #=> Float resp.reserved_node_offerings[0].usage_price #=> Float resp.reserved_node_offerings[0].currency_code #=> String resp.reserved_node_offerings[0].offering_type #=> String resp.reserved_node_offerings[0].recurring_charges #=> Array resp.reserved_node_offerings[0].recurring_charges[0].recurring_charge_amount #=> Float resp.reserved_node_offerings[0].recurring_charges[0].recurring_charge_frequency #=> String resp.reserved_node_offerings[0].reserved_node_offering_type #=> String, one of "Regular", "Upgradable"
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/DescribeReservedNodeOfferings AWS API Documentation
@overload describe_reserved_node_offerings
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 5723 def describe_reserved_node_offerings(params = {}, options = {}) req = build_request(:describe_reserved_node_offerings, params) req.send_request(options) end
Returns the descriptions of the reserved nodes.
@option params [String] :reserved_node_id
Identifier for the node reservation.
@option params [Integer] :max_records
The maximum number of response records to return in each call. If the number of remaining response records exceeds the specified `MaxRecords` value, a value is returned in a `marker` field of the response. You can retrieve the next set of records by retrying the command with the returned marker value. Default: `100` Constraints: minimum 20, maximum 100.
@option params [String] :marker
An optional parameter that specifies the starting point to return a set of response records. When the results of a DescribeReservedNodes request exceed the value specified in `MaxRecords`, Amazon Web Services returns a value in the `Marker` field of the response. You can retrieve the next set of response records by providing the returned marker value in the `Marker` parameter and retrying the request.
@return [Types::ReservedNodesMessage] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::ReservedNodesMessage#marker #marker} => String * {Types::ReservedNodesMessage#reserved_nodes #reserved_nodes} => Array<Types::ReservedNode>
The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
@example Request syntax with placeholder values
resp = client.describe_reserved_nodes({ reserved_node_id: "String", max_records: 1, marker: "String", })
@example Response structure
resp.marker #=> String resp.reserved_nodes #=> Array resp.reserved_nodes[0].reserved_node_id #=> String resp.reserved_nodes[0].reserved_node_offering_id #=> String resp.reserved_nodes[0].node_type #=> String resp.reserved_nodes[0].start_time #=> Time resp.reserved_nodes[0].duration #=> Integer resp.reserved_nodes[0].fixed_price #=> Float resp.reserved_nodes[0].usage_price #=> Float resp.reserved_nodes[0].currency_code #=> String resp.reserved_nodes[0].node_count #=> Integer resp.reserved_nodes[0].state #=> String resp.reserved_nodes[0].offering_type #=> String resp.reserved_nodes[0].recurring_charges #=> Array resp.reserved_nodes[0].recurring_charges[0].recurring_charge_amount #=> Float resp.reserved_nodes[0].recurring_charges[0].recurring_charge_frequency #=> String resp.reserved_nodes[0].reserved_node_offering_type #=> String, one of "Regular", "Upgradable"
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/DescribeReservedNodes AWS API Documentation
@overload describe_reserved_nodes
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 5792 def describe_reserved_nodes(params = {}, options = {}) req = build_request(:describe_reserved_nodes, params) req.send_request(options) end
Returns information about the last resize operation for the specified cluster. If no resize operation has ever been initiated for the specified cluster, a `HTTP 404` error is returned. If a resize operation was initiated and completed, the status of the resize remains as `SUCCEEDED` until the next resize.
A resize operation can be requested using ModifyCluster and specifying a different number or type of nodes for the cluster.
@option params [required, String] :cluster_identifier
The unique identifier of a cluster whose resize progress you are requesting. This parameter is case-sensitive. By default, resize operations for all clusters defined for an Amazon Web Services account are returned.
@return [Types::ResizeProgressMessage] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::ResizeProgressMessage#target_node_type #target_node_type} => String * {Types::ResizeProgressMessage#target_number_of_nodes #target_number_of_nodes} => Integer * {Types::ResizeProgressMessage#target_cluster_type #target_cluster_type} => String * {Types::ResizeProgressMessage#status #status} => String * {Types::ResizeProgressMessage#import_tables_completed #import_tables_completed} => Array<String> * {Types::ResizeProgressMessage#import_tables_in_progress #import_tables_in_progress} => Array<String> * {Types::ResizeProgressMessage#import_tables_not_started #import_tables_not_started} => Array<String> * {Types::ResizeProgressMessage#avg_resize_rate_in_mega_bytes_per_second #avg_resize_rate_in_mega_bytes_per_second} => Float * {Types::ResizeProgressMessage#total_resize_data_in_mega_bytes #total_resize_data_in_mega_bytes} => Integer * {Types::ResizeProgressMessage#progress_in_mega_bytes #progress_in_mega_bytes} => Integer * {Types::ResizeProgressMessage#elapsed_time_in_seconds #elapsed_time_in_seconds} => Integer * {Types::ResizeProgressMessage#estimated_time_to_completion_in_seconds #estimated_time_to_completion_in_seconds} => Integer * {Types::ResizeProgressMessage#resize_type #resize_type} => String * {Types::ResizeProgressMessage#message #message} => String * {Types::ResizeProgressMessage#target_encryption_type #target_encryption_type} => String * {Types::ResizeProgressMessage#data_transfer_progress_percent #data_transfer_progress_percent} => Float
@example Request syntax with placeholder values
resp = client.describe_resize({ cluster_identifier: "String", # required })
@example Response structure
resp.target_node_type #=> String resp.target_number_of_nodes #=> Integer resp.target_cluster_type #=> String resp.status #=> String resp.import_tables_completed #=> Array resp.import_tables_completed[0] #=> String resp.import_tables_in_progress #=> Array resp.import_tables_in_progress[0] #=> String resp.import_tables_not_started #=> Array resp.import_tables_not_started[0] #=> String resp.avg_resize_rate_in_mega_bytes_per_second #=> Float resp.total_resize_data_in_mega_bytes #=> Integer resp.progress_in_mega_bytes #=> Integer resp.elapsed_time_in_seconds #=> Integer resp.estimated_time_to_completion_in_seconds #=> Integer resp.resize_type #=> String resp.message #=> String resp.target_encryption_type #=> String resp.data_transfer_progress_percent #=> Float
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/DescribeResize AWS API Documentation
@overload describe_resize
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 5864 def describe_resize(params = {}, options = {}) req = build_request(:describe_resize, params) req.send_request(options) end
Describes properties of scheduled actions.
@option params [String] :scheduled_action_name
The name of the scheduled action to retrieve.
@option params [String] :target_action_type
The type of the scheduled actions to retrieve.
@option params [Time,DateTime,Date,Integer,String] :start_time
The start time in UTC of the scheduled actions to retrieve. Only active scheduled actions that have invocations after this time are retrieved.
@option params [Time,DateTime,Date,Integer,String] :end_time
The end time in UTC of the scheduled action to retrieve. Only active scheduled actions that have invocations before this time are retrieved.
@option params [Boolean] :active
If true, retrieve only active scheduled actions. If false, retrieve only disabled scheduled actions.
@option params [Array<Types::ScheduledActionFilter>] :filters
List of scheduled action filters.
@option params [String] :marker
An optional parameter that specifies the starting point to return a set of response records. When the results of a DescribeScheduledActions request exceed the value specified in `MaxRecords`, Amazon Web Services returns a value in the `Marker` field of the response. You can retrieve the next set of response records by providing the returned marker value in the `Marker` parameter and retrying the request.
@option params [Integer] :max_records
The maximum number of response records to return in each call. If the number of remaining response records exceeds the specified `MaxRecords` value, a value is returned in a `marker` field of the response. You can retrieve the next set of records by retrying the command with the returned marker value. Default: `100` Constraints: minimum 20, maximum 100.
@return [Types::ScheduledActionsMessage] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::ScheduledActionsMessage#marker #marker} => String * {Types::ScheduledActionsMessage#scheduled_actions #scheduled_actions} => Array<Types::ScheduledAction>
The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
@example Request syntax with placeholder values
resp = client.describe_scheduled_actions({ scheduled_action_name: "String", target_action_type: "ResizeCluster", # accepts ResizeCluster, PauseCluster, ResumeCluster start_time: Time.now, end_time: Time.now, active: false, filters: [ { name: "cluster-identifier", # required, accepts cluster-identifier, iam-role values: ["String"], # required }, ], marker: "String", max_records: 1, })
@example Response structure
resp.marker #=> String resp.scheduled_actions #=> Array resp.scheduled_actions[0].scheduled_action_name #=> String resp.scheduled_actions[0].target_action.resize_cluster.cluster_identifier #=> String resp.scheduled_actions[0].target_action.resize_cluster.cluster_type #=> String resp.scheduled_actions[0].target_action.resize_cluster.node_type #=> String resp.scheduled_actions[0].target_action.resize_cluster.number_of_nodes #=> Integer resp.scheduled_actions[0].target_action.resize_cluster.classic #=> Boolean resp.scheduled_actions[0].target_action.pause_cluster.cluster_identifier #=> String resp.scheduled_actions[0].target_action.resume_cluster.cluster_identifier #=> String resp.scheduled_actions[0].schedule #=> String resp.scheduled_actions[0].iam_role #=> String resp.scheduled_actions[0].scheduled_action_description #=> String resp.scheduled_actions[0].state #=> String, one of "ACTIVE", "DISABLED" resp.scheduled_actions[0].next_invocations #=> Array resp.scheduled_actions[0].next_invocations[0] #=> Time resp.scheduled_actions[0].start_time #=> Time resp.scheduled_actions[0].end_time #=> Time
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/DescribeScheduledActions AWS API Documentation
@overload describe_scheduled_actions
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 5964 def describe_scheduled_actions(params = {}, options = {}) req = build_request(:describe_scheduled_actions, params) req.send_request(options) end
Returns a list of snapshot copy grants owned by the Amazon Web Services account in the destination region.
For more information about managing snapshot copy grants, go to
Cluster Management Guide*.
[1]: docs.aws.amazon.com/redshift/latest/mgmt/working-with-db-encryption.html
@option params [String] :snapshot_copy_grant_name
The name of the snapshot copy grant.
@option params [Integer] :max_records
The maximum number of response records to return in each call. If the number of remaining response records exceeds the specified `MaxRecords` value, a value is returned in a `marker` field of the response. You can retrieve the next set of records by retrying the command with the returned marker value. Default: `100` Constraints: minimum 20, maximum 100.
@option params [String] :marker
An optional parameter that specifies the starting point to return a set of response records. When the results of a `DescribeSnapshotCopyGrant` request exceed the value specified in `MaxRecords`, Amazon Web Services returns a value in the `Marker` field of the response. You can retrieve the next set of response records by providing the returned marker value in the `Marker` parameter and retrying the request. Constraints: You can specify either the **SnapshotCopyGrantName** parameter or the **Marker** parameter, but not both.
@option params [Array<String>] :tag_keys
A tag key or keys for which you want to return all matching resources that are associated with the specified key or keys. For example, suppose that you have resources tagged with keys called `owner` and `environment`. If you specify both of these tag keys in the request, Amazon Redshift returns a response with all resources that have either or both of these tag keys associated with them.
@option params [Array<String>] :tag_values
A tag value or values for which you want to return all matching resources that are associated with the specified value or values. For example, suppose that you have resources tagged with values called `admin` and `test`. If you specify both of these tag values in the request, Amazon Redshift returns a response with all resources that have either or both of these tag values associated with them.
@return [Types::SnapshotCopyGrantMessage] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::SnapshotCopyGrantMessage#marker #marker} => String * {Types::SnapshotCopyGrantMessage#snapshot_copy_grants #snapshot_copy_grants} => Array<Types::SnapshotCopyGrant>
The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
@example Request syntax with placeholder values
resp = client.describe_snapshot_copy_grants({ snapshot_copy_grant_name: "String", max_records: 1, marker: "String", tag_keys: ["String"], tag_values: ["String"], })
@example Response structure
resp.marker #=> String resp.snapshot_copy_grants #=> Array resp.snapshot_copy_grants[0].snapshot_copy_grant_name #=> String resp.snapshot_copy_grants[0].kms_key_id #=> String resp.snapshot_copy_grants[0].tags #=> Array resp.snapshot_copy_grants[0].tags[0].key #=> String resp.snapshot_copy_grants[0].tags[0].value #=> String
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/DescribeSnapshotCopyGrants AWS API Documentation
@overload describe_snapshot_copy_grants
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 6053 def describe_snapshot_copy_grants(params = {}, options = {}) req = build_request(:describe_snapshot_copy_grants, params) req.send_request(options) end
Returns a list of snapshot schedules.
@option params [String] :cluster_identifier
The unique identifier for the cluster whose snapshot schedules you want to view.
@option params [String] :schedule_identifier
A unique identifier for a snapshot schedule.
@option params [Array<String>] :tag_keys
The key value for a snapshot schedule tag.
@option params [Array<String>] :tag_values
The value corresponding to the key of the snapshot schedule tag.
@option params [String] :marker
A value that indicates the starting point for the next set of response records in a subsequent request. If a value is returned in a response, you can retrieve the next set of records by providing this returned marker value in the `marker` parameter and retrying the command. If the `marker` field is empty, all response records have been retrieved for the request.
@option params [Integer] :max_records
The maximum number or response records to return in each call. If the number of remaining response records exceeds the specified `MaxRecords` value, a value is returned in a `marker` field of the response. You can retrieve the next set of records by retrying the command with the returned `marker` value.
@return [Types::DescribeSnapshotSchedulesOutputMessage] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::DescribeSnapshotSchedulesOutputMessage#snapshot_schedules #snapshot_schedules} => Array<Types::SnapshotSchedule> * {Types::DescribeSnapshotSchedulesOutputMessage#marker #marker} => String
The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
@example Request syntax with placeholder values
resp = client.describe_snapshot_schedules({ cluster_identifier: "String", schedule_identifier: "String", tag_keys: ["String"], tag_values: ["String"], marker: "String", max_records: 1, })
@example Response structure
resp.snapshot_schedules #=> Array resp.snapshot_schedules[0].schedule_definitions #=> Array resp.snapshot_schedules[0].schedule_definitions[0] #=> String resp.snapshot_schedules[0].schedule_identifier #=> String resp.snapshot_schedules[0].schedule_description #=> String resp.snapshot_schedules[0].tags #=> Array resp.snapshot_schedules[0].tags[0].key #=> String resp.snapshot_schedules[0].tags[0].value #=> String resp.snapshot_schedules[0].next_invocations #=> Array resp.snapshot_schedules[0].next_invocations[0] #=> Time resp.snapshot_schedules[0].associated_cluster_count #=> Integer resp.snapshot_schedules[0].associated_clusters #=> Array resp.snapshot_schedules[0].associated_clusters[0].cluster_identifier #=> String resp.snapshot_schedules[0].associated_clusters[0].schedule_association_state #=> String, one of "MODIFYING", "ACTIVE", "FAILED" resp.marker #=> String
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/DescribeSnapshotSchedules AWS API Documentation
@overload describe_snapshot_schedules
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 6128 def describe_snapshot_schedules(params = {}, options = {}) req = build_request(:describe_snapshot_schedules, params) req.send_request(options) end
Returns account level backups storage size and provisional storage.
@return [Types::CustomerStorageMessage] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::CustomerStorageMessage#total_backup_size_in_mega_bytes #total_backup_size_in_mega_bytes} => Float * {Types::CustomerStorageMessage#total_provisioned_storage_in_mega_bytes #total_provisioned_storage_in_mega_bytes} => Float
@example Response structure
resp.total_backup_size_in_mega_bytes #=> Float resp.total_provisioned_storage_in_mega_bytes #=> Float
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/DescribeStorage AWS API Documentation
@overload describe_storage
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 6149 def describe_storage(params = {}, options = {}) req = build_request(:describe_storage, params) req.send_request(options) end
Lists the status of one or more table restore requests made using the RestoreTableFromClusterSnapshot API action. If you don't specify a value for the `TableRestoreRequestId` parameter, then `DescribeTableRestoreStatus` returns the status of all table restore requests ordered by the date and time of the request in ascending order. Otherwise `DescribeTableRestoreStatus` returns the status of the table specified by `TableRestoreRequestId`.
@option params [String] :cluster_identifier
The Amazon Redshift cluster that the table is being restored to.
@option params [String] :table_restore_request_id
The identifier of the table restore request to return status for. If you don't specify a `TableRestoreRequestId` value, then `DescribeTableRestoreStatus` returns the status of all in-progress table restore requests.
@option params [Integer] :max_records
The maximum number of records to include in the response. If more records exist than the specified `MaxRecords` value, a pagination token called a marker is included in the response so that the remaining results can be retrieved.
@option params [String] :marker
An optional pagination token provided by a previous `DescribeTableRestoreStatus` request. If this parameter is specified, the response includes only records beyond the marker, up to the value specified by the `MaxRecords` parameter.
@return [Types::TableRestoreStatusMessage] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::TableRestoreStatusMessage#table_restore_status_details #table_restore_status_details} => Array<Types::TableRestoreStatus> * {Types::TableRestoreStatusMessage#marker #marker} => String
The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
@example Request syntax with placeholder values
resp = client.describe_table_restore_status({ cluster_identifier: "String", table_restore_request_id: "String", max_records: 1, marker: "String", })
@example Response structure
resp.table_restore_status_details #=> Array resp.table_restore_status_details[0].table_restore_request_id #=> String resp.table_restore_status_details[0].status #=> String, one of "PENDING", "IN_PROGRESS", "SUCCEEDED", "FAILED", "CANCELED" resp.table_restore_status_details[0].message #=> String resp.table_restore_status_details[0].request_time #=> Time resp.table_restore_status_details[0].progress_in_mega_bytes #=> Integer resp.table_restore_status_details[0].total_data_in_mega_bytes #=> Integer resp.table_restore_status_details[0].cluster_identifier #=> String resp.table_restore_status_details[0].snapshot_identifier #=> String resp.table_restore_status_details[0].source_database_name #=> String resp.table_restore_status_details[0].source_schema_name #=> String resp.table_restore_status_details[0].source_table_name #=> String resp.table_restore_status_details[0].target_database_name #=> String resp.table_restore_status_details[0].target_schema_name #=> String resp.table_restore_status_details[0].new_table_name #=> String resp.marker #=> String
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/DescribeTableRestoreStatus AWS API Documentation
@overload describe_table_restore_status
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 6222 def describe_table_restore_status(params = {}, options = {}) req = build_request(:describe_table_restore_status, params) req.send_request(options) end
Shows usage limits on a cluster. Results are filtered based on the combination of input usage limit identifier, cluster identifier, and feature type parameters:
-
If usage limit identifier, cluster identifier, and feature type are not provided, then all usage limit objects for the current account in the current region are returned.
-
If usage limit identifier is provided, then the corresponding usage limit object is returned.
-
If cluster identifier is provided, then all usage limit objects for the specified cluster are returned.
-
If cluster identifier and feature type are provided, then all usage limit objects for the combination of cluster and feature are returned.
@option params [String] :usage_limit_id
The identifier of the usage limit to describe.
@option params [String] :cluster_identifier
The identifier of the cluster for which you want to describe usage limits.
@option params [String] :feature_type
The feature type for which you want to describe usage limits.
@option params [Integer] :max_records
The maximum number of response records to return in each call. If the number of remaining response records exceeds the specified `MaxRecords` value, a value is returned in a `marker` field of the response. You can retrieve the next set of records by retrying the command with the returned marker value. Default: `100` Constraints: minimum 20, maximum 100.
@option params [String] :marker
An optional parameter that specifies the starting point to return a set of response records. When the results of a DescribeUsageLimits request exceed the value specified in `MaxRecords`, Amazon Web Services returns a value in the `Marker` field of the response. You can retrieve the next set of response records by providing the returned marker value in the `Marker` parameter and retrying the request.
@option params [Array<String>] :tag_keys
A tag key or keys for which you want to return all matching usage limit objects that are associated with the specified key or keys. For example, suppose that you have parameter groups that are tagged with keys called `owner` and `environment`. If you specify both of these tag keys in the request, Amazon Redshift returns a response with the usage limit objects have either or both of these tag keys associated with them.
@option params [Array<String>] :tag_values
A tag value or values for which you want to return all matching usage limit objects that are associated with the specified tag value or values. For example, suppose that you have parameter groups that are tagged with values called `admin` and `test`. If you specify both of these tag values in the request, Amazon Redshift returns a response with the usage limit objects that have either or both of these tag values associated with them.
@return [Types::UsageLimitList] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::UsageLimitList#usage_limits #usage_limits} => Array<Types::UsageLimit> * {Types::UsageLimitList#marker #marker} => String
The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
@example Request syntax with placeholder values
resp = client.describe_usage_limits({ usage_limit_id: "String", cluster_identifier: "String", feature_type: "spectrum", # accepts spectrum, concurrency-scaling max_records: 1, marker: "String", tag_keys: ["String"], tag_values: ["String"], })
@example Response structure
resp.usage_limits #=> Array resp.usage_limits[0].usage_limit_id #=> String resp.usage_limits[0].cluster_identifier #=> String resp.usage_limits[0].feature_type #=> String, one of "spectrum", "concurrency-scaling" resp.usage_limits[0].limit_type #=> String, one of "time", "data-scanned" resp.usage_limits[0].amount #=> Integer resp.usage_limits[0].period #=> String, one of "daily", "weekly", "monthly" resp.usage_limits[0].breach_action #=> String, one of "log", "emit-metric", "disable" resp.usage_limits[0].tags #=> Array resp.usage_limits[0].tags[0].key #=> String resp.usage_limits[0].tags[0].value #=> String resp.marker #=> String
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/DescribeUsageLimits AWS API Documentation
@overload describe_usage_limits
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 6461 def describe_usage_limits(params = {}, options = {}) req = build_request(:describe_usage_limits, params) req.send_request(options) end
Stops logging information, such as queries and connection attempts, for the specified Amazon Redshift
cluster.
@option params [required, String] :cluster_identifier
The identifier of the cluster on which logging is to be stopped. Example: `examplecluster`
@return [Types::LoggingStatus] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::LoggingStatus#logging_enabled #logging_enabled} => Boolean * {Types::LoggingStatus#bucket_name #bucket_name} => String * {Types::LoggingStatus#s3_key_prefix #s3_key_prefix} => String * {Types::LoggingStatus#last_successful_delivery_time #last_successful_delivery_time} => Time * {Types::LoggingStatus#last_failure_time #last_failure_time} => Time * {Types::LoggingStatus#last_failure_message #last_failure_message} => String
@example Request syntax with placeholder values
resp = client.disable_logging({ cluster_identifier: "String", # required })
@example Response structure
resp.logging_enabled #=> Boolean resp.bucket_name #=> String resp.s3_key_prefix #=> String resp.last_successful_delivery_time #=> Time resp.last_failure_time #=> Time resp.last_failure_message #=> String
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/DisableLogging AWS API Documentation
@overload disable_logging
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 6502 def disable_logging(params = {}, options = {}) req = build_request(:disable_logging, params) req.send_request(options) end
Disables the automatic copying of snapshots from one region to another region for a specified cluster.
If your cluster and its snapshots are encrypted using a customer master key (CMK) from Key Management Service, use DeleteSnapshotCopyGrant to delete the grant that grants Amazon Redshift
permission to the CMK in the destination region.
@option params [required, String] :cluster_identifier
The unique identifier of the source cluster that you want to disable copying of snapshots to a destination region. Constraints: Must be the valid name of an existing cluster that has cross-region snapshot copy enabled.
@return [Types::DisableSnapshotCopyResult] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::DisableSnapshotCopyResult#cluster #cluster} => Types::Cluster
@example Request syntax with placeholder values
resp = client.disable_snapshot_copy({ cluster_identifier: "String", # required })
@example Response structure
resp.cluster.cluster_identifier #=> String resp.cluster.node_type #=> String resp.cluster.cluster_status #=> String resp.cluster.cluster_availability_status #=> String resp.cluster.modify_status #=> String resp.cluster.master_username #=> String resp.cluster.db_name #=> String resp.cluster.endpoint.address #=> String resp.cluster.endpoint.port #=> Integer resp.cluster.endpoint.vpc_endpoints #=> Array resp.cluster.endpoint.vpc_endpoints[0].vpc_endpoint_id #=> String resp.cluster.endpoint.vpc_endpoints[0].vpc_id #=> String resp.cluster.endpoint.vpc_endpoints[0].network_interfaces #=> Array resp.cluster.endpoint.vpc_endpoints[0].network_interfaces[0].network_interface_id #=> String resp.cluster.endpoint.vpc_endpoints[0].network_interfaces[0].subnet_id #=> String resp.cluster.endpoint.vpc_endpoints[0].network_interfaces[0].private_ip_address #=> String resp.cluster.endpoint.vpc_endpoints[0].network_interfaces[0].availability_zone #=> String resp.cluster.cluster_create_time #=> Time resp.cluster.automated_snapshot_retention_period #=> Integer resp.cluster.manual_snapshot_retention_period #=> Integer resp.cluster.cluster_security_groups #=> Array resp.cluster.cluster_security_groups[0].cluster_security_group_name #=> String resp.cluster.cluster_security_groups[0].status #=> String resp.cluster.vpc_security_groups #=> Array resp.cluster.vpc_security_groups[0].vpc_security_group_id #=> String resp.cluster.vpc_security_groups[0].status #=> String resp.cluster.cluster_parameter_groups #=> Array resp.cluster.cluster_parameter_groups[0].parameter_group_name #=> String resp.cluster.cluster_parameter_groups[0].parameter_apply_status #=> String resp.cluster.cluster_parameter_groups[0].cluster_parameter_status_list #=> Array resp.cluster.cluster_parameter_groups[0].cluster_parameter_status_list[0].parameter_name #=> String resp.cluster.cluster_parameter_groups[0].cluster_parameter_status_list[0].parameter_apply_status #=> String resp.cluster.cluster_parameter_groups[0].cluster_parameter_status_list[0].parameter_apply_error_description #=> String resp.cluster.cluster_subnet_group_name #=> String resp.cluster.vpc_id #=> String resp.cluster.availability_zone #=> String resp.cluster.preferred_maintenance_window #=> String resp.cluster.pending_modified_values.master_user_password #=> String resp.cluster.pending_modified_values.node_type #=> String resp.cluster.pending_modified_values.number_of_nodes #=> Integer resp.cluster.pending_modified_values.cluster_type #=> String resp.cluster.pending_modified_values.cluster_version #=> String resp.cluster.pending_modified_values.automated_snapshot_retention_period #=> Integer resp.cluster.pending_modified_values.cluster_identifier #=> String resp.cluster.pending_modified_values.publicly_accessible #=> Boolean resp.cluster.pending_modified_values.enhanced_vpc_routing #=> Boolean resp.cluster.pending_modified_values.maintenance_track_name #=> String resp.cluster.pending_modified_values.encryption_type #=> String resp.cluster.cluster_version #=> String resp.cluster.allow_version_upgrade #=> Boolean resp.cluster.number_of_nodes #=> Integer resp.cluster.publicly_accessible #=> Boolean resp.cluster.encrypted #=> Boolean resp.cluster.restore_status.status #=> String resp.cluster.restore_status.current_restore_rate_in_mega_bytes_per_second #=> Float resp.cluster.restore_status.snapshot_size_in_mega_bytes #=> Integer resp.cluster.restore_status.progress_in_mega_bytes #=> Integer resp.cluster.restore_status.elapsed_time_in_seconds #=> Integer resp.cluster.restore_status.estimated_time_to_completion_in_seconds #=> Integer resp.cluster.data_transfer_progress.status #=> String resp.cluster.data_transfer_progress.current_rate_in_mega_bytes_per_second #=> Float resp.cluster.data_transfer_progress.total_data_in_mega_bytes #=> Integer resp.cluster.data_transfer_progress.data_transferred_in_mega_bytes #=> Integer resp.cluster.data_transfer_progress.estimated_time_to_completion_in_seconds #=> Integer resp.cluster.data_transfer_progress.elapsed_time_in_seconds #=> Integer resp.cluster.hsm_status.hsm_client_certificate_identifier #=> String resp.cluster.hsm_status.hsm_configuration_identifier #=> String resp.cluster.hsm_status.status #=> String resp.cluster.cluster_snapshot_copy_status.destination_region #=> String resp.cluster.cluster_snapshot_copy_status.retention_period #=> Integer resp.cluster.cluster_snapshot_copy_status.manual_snapshot_retention_period #=> Integer resp.cluster.cluster_snapshot_copy_status.snapshot_copy_grant_name #=> String resp.cluster.cluster_public_key #=> String resp.cluster.cluster_nodes #=> Array resp.cluster.cluster_nodes[0].node_role #=> String resp.cluster.cluster_nodes[0].private_ip_address #=> String resp.cluster.cluster_nodes[0].public_ip_address #=> String resp.cluster.elastic_ip_status.elastic_ip #=> String resp.cluster.elastic_ip_status.status #=> String resp.cluster.cluster_revision_number #=> String resp.cluster.tags #=> Array resp.cluster.tags[0].key #=> String resp.cluster.tags[0].value #=> String resp.cluster.kms_key_id #=> String resp.cluster.enhanced_vpc_routing #=> Boolean resp.cluster.iam_roles #=> Array resp.cluster.iam_roles[0].iam_role_arn #=> String resp.cluster.iam_roles[0].apply_status #=> String resp.cluster.pending_actions #=> Array resp.cluster.pending_actions[0] #=> String resp.cluster.maintenance_track_name #=> String resp.cluster.elastic_resize_number_of_node_options #=> String resp.cluster.deferred_maintenance_windows #=> Array resp.cluster.deferred_maintenance_windows[0].defer_maintenance_identifier #=> String resp.cluster.deferred_maintenance_windows[0].defer_maintenance_start_time #=> Time resp.cluster.deferred_maintenance_windows[0].defer_maintenance_end_time #=> Time resp.cluster.snapshot_schedule_identifier #=> String resp.cluster.snapshot_schedule_state #=> String, one of "MODIFYING", "ACTIVE", "FAILED" resp.cluster.expected_next_snapshot_schedule_time #=> Time resp.cluster.expected_next_snapshot_schedule_time_status #=> String resp.cluster.next_maintenance_window_start_time #=> Time resp.cluster.resize_info.resize_type #=> String resp.cluster.resize_info.allow_cancel_resize #=> Boolean resp.cluster.availability_zone_relocation_status #=> String resp.cluster.cluster_namespace_arn #=> String resp.cluster.total_storage_capacity_in_mega_bytes #=> Integer resp.cluster.aqua_configuration.aqua_status #=> String, one of "enabled", "disabled", "applying" resp.cluster.aqua_configuration.aqua_configuration_status #=> String, one of "enabled", "disabled", "auto"
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/DisableSnapshotCopy AWS API Documentation
@overload disable_snapshot_copy
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 6647 def disable_snapshot_copy(params = {}, options = {}) req = build_request(:disable_snapshot_copy, params) req.send_request(options) end
Starts logging information, such as queries and connection attempts, for the specified Amazon Redshift
cluster.
@option params [required, String] :cluster_identifier
The identifier of the cluster on which logging is to be started. Example: `examplecluster`
@option params [required, String] :bucket_name
The name of an existing S3 bucket where the log files are to be stored. Constraints: * Must be in the same region as the cluster * The cluster must have read bucket and put object permissions
@option params [String] :s3_key_prefix
The prefix applied to the log file names. Constraints: * Cannot exceed 512 characters * Cannot contain spaces( ), double quotes ("), single quotes ('), a backslash (\\), or control characters. The hexadecimal codes for invalid characters are: * x00 to x20 * x22 * x27 * x5c * x7f or larger
@return [Types::LoggingStatus] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::LoggingStatus#logging_enabled #logging_enabled} => Boolean * {Types::LoggingStatus#bucket_name #bucket_name} => String * {Types::LoggingStatus#s3_key_prefix #s3_key_prefix} => String * {Types::LoggingStatus#last_successful_delivery_time #last_successful_delivery_time} => Time * {Types::LoggingStatus#last_failure_time #last_failure_time} => Time * {Types::LoggingStatus#last_failure_message #last_failure_message} => String
@example Request syntax with placeholder values
resp = client.enable_logging({ cluster_identifier: "String", # required bucket_name: "String", # required s3_key_prefix: "String", })
@example Response structure
resp.logging_enabled #=> Boolean resp.bucket_name #=> String resp.s3_key_prefix #=> String resp.last_successful_delivery_time #=> Time resp.last_failure_time #=> Time resp.last_failure_message #=> String
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/EnableLogging AWS API Documentation
@overload enable_logging
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 6771 def enable_logging(params = {}, options = {}) req = build_request(:enable_logging, params) req.send_request(options) end
Enables the automatic copy of snapshots from one region to another region for a specified cluster.
@option params [required, String] :cluster_identifier
The unique identifier of the source cluster to copy snapshots from. Constraints: Must be the valid name of an existing cluster that does not already have cross-region snapshot copy enabled.
@option params [required, String] :destination_region
The destination Amazon Web Services Region that you want to copy snapshots to. Constraints: Must be the name of a valid Amazon Web Services Region. For more information, see [Regions and Endpoints][1] in the Amazon Web Services General Reference. [1]: https://docs.aws.amazon.com/general/latest/gr/rande.html#redshift_region
@option params [Integer] :retention_period
The number of days to retain automated snapshots in the destination region after they are copied from the source region. Default: 7. Constraints: Must be at least 1 and no more than 35.
@option params [String] :snapshot_copy_grant_name
The name of the snapshot copy grant to use when snapshots of an Amazon Web Services KMS-encrypted cluster are copied to the destination region.
@option params [Integer] :manual_snapshot_retention_period
The number of days to retain newly copied snapshots in the destination Amazon Web Services Region after they are copied from the source Amazon Web Services Region. If the value is -1, the manual snapshot is retained indefinitely. The value must be either -1 or an integer between 1 and 3,653.
@return [Types::EnableSnapshotCopyResult] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::EnableSnapshotCopyResult#cluster #cluster} => Types::Cluster
@example Request syntax with placeholder values
resp = client.enable_snapshot_copy({ cluster_identifier: "String", # required destination_region: "String", # required retention_period: 1, snapshot_copy_grant_name: "String", manual_snapshot_retention_period: 1, })
@example Response structure
resp.cluster.cluster_identifier #=> String resp.cluster.node_type #=> String resp.cluster.cluster_status #=> String resp.cluster.cluster_availability_status #=> String resp.cluster.modify_status #=> String resp.cluster.master_username #=> String resp.cluster.db_name #=> String resp.cluster.endpoint.address #=> String resp.cluster.endpoint.port #=> Integer resp.cluster.endpoint.vpc_endpoints #=> Array resp.cluster.endpoint.vpc_endpoints[0].vpc_endpoint_id #=> String resp.cluster.endpoint.vpc_endpoints[0].vpc_id #=> String resp.cluster.endpoint.vpc_endpoints[0].network_interfaces #=> Array resp.cluster.endpoint.vpc_endpoints[0].network_interfaces[0].network_interface_id #=> String resp.cluster.endpoint.vpc_endpoints[0].network_interfaces[0].subnet_id #=> String resp.cluster.endpoint.vpc_endpoints[0].network_interfaces[0].private_ip_address #=> String resp.cluster.endpoint.vpc_endpoints[0].network_interfaces[0].availability_zone #=> String resp.cluster.cluster_create_time #=> Time resp.cluster.automated_snapshot_retention_period #=> Integer resp.cluster.manual_snapshot_retention_period #=> Integer resp.cluster.cluster_security_groups #=> Array resp.cluster.cluster_security_groups[0].cluster_security_group_name #=> String resp.cluster.cluster_security_groups[0].status #=> String resp.cluster.vpc_security_groups #=> Array resp.cluster.vpc_security_groups[0].vpc_security_group_id #=> String resp.cluster.vpc_security_groups[0].status #=> String resp.cluster.cluster_parameter_groups #=> Array resp.cluster.cluster_parameter_groups[0].parameter_group_name #=> String resp.cluster.cluster_parameter_groups[0].parameter_apply_status #=> String resp.cluster.cluster_parameter_groups[0].cluster_parameter_status_list #=> Array resp.cluster.cluster_parameter_groups[0].cluster_parameter_status_list[0].parameter_name #=> String resp.cluster.cluster_parameter_groups[0].cluster_parameter_status_list[0].parameter_apply_status #=> String resp.cluster.cluster_parameter_groups[0].cluster_parameter_status_list[0].parameter_apply_error_description #=> String resp.cluster.cluster_subnet_group_name #=> String resp.cluster.vpc_id #=> String resp.cluster.availability_zone #=> String resp.cluster.preferred_maintenance_window #=> String resp.cluster.pending_modified_values.master_user_password #=> String resp.cluster.pending_modified_values.node_type #=> String resp.cluster.pending_modified_values.number_of_nodes #=> Integer resp.cluster.pending_modified_values.cluster_type #=> String resp.cluster.pending_modified_values.cluster_version #=> String resp.cluster.pending_modified_values.automated_snapshot_retention_period #=> Integer resp.cluster.pending_modified_values.cluster_identifier #=> String resp.cluster.pending_modified_values.publicly_accessible #=> Boolean resp.cluster.pending_modified_values.enhanced_vpc_routing #=> Boolean resp.cluster.pending_modified_values.maintenance_track_name #=> String resp.cluster.pending_modified_values.encryption_type #=> String resp.cluster.cluster_version #=> String resp.cluster.allow_version_upgrade #=> Boolean resp.cluster.number_of_nodes #=> Integer resp.cluster.publicly_accessible #=> Boolean resp.cluster.encrypted #=> Boolean resp.cluster.restore_status.status #=> String resp.cluster.restore_status.current_restore_rate_in_mega_bytes_per_second #=> Float resp.cluster.restore_status.snapshot_size_in_mega_bytes #=> Integer resp.cluster.restore_status.progress_in_mega_bytes #=> Integer resp.cluster.restore_status.elapsed_time_in_seconds #=> Integer resp.cluster.restore_status.estimated_time_to_completion_in_seconds #=> Integer resp.cluster.data_transfer_progress.status #=> String resp.cluster.data_transfer_progress.current_rate_in_mega_bytes_per_second #=> Float resp.cluster.data_transfer_progress.total_data_in_mega_bytes #=> Integer resp.cluster.data_transfer_progress.data_transferred_in_mega_bytes #=> Integer resp.cluster.data_transfer_progress.estimated_time_to_completion_in_seconds #=> Integer resp.cluster.data_transfer_progress.elapsed_time_in_seconds #=> Integer resp.cluster.hsm_status.hsm_client_certificate_identifier #=> String resp.cluster.hsm_status.hsm_configuration_identifier #=> String resp.cluster.hsm_status.status #=> String resp.cluster.cluster_snapshot_copy_status.destination_region #=> String resp.cluster.cluster_snapshot_copy_status.retention_period #=> Integer resp.cluster.cluster_snapshot_copy_status.manual_snapshot_retention_period #=> Integer resp.cluster.cluster_snapshot_copy_status.snapshot_copy_grant_name #=> String resp.cluster.cluster_public_key #=> String resp.cluster.cluster_nodes #=> Array resp.cluster.cluster_nodes[0].node_role #=> String resp.cluster.cluster_nodes[0].private_ip_address #=> String resp.cluster.cluster_nodes[0].public_ip_address #=> String resp.cluster.elastic_ip_status.elastic_ip #=> String resp.cluster.elastic_ip_status.status #=> String resp.cluster.cluster_revision_number #=> String resp.cluster.tags #=> Array resp.cluster.tags[0].key #=> String resp.cluster.tags[0].value #=> String resp.cluster.kms_key_id #=> String resp.cluster.enhanced_vpc_routing #=> Boolean resp.cluster.iam_roles #=> Array resp.cluster.iam_roles[0].iam_role_arn #=> String resp.cluster.iam_roles[0].apply_status #=> String resp.cluster.pending_actions #=> Array resp.cluster.pending_actions[0] #=> String resp.cluster.maintenance_track_name #=> String resp.cluster.elastic_resize_number_of_node_options #=> String resp.cluster.deferred_maintenance_windows #=> Array resp.cluster.deferred_maintenance_windows[0].defer_maintenance_identifier #=> String resp.cluster.deferred_maintenance_windows[0].defer_maintenance_start_time #=> Time resp.cluster.deferred_maintenance_windows[0].defer_maintenance_end_time #=> Time resp.cluster.snapshot_schedule_identifier #=> String resp.cluster.snapshot_schedule_state #=> String, one of "MODIFYING", "ACTIVE", "FAILED" resp.cluster.expected_next_snapshot_schedule_time #=> Time resp.cluster.expected_next_snapshot_schedule_time_status #=> String resp.cluster.next_maintenance_window_start_time #=> Time resp.cluster.resize_info.resize_type #=> String resp.cluster.resize_info.allow_cancel_resize #=> Boolean resp.cluster.availability_zone_relocation_status #=> String resp.cluster.cluster_namespace_arn #=> String resp.cluster.total_storage_capacity_in_mega_bytes #=> Integer resp.cluster.aqua_configuration.aqua_status #=> String, one of "enabled", "disabled", "applying" resp.cluster.aqua_configuration.aqua_configuration_status #=> String, one of "enabled", "disabled", "auto"
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/EnableSnapshotCopy AWS API Documentation
@overload enable_snapshot_copy
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 6947 def enable_snapshot_copy(params = {}, options = {}) req = build_request(:enable_snapshot_copy, params) req.send_request(options) end
Returns a database user name and temporary password with temporary authorization to log on to an Amazon Redshift
database. The action returns the database user name prefixed with `IAM:` if `AutoCreate` is `False` or `IAMA:` if `AutoCreate` is `True`. You can optionally specify one or more database user groups that the user will join at log on. By default, the temporary credentials expire in 900 seconds. You can optionally specify a duration between 900 seconds (15 minutes) and 3600 seconds (60 minutes). For more information, see [Using IAM Authentication to Generate Database User Credentials] in the Amazon Redshift
Cluster Management Guide.
The Identity and Access Management (IAM) user or role that runs GetClusterCredentials must have an IAM policy attached that allows access to all necessary actions and resources. For more information about permissions, see [Resource Policies for GetClusterCredentials] in the Amazon Redshift
Cluster Management Guide.
If the `DbGroups` parameter is specified, the IAM policy must allow the `redshift:JoinGroup` action with access to the listed `dbgroups`.
In addition, if the `AutoCreate` parameter is set to `True`, then the policy must include the `redshift:CreateClusterUser` privilege.
If the `DbName` parameter is specified, the IAM policy must allow access to the resource `dbname` for the specified database name.
[1]: docs.aws.amazon.com/redshift/latest/mgmt/generating-user-credentials.html [2]: docs.aws.amazon.com/redshift/latest/mgmt/redshift-iam-access-control-identity-based.html#redshift-policy-resources.getclustercredentials-resources
@option params [required, String] :db_user
The name of a database user. If a user name matching `DbUser` exists in the database, the temporary user credentials have the same permissions as the existing user. If `DbUser` doesn't exist in the database and `Autocreate` is `True`, a new user is created using the value for `DbUser` with PUBLIC permissions. If a database user matching the value for `DbUser` doesn't exist and `Autocreate` is `False`, then the command succeeds but the connection attempt will fail because the user doesn't exist in the database. For more information, see [CREATE USER][1] in the Amazon Redshift Database Developer Guide. Constraints: * Must be 1 to 64 alphanumeric characters or hyphens. The user name can't be `PUBLIC`. * Must contain only lowercase letters, numbers, underscore, plus sign, period (dot), at symbol (@), or hyphen. * First character must be a letter. * Must not contain a colon ( : ) or slash ( / ). * Cannot be a reserved word. A list of reserved words can be found in [Reserved Words][2] in the Amazon Redshift Database Developer Guide. [1]: https://docs.aws.amazon.com/redshift/latest/dg/r_CREATE_USER.html [2]: http://docs.aws.amazon.com/redshift/latest/dg/r_pg_keywords.html
@option params [String] :db_name
The name of a database that `DbUser` is authorized to log on to. If `DbName` is not specified, `DbUser` can log on to any existing database. Constraints: * Must be 1 to 64 alphanumeric characters or hyphens * Must contain only lowercase letters, numbers, underscore, plus sign, period (dot), at symbol (@), or hyphen. * First character must be a letter. * Must not contain a colon ( : ) or slash ( / ). * Cannot be a reserved word. A list of reserved words can be found in [Reserved Words][1] in the Amazon Redshift Database Developer Guide. [1]: http://docs.aws.amazon.com/redshift/latest/dg/r_pg_keywords.html
@option params [required, String] :cluster_identifier
The unique identifier of the cluster that contains the database for which your are requesting credentials. This parameter is case sensitive.
@option params [Integer] :duration_seconds
The number of seconds until the returned temporary password expires. Constraint: minimum 900, maximum 3600. Default: 900
@option params [Boolean] :auto_create
Create a database user with the name specified for the user named in `DbUser` if one does not exist.
@option params [Array<String>] :db_groups
A list of the names of existing database groups that the user named in `DbUser` will join for the current session, in addition to any group memberships for an existing user. If not specified, a new user is added only to PUBLIC. Database group name constraints * Must be 1 to 64 alphanumeric characters or hyphens * Must contain only lowercase letters, numbers, underscore, plus sign, period (dot), at symbol (@), or hyphen. * First character must be a letter. * Must not contain a colon ( : ) or slash ( / ). * Cannot be a reserved word. A list of reserved words can be found in [Reserved Words][1] in the Amazon Redshift Database Developer Guide. [1]: http://docs.aws.amazon.com/redshift/latest/dg/r_pg_keywords.html
@return [Types::ClusterCredentials] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::ClusterCredentials#db_user #db_user} => String * {Types::ClusterCredentials#db_password #db_password} => String * {Types::ClusterCredentials#expiration #expiration} => Time
@example Request syntax with placeholder values
resp = client.get_cluster_credentials({ db_user: "String", # required db_name: "String", cluster_identifier: "String", # required duration_seconds: 1, auto_create: false, db_groups: ["String"], })
@example Response structure
resp.db_user #=> String resp.db_password #=> String resp.expiration #=> Time
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/GetClusterCredentials AWS API Documentation
@overload get_cluster_credentials
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 7107 def get_cluster_credentials(params = {}, options = {}) req = build_request(:get_cluster_credentials, params) req.send_request(options) end
Returns an array of DC2 ReservedNodeOfferings that matches the payment type, term, and usage price of the given DC1 reserved node.
@option params [required, String] :reserved_node_id
A string representing the node identifier for the DC1 Reserved Node to be exchanged.
@option params [Integer] :max_records
An integer setting the maximum number of ReservedNodeOfferings to retrieve.
@option params [String] :marker
A value that indicates the starting point for the next set of ReservedNodeOfferings.
@return [Types::GetReservedNodeExchangeOfferingsOutputMessage] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::GetReservedNodeExchangeOfferingsOutputMessage#marker #marker} => String * {Types::GetReservedNodeExchangeOfferingsOutputMessage#reserved_node_offerings #reserved_node_offerings} => Array<Types::ReservedNodeOffering>
The returned {Seahorse::Client::Response response} is a pageable response and is Enumerable. For details on usage see {Aws::PageableResponse PageableResponse}.
@example Request syntax with placeholder values
resp = client.get_reserved_node_exchange_offerings({ reserved_node_id: "String", # required max_records: 1, marker: "String", })
@example Response structure
resp.marker #=> String resp.reserved_node_offerings #=> Array resp.reserved_node_offerings[0].reserved_node_offering_id #=> String resp.reserved_node_offerings[0].node_type #=> String resp.reserved_node_offerings[0].duration #=> Integer resp.reserved_node_offerings[0].fixed_price #=> Float resp.reserved_node_offerings[0].usage_price #=> Float resp.reserved_node_offerings[0].currency_code #=> String resp.reserved_node_offerings[0].offering_type #=> String resp.reserved_node_offerings[0].recurring_charges #=> Array resp.reserved_node_offerings[0].recurring_charges[0].recurring_charge_amount #=> Float resp.reserved_node_offerings[0].recurring_charges[0].recurring_charge_frequency #=> String resp.reserved_node_offerings[0].reserved_node_offering_type #=> String, one of "Regular", "Upgradable"
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/GetReservedNodeExchangeOfferings AWS API Documentation
@overload get_reserved_node_exchange_offerings
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 7162 def get_reserved_node_exchange_offerings(params = {}, options = {}) req = build_request(:get_reserved_node_exchange_offerings, params) req.send_request(options) end
Modifies whether a cluster can use AQUA (Advanced Query Accelerator).
@option params [required, String] :cluster_identifier
The identifier of the cluster to be modified.
@option params [String] :aqua_configuration_status
The new value of AQUA configuration status. Possible values include the following. * enabled - Use AQUA if it is available for the current Amazon Web Services Region and Amazon Redshift node type. * disabled - Don't use AQUA. * auto - Amazon Redshift determines whether to use AQUA.
@return [Types::ModifyAquaOutputMessage] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::ModifyAquaOutputMessage#aqua_configuration #aqua_configuration} => Types::AquaConfiguration
@example Request syntax with placeholder values
resp = client.modify_aqua_configuration({ cluster_identifier: "String", # required aqua_configuration_status: "enabled", # accepts enabled, disabled, auto })
@example Response structure
resp.aqua_configuration.aqua_status #=> String, one of "enabled", "disabled", "applying" resp.aqua_configuration.aqua_configuration_status #=> String, one of "enabled", "disabled", "auto"
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/ModifyAquaConfiguration AWS API Documentation
@overload modify_aqua_configuration
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 7203 def modify_aqua_configuration(params = {}, options = {}) req = build_request(:modify_aqua_configuration, params) req.send_request(options) end
Modifies an authentication profile.
@option params [required, String] :authentication_profile_name
The name of the authentication profile to replace.
@option params [required, String] :authentication_profile_content
The new content of the authentication profile in JSON format. The maximum length of the JSON string is determined by a quota for your account.
@return [Types::ModifyAuthenticationProfileResult] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::ModifyAuthenticationProfileResult#authentication_profile_name #authentication_profile_name} => String * {Types::ModifyAuthenticationProfileResult#authentication_profile_content #authentication_profile_content} => String
@example Request syntax with placeholder values
resp = client.modify_authentication_profile({ authentication_profile_name: "AuthenticationProfileNameString", # required authentication_profile_content: "String", # required })
@example Response structure
resp.authentication_profile_name #=> String resp.authentication_profile_content #=> String
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/ModifyAuthenticationProfile AWS API Documentation
@overload modify_authentication_profile
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 7239 def modify_authentication_profile(params = {}, options = {}) req = build_request(:modify_authentication_profile, params) req.send_request(options) end
Modifies the settings for a cluster.
You can also change node type and the number of nodes to scale up or down the cluster. When resizing a cluster, you must specify both the number of nodes and the node type even if one of the parameters does not change.
You can add another security or parameter group, or change the admin user password. Resetting a cluster password or modifying the security groups associated with a cluster do not need a reboot. However, modifying a parameter group requires a reboot for parameters to take effect. For more information about managing clusters, go to [Amazon Redshift
Clusters] in the *Amazon Redshift
Cluster Management Guide*.
[1]: docs.aws.amazon.com/redshift/latest/mgmt/working-with-clusters.html
@option params [required, String] :cluster_identifier
The unique identifier of the cluster to be modified. Example: `examplecluster`
@option params [String] :cluster_type
The new cluster type. When you submit your cluster resize request, your existing cluster goes into a read-only mode. After Amazon Redshift provisions a new cluster based on your resize requirements, there will be outage for a period while the old cluster is deleted and your connection is switched to the new cluster. You can use DescribeResize to track the progress of the resize request. Valid Values: ` multi-node | single-node `
@option params [String] :node_type
The new node type of the cluster. If you specify a new node type, you must also specify the number of nodes parameter. For more information about resizing clusters, go to [Resizing Clusters in Amazon Redshift][1] in the *Amazon Redshift Cluster Management Guide*. Valid Values: `ds2.xlarge` \| `ds2.8xlarge` \| `dc1.large` \| `dc1.8xlarge` \| `dc2.large` \| `dc2.8xlarge` \| `ra3.xlplus` \| `ra3.4xlarge` \| `ra3.16xlarge` [1]: https://docs.aws.amazon.com/redshift/latest/mgmt/rs-resize-tutorial.html
@option params [Integer] :number_of_nodes
The new number of nodes of the cluster. If you specify a new number of nodes, you must also specify the node type parameter. For more information about resizing clusters, go to [Resizing Clusters in Amazon Redshift][1] in the *Amazon Redshift Cluster Management Guide*. Valid Values: Integer greater than `0`. [1]: https://docs.aws.amazon.com/redshift/latest/mgmt/rs-resize-tutorial.html
@option params [Array<String>] :cluster_security_groups
A list of cluster security groups to be authorized on this cluster. This change is asynchronously applied as soon as possible. Security groups currently associated with the cluster, and not in the list of groups to apply, will be revoked from the cluster. Constraints: * Must be 1 to 255 alphanumeric characters or hyphens * First character must be a letter * Cannot end with a hyphen or contain two consecutive hyphens
@option params [Array<String>] :vpc_security_group_ids
A list of virtual private cloud (VPC) security groups to be associated with the cluster. This change is asynchronously applied as soon as possible.
@option params [String] :master_user_password
The new password for the cluster admin user. This change is asynchronously applied as soon as possible. Between the time of the request and the completion of the request, the `MasterUserPassword` element exists in the `PendingModifiedValues` element of the operation response. <note markdown="1"> Operations never return the password, so this operation provides a way to regain access to the admin user account for a cluster if the password is lost. </note> Default: Uses existing setting. Constraints: * Must be between 8 and 64 characters in length. * Must contain at least one uppercase letter. * Must contain at least one lowercase letter. * Must contain one number. * Can be any printable ASCII character (ASCII code 33 to 126) except ' (single quote), " (double quote), \\, /, @, or space.
@option params [String] :cluster_parameter_group_name
The name of the cluster parameter group to apply to this cluster. This change is applied only after the cluster is rebooted. To reboot a cluster use RebootCluster. Default: Uses existing setting. Constraints: The cluster parameter group must be in the same parameter group family that matches the cluster version.
@option params [Integer] :automated_snapshot_retention_period
The number of days that automated snapshots are retained. If the value is 0, automated snapshots are disabled. Even if automated snapshots are disabled, you can still create manual snapshots when you want with CreateClusterSnapshot. If you decrease the automated snapshot retention period from its current value, existing automated snapshots that fall outside of the new retention period will be immediately deleted. You can't disable automated snapshots for RA3 node types. Set the automated retention period from 1-35 days. Default: Uses existing setting. Constraints: Must be a value from 0 to 35.
@option params [Integer] :manual_snapshot_retention_period
The default for number of days that a newly created manual snapshot is retained. If the value is -1, the manual snapshot is retained indefinitely. This value doesn't retroactively change the retention periods of existing manual snapshots. The value must be either -1 or an integer between 1 and 3,653. The default value is -1.
@option params [String] :preferred_maintenance_window
The weekly time range (in UTC) during which system maintenance can occur, if necessary. If system maintenance is necessary during the window, it may result in an outage. This maintenance window change is made immediately. If the new maintenance window indicates the current time, there must be at least 120 minutes between the current time and end of the window in order to ensure that pending changes are applied. Default: Uses existing setting. Format: ddd:hh24:mi-ddd:hh24:mi, for example `wed:07:30-wed:08:00`. Valid Days: Mon \| Tue \| Wed \| Thu \| Fri \| Sat \| Sun Constraints: Must be at least 30 minutes.
@option params [String] :cluster_version
The new version number of the Amazon Redshift engine to upgrade to. For major version upgrades, if a non-default cluster parameter group is currently in use, a new cluster parameter group in the cluster parameter group family for the new version must be specified. The new cluster parameter group can be the default for that cluster parameter group family. For more information about parameters and parameter groups, go to [Amazon Redshift Parameter Groups][1] in the *Amazon Redshift Cluster Management Guide*. Example: `1.0` [1]: https://docs.aws.amazon.com/redshift/latest/mgmt/working-with-parameter-groups.html
@option params [Boolean] :allow_version_upgrade
If `true`, major version upgrades will be applied automatically to the cluster during the maintenance window. Default: `false`
@option params [String] :hsm_client_certificate_identifier
Specifies the name of the HSM client certificate the Amazon Redshift cluster uses to retrieve the data encryption keys stored in an HSM.
@option params [String] :hsm_configuration_identifier
Specifies the name of the HSM configuration that contains the information the Amazon Redshift cluster can use to retrieve and store keys in an HSM.
@option params [String] :new_cluster_identifier
The new identifier for the cluster. Constraints: * Must contain from 1 to 63 alphanumeric characters or hyphens. * Alphabetic characters must be lowercase. * First character must be a letter. * Cannot end with a hyphen or contain two consecutive hyphens. * Must be unique for all clusters within an Amazon Web Services account. Example: `examplecluster`
@option params [Boolean] :publicly_accessible
If `true`, the cluster can be accessed from a public network. Only clusters in VPCs can be set to be publicly available.
@option params [String] :elastic_ip
The Elastic IP (EIP) address for the cluster. Constraints: The cluster must be provisioned in EC2-VPC and publicly-accessible through an Internet gateway. For more information about provisioning clusters in EC2-VPC, go to [Supported Platforms to Launch Your Cluster][1] in the Amazon Redshift Cluster Management Guide. [1]: https://docs.aws.amazon.com/redshift/latest/mgmt/working-with-clusters.html#cluster-platforms
@option params [Boolean] :enhanced_vpc_routing
An option that specifies whether to create the cluster with enhanced VPC routing enabled. To create a cluster that uses enhanced VPC routing, the cluster must be in a VPC. For more information, see [Enhanced VPC Routing][1] in the Amazon Redshift Cluster Management Guide. If this option is `true`, enhanced VPC routing is enabled. Default: false [1]: https://docs.aws.amazon.com/redshift/latest/mgmt/enhanced-vpc-routing.html
@option params [String] :maintenance_track_name
The name for the maintenance track that you want to assign for the cluster. This name change is asynchronous. The new track name stays in the `PendingModifiedValues` for the cluster until the next maintenance window. When the maintenance track changes, the cluster is switched to the latest cluster release available for the maintenance track. At this point, the maintenance track name is applied.
@option params [Boolean] :encrypted
Indicates whether the cluster is encrypted. If the value is encrypted (true) and you provide a value for the `KmsKeyId` parameter, we encrypt the cluster with the provided `KmsKeyId`. If you don't provide a `KmsKeyId`, we encrypt with the default key. If the value is not encrypted (false), then the cluster is decrypted.
@option params [String] :kms_key_id
The Key Management Service (KMS) key ID of the encryption key that you want to use to encrypt data in the cluster.
@option params [Boolean] :availability_zone_relocation
The option to enable relocation for an Amazon Redshift cluster between Availability Zones after the cluster modification is complete.
@option params [String] :availability_zone
The option to initiate relocation for an Amazon Redshift cluster to the target Availability Zone.
@option params [Integer] :port
The option to change the port of an Amazon Redshift cluster.
@return [Types::ModifyClusterResult] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::ModifyClusterResult#cluster #cluster} => Types::Cluster
@example Request syntax with placeholder values
resp = client.modify_cluster({ cluster_identifier: "String", # required cluster_type: "String", node_type: "String", number_of_nodes: 1, cluster_security_groups: ["String"], vpc_security_group_ids: ["String"], master_user_password: "String", cluster_parameter_group_name: "String", automated_snapshot_retention_period: 1, manual_snapshot_retention_period: 1, preferred_maintenance_window: "String", cluster_version: "String", allow_version_upgrade: false, hsm_client_certificate_identifier: "String", hsm_configuration_identifier: "String", new_cluster_identifier: "String", publicly_accessible: false, elastic_ip: "String", enhanced_vpc_routing: false, maintenance_track_name: "String", encrypted: false, kms_key_id: "String", availability_zone_relocation: false, availability_zone: "String", port: 1, })
@example Response structure
resp.cluster.cluster_identifier #=> String resp.cluster.node_type #=> String resp.cluster.cluster_status #=> String resp.cluster.cluster_availability_status #=> String resp.cluster.modify_status #=> String resp.cluster.master_username #=> String resp.cluster.db_name #=> String resp.cluster.endpoint.address #=> String resp.cluster.endpoint.port #=> Integer resp.cluster.endpoint.vpc_endpoints #=> Array resp.cluster.endpoint.vpc_endpoints[0].vpc_endpoint_id #=> String resp.cluster.endpoint.vpc_endpoints[0].vpc_id #=> String resp.cluster.endpoint.vpc_endpoints[0].network_interfaces #=> Array resp.cluster.endpoint.vpc_endpoints[0].network_interfaces[0].network_interface_id #=> String resp.cluster.endpoint.vpc_endpoints[0].network_interfaces[0].subnet_id #=> String resp.cluster.endpoint.vpc_endpoints[0].network_interfaces[0].private_ip_address #=> String resp.cluster.endpoint.vpc_endpoints[0].network_interfaces[0].availability_zone #=> String resp.cluster.cluster_create_time #=> Time resp.cluster.automated_snapshot_retention_period #=> Integer resp.cluster.manual_snapshot_retention_period #=> Integer resp.cluster.cluster_security_groups #=> Array resp.cluster.cluster_security_groups[0].cluster_security_group_name #=> String resp.cluster.cluster_security_groups[0].status #=> String resp.cluster.vpc_security_groups #=> Array resp.cluster.vpc_security_groups[0].vpc_security_group_id #=> String resp.cluster.vpc_security_groups[0].status #=> String resp.cluster.cluster_parameter_groups #=> Array resp.cluster.cluster_parameter_groups[0].parameter_group_name #=> String resp.cluster.cluster_parameter_groups[0].parameter_apply_status #=> String resp.cluster.cluster_parameter_groups[0].cluster_parameter_status_list #=> Array resp.cluster.cluster_parameter_groups[0].cluster_parameter_status_list[0].parameter_name #=> String resp.cluster.cluster_parameter_groups[0].cluster_parameter_status_list[0].parameter_apply_status #=> String resp.cluster.cluster_parameter_groups[0].cluster_parameter_status_list[0].parameter_apply_error_description #=> String resp.cluster.cluster_subnet_group_name #=> String resp.cluster.vpc_id #=> String resp.cluster.availability_zone #=> String resp.cluster.preferred_maintenance_window #=> String resp.cluster.pending_modified_values.master_user_password #=> String resp.cluster.pending_modified_values.node_type #=> String resp.cluster.pending_modified_values.number_of_nodes #=> Integer resp.cluster.pending_modified_values.cluster_type #=> String resp.cluster.pending_modified_values.cluster_version #=> String resp.cluster.pending_modified_values.automated_snapshot_retention_period #=> Integer resp.cluster.pending_modified_values.cluster_identifier #=> String resp.cluster.pending_modified_values.publicly_accessible #=> Boolean resp.cluster.pending_modified_values.enhanced_vpc_routing #=> Boolean resp.cluster.pending_modified_values.maintenance_track_name #=> String resp.cluster.pending_modified_values.encryption_type #=> String resp.cluster.cluster_version #=> String resp.cluster.allow_version_upgrade #=> Boolean resp.cluster.number_of_nodes #=> Integer resp.cluster.publicly_accessible #=> Boolean resp.cluster.encrypted #=> Boolean resp.cluster.restore_status.status #=> String resp.cluster.restore_status.current_restore_rate_in_mega_bytes_per_second #=> Float resp.cluster.restore_status.snapshot_size_in_mega_bytes #=> Integer resp.cluster.restore_status.progress_in_mega_bytes #=> Integer resp.cluster.restore_status.elapsed_time_in_seconds #=> Integer resp.cluster.restore_status.estimated_time_to_completion_in_seconds #=> Integer resp.cluster.data_transfer_progress.status #=> String resp.cluster.data_transfer_progress.current_rate_in_mega_bytes_per_second #=> Float resp.cluster.data_transfer_progress.total_data_in_mega_bytes #=> Integer resp.cluster.data_transfer_progress.data_transferred_in_mega_bytes #=> Integer resp.cluster.data_transfer_progress.estimated_time_to_completion_in_seconds #=> Integer resp.cluster.data_transfer_progress.elapsed_time_in_seconds #=> Integer resp.cluster.hsm_status.hsm_client_certificate_identifier #=> String resp.cluster.hsm_status.hsm_configuration_identifier #=> String resp.cluster.hsm_status.status #=> String resp.cluster.cluster_snapshot_copy_status.destination_region #=> String resp.cluster.cluster_snapshot_copy_status.retention_period #=> Integer resp.cluster.cluster_snapshot_copy_status.manual_snapshot_retention_period #=> Integer resp.cluster.cluster_snapshot_copy_status.snapshot_copy_grant_name #=> String resp.cluster.cluster_public_key #=> String resp.cluster.cluster_nodes #=> Array resp.cluster.cluster_nodes[0].node_role #=> String resp.cluster.cluster_nodes[0].private_ip_address #=> String resp.cluster.cluster_nodes[0].public_ip_address #=> String resp.cluster.elastic_ip_status.elastic_ip #=> String resp.cluster.elastic_ip_status.status #=> String resp.cluster.cluster_revision_number #=> String resp.cluster.tags #=> Array resp.cluster.tags[0].key #=> String resp.cluster.tags[0].value #=> String resp.cluster.kms_key_id #=> String resp.cluster.enhanced_vpc_routing #=> Boolean resp.cluster.iam_roles #=> Array resp.cluster.iam_roles[0].iam_role_arn #=> String resp.cluster.iam_roles[0].apply_status #=> String resp.cluster.pending_actions #=> Array resp.cluster.pending_actions[0] #=> String resp.cluster.maintenance_track_name #=> String resp.cluster.elastic_resize_number_of_node_options #=> String resp.cluster.deferred_maintenance_windows #=> Array resp.cluster.deferred_maintenance_windows[0].defer_maintenance_identifier #=> String resp.cluster.deferred_maintenance_windows[0].defer_maintenance_start_time #=> Time resp.cluster.deferred_maintenance_windows[0].defer_maintenance_end_time #=> Time resp.cluster.snapshot_schedule_identifier #=> String resp.cluster.snapshot_schedule_state #=> String, one of "MODIFYING", "ACTIVE", "FAILED" resp.cluster.expected_next_snapshot_schedule_time #=> Time resp.cluster.expected_next_snapshot_schedule_time_status #=> String resp.cluster.next_maintenance_window_start_time #=> Time resp.cluster.resize_info.resize_type #=> String resp.cluster.resize_info.allow_cancel_resize #=> Boolean resp.cluster.availability_zone_relocation_status #=> String resp.cluster.cluster_namespace_arn #=> String resp.cluster.total_storage_capacity_in_mega_bytes #=> Integer resp.cluster.aqua_configuration.aqua_status #=> String, one of "enabled", "disabled", "applying" resp.cluster.aqua_configuration.aqua_configuration_status #=> String, one of "enabled", "disabled", "auto"
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/ModifyCluster AWS API Documentation
@overload modify_cluster
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 7675 def modify_cluster(params = {}, options = {}) req = build_request(:modify_cluster, params) req.send_request(options) end
Modifies the database revision of a cluster. The database revision is a unique revision of the database running in a cluster.
@option params [required, String] :cluster_identifier
The unique identifier of a cluster whose database revision you want to modify. Example: `examplecluster`
@option params [required, String] :revision_target
The identifier of the database revision. You can retrieve this value from the response to the DescribeClusterDbRevisions request.
@return [Types::ModifyClusterDbRevisionResult] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::ModifyClusterDbRevisionResult#cluster #cluster} => Types::Cluster
@example Request syntax with placeholder values
resp = client.modify_cluster_db_revision({ cluster_identifier: "String", # required revision_target: "String", # required })
@example Response structure
resp.cluster.cluster_identifier #=> String resp.cluster.node_type #=> String resp.cluster.cluster_status #=> String resp.cluster.cluster_availability_status #=> String resp.cluster.modify_status #=> String resp.cluster.master_username #=> String resp.cluster.db_name #=> String resp.cluster.endpoint.address #=> String resp.cluster.endpoint.port #=> Integer resp.cluster.endpoint.vpc_endpoints #=> Array resp.cluster.endpoint.vpc_endpoints[0].vpc_endpoint_id #=> String resp.cluster.endpoint.vpc_endpoints[0].vpc_id #=> String resp.cluster.endpoint.vpc_endpoints[0].network_interfaces #=> Array resp.cluster.endpoint.vpc_endpoints[0].network_interfaces[0].network_interface_id #=> String resp.cluster.endpoint.vpc_endpoints[0].network_interfaces[0].subnet_id #=> String resp.cluster.endpoint.vpc_endpoints[0].network_interfaces[0].private_ip_address #=> String resp.cluster.endpoint.vpc_endpoints[0].network_interfaces[0].availability_zone #=> String resp.cluster.cluster_create_time #=> Time resp.cluster.automated_snapshot_retention_period #=> Integer resp.cluster.manual_snapshot_retention_period #=> Integer resp.cluster.cluster_security_groups #=> Array resp.cluster.cluster_security_groups[0].cluster_security_group_name #=> String resp.cluster.cluster_security_groups[0].status #=> String resp.cluster.vpc_security_groups #=> Array resp.cluster.vpc_security_groups[0].vpc_security_group_id #=> String resp.cluster.vpc_security_groups[0].status #=> String resp.cluster.cluster_parameter_groups #=> Array resp.cluster.cluster_parameter_groups[0].parameter_group_name #=> String resp.cluster.cluster_parameter_groups[0].parameter_apply_status #=> String resp.cluster.cluster_parameter_groups[0].cluster_parameter_status_list #=> Array resp.cluster.cluster_parameter_groups[0].cluster_parameter_status_list[0].parameter_name #=> String resp.cluster.cluster_parameter_groups[0].cluster_parameter_status_list[0].parameter_apply_status #=> String resp.cluster.cluster_parameter_groups[0].cluster_parameter_status_list[0].parameter_apply_error_description #=> String resp.cluster.cluster_subnet_group_name #=> String resp.cluster.vpc_id #=> String resp.cluster.availability_zone #=> String resp.cluster.preferred_maintenance_window #=> String resp.cluster.pending_modified_values.master_user_password #=> String resp.cluster.pending_modified_values.node_type #=> String resp.cluster.pending_modified_values.number_of_nodes #=> Integer resp.cluster.pending_modified_values.cluster_type #=> String resp.cluster.pending_modified_values.cluster_version #=> String resp.cluster.pending_modified_values.automated_snapshot_retention_period #=> Integer resp.cluster.pending_modified_values.cluster_identifier #=> String resp.cluster.pending_modified_values.publicly_accessible #=> Boolean resp.cluster.pending_modified_values.enhanced_vpc_routing #=> Boolean resp.cluster.pending_modified_values.maintenance_track_name #=> String resp.cluster.pending_modified_values.encryption_type #=> String resp.cluster.cluster_version #=> String resp.cluster.allow_version_upgrade #=> Boolean resp.cluster.number_of_nodes #=> Integer resp.cluster.publicly_accessible #=> Boolean resp.cluster.encrypted #=> Boolean resp.cluster.restore_status.status #=> String resp.cluster.restore_status.current_restore_rate_in_mega_bytes_per_second #=> Float resp.cluster.restore_status.snapshot_size_in_mega_bytes #=> Integer resp.cluster.restore_status.progress_in_mega_bytes #=> Integer resp.cluster.restore_status.elapsed_time_in_seconds #=> Integer resp.cluster.restore_status.estimated_time_to_completion_in_seconds #=> Integer resp.cluster.data_transfer_progress.status #=> String resp.cluster.data_transfer_progress.current_rate_in_mega_bytes_per_second #=> Float resp.cluster.data_transfer_progress.total_data_in_mega_bytes #=> Integer resp.cluster.data_transfer_progress.data_transferred_in_mega_bytes #=> Integer resp.cluster.data_transfer_progress.estimated_time_to_completion_in_seconds #=> Integer resp.cluster.data_transfer_progress.elapsed_time_in_seconds #=> Integer resp.cluster.hsm_status.hsm_client_certificate_identifier #=> String resp.cluster.hsm_status.hsm_configuration_identifier #=> String resp.cluster.hsm_status.status #=> String resp.cluster.cluster_snapshot_copy_status.destination_region #=> String resp.cluster.cluster_snapshot_copy_status.retention_period #=> Integer resp.cluster.cluster_snapshot_copy_status.manual_snapshot_retention_period #=> Integer resp.cluster.cluster_snapshot_copy_status.snapshot_copy_grant_name #=> String resp.cluster.cluster_public_key #=> String resp.cluster.cluster_nodes #=> Array resp.cluster.cluster_nodes[0].node_role #=> String resp.cluster.cluster_nodes[0].private_ip_address #=> String resp.cluster.cluster_nodes[0].public_ip_address #=> String resp.cluster.elastic_ip_status.elastic_ip #=> String resp.cluster.elastic_ip_status.status #=> String resp.cluster.cluster_revision_number #=> String resp.cluster.tags #=> Array resp.cluster.tags[0].key #=> String resp.cluster.tags[0].value #=> String resp.cluster.kms_key_id #=> String resp.cluster.enhanced_vpc_routing #=> Boolean resp.cluster.iam_roles #=> Array resp.cluster.iam_roles[0].iam_role_arn #=> String resp.cluster.iam_roles[0].apply_status #=> String resp.cluster.pending_actions #=> Array resp.cluster.pending_actions[0] #=> String resp.cluster.maintenance_track_name #=> String resp.cluster.elastic_resize_number_of_node_options #=> String resp.cluster.deferred_maintenance_windows #=> Array resp.cluster.deferred_maintenance_windows[0].defer_maintenance_identifier #=> String resp.cluster.deferred_maintenance_windows[0].defer_maintenance_start_time #=> Time resp.cluster.deferred_maintenance_windows[0].defer_maintenance_end_time #=> Time resp.cluster.snapshot_schedule_identifier #=> String resp.cluster.snapshot_schedule_state #=> String, one of "MODIFYING", "ACTIVE", "FAILED" resp.cluster.expected_next_snapshot_schedule_time #=> Time resp.cluster.expected_next_snapshot_schedule_time_status #=> String resp.cluster.next_maintenance_window_start_time #=> Time resp.cluster.resize_info.resize_type #=> String resp.cluster.resize_info.allow_cancel_resize #=> Boolean resp.cluster.availability_zone_relocation_status #=> String resp.cluster.cluster_namespace_arn #=> String resp.cluster.total_storage_capacity_in_mega_bytes #=> Integer resp.cluster.aqua_configuration.aqua_status #=> String, one of "enabled", "disabled", "applying" resp.cluster.aqua_configuration.aqua_configuration_status #=> String, one of "enabled", "disabled", "auto"
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/ModifyClusterDbRevision AWS API Documentation
@overload modify_cluster_db_revision
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 7819 def modify_cluster_db_revision(params = {}, options = {}) req = build_request(:modify_cluster_db_revision, params) req.send_request(options) end
Modifies the list of Identity and Access Management (IAM) roles that can be used by the cluster to access other Amazon Web Services services.
A cluster can have up to 10 IAM roles associated at any time.
@option params [required, String] :cluster_identifier
The unique identifier of the cluster for which you want to associate or disassociate IAM roles.
@option params [Array<String>] :add_iam_roles
Zero or more IAM roles to associate with the cluster. The roles must be in their Amazon Resource Name (ARN) format. You can associate up to 10 IAM roles with a single cluster in a single request.
@option params [Array<String>] :remove_iam_roles
Zero or more IAM roles in ARN format to disassociate from the cluster. You can disassociate up to 10 IAM roles from a single cluster in a single request.
@return [Types::ModifyClusterIamRolesResult] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::ModifyClusterIamRolesResult#cluster #cluster} => Types::Cluster
@example Request syntax with placeholder values
resp = client.modify_cluster_iam_roles({ cluster_identifier: "String", # required add_iam_roles: ["String"], remove_iam_roles: ["String"], })
@example Response structure
resp.cluster.cluster_identifier #=> String resp.cluster.node_type #=> String resp.cluster.cluster_status #=> String resp.cluster.cluster_availability_status #=> String resp.cluster.modify_status #=> String resp.cluster.master_username #=> String resp.cluster.db_name #=> String resp.cluster.endpoint.address #=> String resp.cluster.endpoint.port #=> Integer resp.cluster.endpoint.vpc_endpoints #=> Array resp.cluster.endpoint.vpc_endpoints[0].vpc_endpoint_id #=> String resp.cluster.endpoint.vpc_endpoints[0].vpc_id #=> String resp.cluster.endpoint.vpc_endpoints[0].network_interfaces #=> Array resp.cluster.endpoint.vpc_endpoints[0].network_interfaces[0].network_interface_id #=> String resp.cluster.endpoint.vpc_endpoints[0].network_interfaces[0].subnet_id #=> String resp.cluster.endpoint.vpc_endpoints[0].network_interfaces[0].private_ip_address #=> String resp.cluster.endpoint.vpc_endpoints[0].network_interfaces[0].availability_zone #=> String resp.cluster.cluster_create_time #=> Time resp.cluster.automated_snapshot_retention_period #=> Integer resp.cluster.manual_snapshot_retention_period #=> Integer resp.cluster.cluster_security_groups #=> Array resp.cluster.cluster_security_groups[0].cluster_security_group_name #=> String resp.cluster.cluster_security_groups[0].status #=> String resp.cluster.vpc_security_groups #=> Array resp.cluster.vpc_security_groups[0].vpc_security_group_id #=> String resp.cluster.vpc_security_groups[0].status #=> String resp.cluster.cluster_parameter_groups #=> Array resp.cluster.cluster_parameter_groups[0].parameter_group_name #=> String resp.cluster.cluster_parameter_groups[0].parameter_apply_status #=> String resp.cluster.cluster_parameter_groups[0].cluster_parameter_status_list #=> Array resp.cluster.cluster_parameter_groups[0].cluster_parameter_status_list[0].parameter_name #=> String resp.cluster.cluster_parameter_groups[0].cluster_parameter_status_list[0].parameter_apply_status #=> String resp.cluster.cluster_parameter_groups[0].cluster_parameter_status_list[0].parameter_apply_error_description #=> String resp.cluster.cluster_subnet_group_name #=> String resp.cluster.vpc_id #=> String resp.cluster.availability_zone #=> String resp.cluster.preferred_maintenance_window #=> String resp.cluster.pending_modified_values.master_user_password #=> String resp.cluster.pending_modified_values.node_type #=> String resp.cluster.pending_modified_values.number_of_nodes #=> Integer resp.cluster.pending_modified_values.cluster_type #=> String resp.cluster.pending_modified_values.cluster_version #=> String resp.cluster.pending_modified_values.automated_snapshot_retention_period #=> Integer resp.cluster.pending_modified_values.cluster_identifier #=> String resp.cluster.pending_modified_values.publicly_accessible #=> Boolean resp.cluster.pending_modified_values.enhanced_vpc_routing #=> Boolean resp.cluster.pending_modified_values.maintenance_track_name #=> String resp.cluster.pending_modified_values.encryption_type #=> String resp.cluster.cluster_version #=> String resp.cluster.allow_version_upgrade #=> Boolean resp.cluster.number_of_nodes #=> Integer resp.cluster.publicly_accessible #=> Boolean resp.cluster.encrypted #=> Boolean resp.cluster.restore_status.status #=> String resp.cluster.restore_status.current_restore_rate_in_mega_bytes_per_second #=> Float resp.cluster.restore_status.snapshot_size_in_mega_bytes #=> Integer resp.cluster.restore_status.progress_in_mega_bytes #=> Integer resp.cluster.restore_status.elapsed_time_in_seconds #=> Integer resp.cluster.restore_status.estimated_time_to_completion_in_seconds #=> Integer resp.cluster.data_transfer_progress.status #=> String resp.cluster.data_transfer_progress.current_rate_in_mega_bytes_per_second #=> Float resp.cluster.data_transfer_progress.total_data_in_mega_bytes #=> Integer resp.cluster.data_transfer_progress.data_transferred_in_mega_bytes #=> Integer resp.cluster.data_transfer_progress.estimated_time_to_completion_in_seconds #=> Integer resp.cluster.data_transfer_progress.elapsed_time_in_seconds #=> Integer resp.cluster.hsm_status.hsm_client_certificate_identifier #=> String resp.cluster.hsm_status.hsm_configuration_identifier #=> String resp.cluster.hsm_status.status #=> String resp.cluster.cluster_snapshot_copy_status.destination_region #=> String resp.cluster.cluster_snapshot_copy_status.retention_period #=> Integer resp.cluster.cluster_snapshot_copy_status.manual_snapshot_retention_period #=> Integer resp.cluster.cluster_snapshot_copy_status.snapshot_copy_grant_name #=> String resp.cluster.cluster_public_key #=> String resp.cluster.cluster_nodes #=> Array resp.cluster.cluster_nodes[0].node_role #=> String resp.cluster.cluster_nodes[0].private_ip_address #=> String resp.cluster.cluster_nodes[0].public_ip_address #=> String resp.cluster.elastic_ip_status.elastic_ip #=> String resp.cluster.elastic_ip_status.status #=> String resp.cluster.cluster_revision_number #=> String resp.cluster.tags #=> Array resp.cluster.tags[0].key #=> String resp.cluster.tags[0].value #=> String resp.cluster.kms_key_id #=> String resp.cluster.enhanced_vpc_routing #=> Boolean resp.cluster.iam_roles #=> Array resp.cluster.iam_roles[0].iam_role_arn #=> String resp.cluster.iam_roles[0].apply_status #=> String resp.cluster.pending_actions #=> Array resp.cluster.pending_actions[0] #=> String resp.cluster.maintenance_track_name #=> String resp.cluster.elastic_resize_number_of_node_options #=> String resp.cluster.deferred_maintenance_windows #=> Array resp.cluster.deferred_maintenance_windows[0].defer_maintenance_identifier #=> String resp.cluster.deferred_maintenance_windows[0].defer_maintenance_start_time #=> Time resp.cluster.deferred_maintenance_windows[0].defer_maintenance_end_time #=> Time resp.cluster.snapshot_schedule_identifier #=> String resp.cluster.snapshot_schedule_state #=> String, one of "MODIFYING", "ACTIVE", "FAILED" resp.cluster.expected_next_snapshot_schedule_time #=> Time resp.cluster.expected_next_snapshot_schedule_time_status #=> String resp.cluster.next_maintenance_window_start_time #=> Time resp.cluster.resize_info.resize_type #=> String resp.cluster.resize_info.allow_cancel_resize #=> Boolean resp.cluster.availability_zone_relocation_status #=> String resp.cluster.cluster_namespace_arn #=> String resp.cluster.total_storage_capacity_in_mega_bytes #=> Integer resp.cluster.aqua_configuration.aqua_status #=> String, one of "enabled", "disabled", "applying" resp.cluster.aqua_configuration.aqua_configuration_status #=> String, one of "enabled", "disabled", "auto"
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/ModifyClusterIamRoles AWS API Documentation
@overload modify_cluster_iam_roles
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 7971 def modify_cluster_iam_roles(params = {}, options = {}) req = build_request(:modify_cluster_iam_roles, params) req.send_request(options) end
Modifies the maintenance settings of a cluster.
@option params [required, String] :cluster_identifier
A unique identifier for the cluster.
@option params [Boolean] :defer_maintenance
A boolean indicating whether to enable the deferred maintenance window.
@option params [String] :defer_maintenance_identifier
A unique identifier for the deferred maintenance window.
@option params [Time,DateTime,Date,Integer,String] :defer_maintenance_start_time
A timestamp indicating the start time for the deferred maintenance window.
@option params [Time,DateTime,Date,Integer,String] :defer_maintenance_end_time
A timestamp indicating end time for the deferred maintenance window. If you specify an end time, you can't specify a duration.
@option params [Integer] :defer_maintenance_duration
An integer indicating the duration of the maintenance window in days. If you specify a duration, you can't specify an end time. The duration must be 45 days or less.
@return [Types::ModifyClusterMaintenanceResult] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::ModifyClusterMaintenanceResult#cluster #cluster} => Types::Cluster
@example Request syntax with placeholder values
resp = client.modify_cluster_maintenance({ cluster_identifier: "String", # required defer_maintenance: false, defer_maintenance_identifier: "String", defer_maintenance_start_time: Time.now, defer_maintenance_end_time: Time.now, defer_maintenance_duration: 1, })
@example Response structure
resp.cluster.cluster_identifier #=> String resp.cluster.node_type #=> String resp.cluster.cluster_status #=> String resp.cluster.cluster_availability_status #=> String resp.cluster.modify_status #=> String resp.cluster.master_username #=> String resp.cluster.db_name #=> String resp.cluster.endpoint.address #=> String resp.cluster.endpoint.port #=> Integer resp.cluster.endpoint.vpc_endpoints #=> Array resp.cluster.endpoint.vpc_endpoints[0].vpc_endpoint_id #=> String resp.cluster.endpoint.vpc_endpoints[0].vpc_id #=> String resp.cluster.endpoint.vpc_endpoints[0].network_interfaces #=> Array resp.cluster.endpoint.vpc_endpoints[0].network_interfaces[0].network_interface_id #=> String resp.cluster.endpoint.vpc_endpoints[0].network_interfaces[0].subnet_id #=> String resp.cluster.endpoint.vpc_endpoints[0].network_interfaces[0].private_ip_address #=> String resp.cluster.endpoint.vpc_endpoints[0].network_interfaces[0].availability_zone #=> String resp.cluster.cluster_create_time #=> Time resp.cluster.automated_snapshot_retention_period #=> Integer resp.cluster.manual_snapshot_retention_period #=> Integer resp.cluster.cluster_security_groups #=> Array resp.cluster.cluster_security_groups[0].cluster_security_group_name #=> String resp.cluster.cluster_security_groups[0].status #=> String resp.cluster.vpc_security_groups #=> Array resp.cluster.vpc_security_groups[0].vpc_security_group_id #=> String resp.cluster.vpc_security_groups[0].status #=> String resp.cluster.cluster_parameter_groups #=> Array resp.cluster.cluster_parameter_groups[0].parameter_group_name #=> String resp.cluster.cluster_parameter_groups[0].parameter_apply_status #=> String resp.cluster.cluster_parameter_groups[0].cluster_parameter_status_list #=> Array resp.cluster.cluster_parameter_groups[0].cluster_parameter_status_list[0].parameter_name #=> String resp.cluster.cluster_parameter_groups[0].cluster_parameter_status_list[0].parameter_apply_status #=> String resp.cluster.cluster_parameter_groups[0].cluster_parameter_status_list[0].parameter_apply_error_description #=> String resp.cluster.cluster_subnet_group_name #=> String resp.cluster.vpc_id #=> String resp.cluster.availability_zone #=> String resp.cluster.preferred_maintenance_window #=> String resp.cluster.pending_modified_values.master_user_password #=> String resp.cluster.pending_modified_values.node_type #=> String resp.cluster.pending_modified_values.number_of_nodes #=> Integer resp.cluster.pending_modified_values.cluster_type #=> String resp.cluster.pending_modified_values.cluster_version #=> String resp.cluster.pending_modified_values.automated_snapshot_retention_period #=> Integer resp.cluster.pending_modified_values.cluster_identifier #=> String resp.cluster.pending_modified_values.publicly_accessible #=> Boolean resp.cluster.pending_modified_values.enhanced_vpc_routing #=> Boolean resp.cluster.pending_modified_values.maintenance_track_name #=> String resp.cluster.pending_modified_values.encryption_type #=> String resp.cluster.cluster_version #=> String resp.cluster.allow_version_upgrade #=> Boolean resp.cluster.number_of_nodes #=> Integer resp.cluster.publicly_accessible #=> Boolean resp.cluster.encrypted #=> Boolean resp.cluster.restore_status.status #=> String resp.cluster.restore_status.current_restore_rate_in_mega_bytes_per_second #=> Float resp.cluster.restore_status.snapshot_size_in_mega_bytes #=> Integer resp.cluster.restore_status.progress_in_mega_bytes #=> Integer resp.cluster.restore_status.elapsed_time_in_seconds #=> Integer resp.cluster.restore_status.estimated_time_to_completion_in_seconds #=> Integer resp.cluster.data_transfer_progress.status #=> String resp.cluster.data_transfer_progress.current_rate_in_mega_bytes_per_second #=> Float resp.cluster.data_transfer_progress.total_data_in_mega_bytes #=> Integer resp.cluster.data_transfer_progress.data_transferred_in_mega_bytes #=> Integer resp.cluster.data_transfer_progress.estimated_time_to_completion_in_seconds #=> Integer resp.cluster.data_transfer_progress.elapsed_time_in_seconds #=> Integer resp.cluster.hsm_status.hsm_client_certificate_identifier #=> String resp.cluster.hsm_status.hsm_configuration_identifier #=> String resp.cluster.hsm_status.status #=> String resp.cluster.cluster_snapshot_copy_status.destination_region #=> String resp.cluster.cluster_snapshot_copy_status.retention_period #=> Integer resp.cluster.cluster_snapshot_copy_status.manual_snapshot_retention_period #=> Integer resp.cluster.cluster_snapshot_copy_status.snapshot_copy_grant_name #=> String resp.cluster.cluster_public_key #=> String resp.cluster.cluster_nodes #=> Array resp.cluster.cluster_nodes[0].node_role #=> String resp.cluster.cluster_nodes[0].private_ip_address #=> String resp.cluster.cluster_nodes[0].public_ip_address #=> String resp.cluster.elastic_ip_status.elastic_ip #=> String resp.cluster.elastic_ip_status.status #=> String resp.cluster.cluster_revision_number #=> String resp.cluster.tags #=> Array resp.cluster.tags[0].key #=> String resp.cluster.tags[0].value #=> String resp.cluster.kms_key_id #=> String resp.cluster.enhanced_vpc_routing #=> Boolean resp.cluster.iam_roles #=> Array resp.cluster.iam_roles[0].iam_role_arn #=> String resp.cluster.iam_roles[0].apply_status #=> String resp.cluster.pending_actions #=> Array resp.cluster.pending_actions[0] #=> String resp.cluster.maintenance_track_name #=> String resp.cluster.elastic_resize_number_of_node_options #=> String resp.cluster.deferred_maintenance_windows #=> Array resp.cluster.deferred_maintenance_windows[0].defer_maintenance_identifier #=> String resp.cluster.deferred_maintenance_windows[0].defer_maintenance_start_time #=> Time resp.cluster.deferred_maintenance_windows[0].defer_maintenance_end_time #=> Time resp.cluster.snapshot_schedule_identifier #=> String resp.cluster.snapshot_schedule_state #=> String, one of "MODIFYING", "ACTIVE", "FAILED" resp.cluster.expected_next_snapshot_schedule_time #=> Time resp.cluster.expected_next_snapshot_schedule_time_status #=> String resp.cluster.next_maintenance_window_start_time #=> Time resp.cluster.resize_info.resize_type #=> String resp.cluster.resize_info.allow_cancel_resize #=> Boolean resp.cluster.availability_zone_relocation_status #=> String resp.cluster.cluster_namespace_arn #=> String resp.cluster.total_storage_capacity_in_mega_bytes #=> Integer resp.cluster.aqua_configuration.aqua_status #=> String, one of "enabled", "disabled", "applying" resp.cluster.aqua_configuration.aqua_configuration_status #=> String, one of "enabled", "disabled", "auto"
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/ModifyClusterMaintenance AWS API Documentation
@overload modify_cluster_maintenance
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 8131 def modify_cluster_maintenance(params = {}, options = {}) req = build_request(:modify_cluster_maintenance, params) req.send_request(options) end
Modifies the parameters of a parameter group. For the parameters parameter, it can't contain ASCII characters.
For more information about parameters and parameter groups, go to
Management Guide*.
[1]: docs.aws.amazon.com/redshift/latest/mgmt/working-with-parameter-groups.html
@option params [required, String] :parameter_group_name
The name of the parameter group to be modified.
@option params [required, Array<Types::Parameter>] :parameters
An array of parameters to be modified. A maximum of 20 parameters can be modified in a single request. For each parameter to be modified, you must supply at least the parameter name and parameter value; other name-value pairs of the parameter are optional. For the workload management (WLM) configuration, you must supply all the name-value pairs in the wlm\_json\_configuration parameter.
@return [Types::ClusterParameterGroupNameMessage] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::ClusterParameterGroupNameMessage#parameter_group_name #parameter_group_name} => String * {Types::ClusterParameterGroupNameMessage#parameter_group_status #parameter_group_status} => String
@example Request syntax with placeholder values
resp = client.modify_cluster_parameter_group({ parameter_group_name: "String", # required parameters: [ # required { parameter_name: "String", parameter_value: "String", description: "String", source: "String", data_type: "String", allowed_values: "String", apply_type: "static", # accepts static, dynamic is_modifiable: false, minimum_engine_version: "String", }, ], })
@example Response structure
resp.parameter_group_name #=> String resp.parameter_group_status #=> String
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/ModifyClusterParameterGroup AWS API Documentation
@overload modify_cluster_parameter_group
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 8194 def modify_cluster_parameter_group(params = {}, options = {}) req = build_request(:modify_cluster_parameter_group, params) req.send_request(options) end
Modifies the settings for a snapshot.
This exanmple modifies the manual retention period setting for a cluster snapshot.
@option params [required, String] :snapshot_identifier
The identifier of the snapshot whose setting you want to modify.
@option params [Integer] :manual_snapshot_retention_period
The number of days that a manual snapshot is retained. If the value is -1, the manual snapshot is retained indefinitely. If the manual snapshot falls outside of the new retention period, you can specify the force option to immediately delete the snapshot. The value must be either -1 or an integer between 1 and 3,653.
@option params [Boolean] :force
A Boolean option to override an exception if the retention period has already passed.
@return [Types::ModifyClusterSnapshotResult] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::ModifyClusterSnapshotResult#snapshot #snapshot} => Types::Snapshot
@example Request syntax with placeholder values
resp = client.modify_cluster_snapshot({ snapshot_identifier: "String", # required manual_snapshot_retention_period: 1, force: false, })
@example Response structure
resp.snapshot.snapshot_identifier #=> String resp.snapshot.cluster_identifier #=> String resp.snapshot.snapshot_create_time #=> Time resp.snapshot.status #=> String resp.snapshot.port #=> Integer resp.snapshot.availability_zone #=> String resp.snapshot.cluster_create_time #=> Time resp.snapshot.master_username #=> String resp.snapshot.cluster_version #=> String resp.snapshot.engine_full_version #=> String resp.snapshot.snapshot_type #=> String resp.snapshot.node_type #=> String resp.snapshot.number_of_nodes #=> Integer resp.snapshot.db_name #=> String resp.snapshot.vpc_id #=> String resp.snapshot.encrypted #=> Boolean resp.snapshot.kms_key_id #=> String resp.snapshot.encrypted_with_hsm #=> Boolean resp.snapshot.accounts_with_restore_access #=> Array resp.snapshot.accounts_with_restore_access[0].account_id #=> String resp.snapshot.accounts_with_restore_access[0].account_alias #=> String resp.snapshot.owner_account #=> String resp.snapshot.total_backup_size_in_mega_bytes #=> Float resp.snapshot.actual_incremental_backup_size_in_mega_bytes #=> Float resp.snapshot.backup_progress_in_mega_bytes #=> Float resp.snapshot.current_backup_rate_in_mega_bytes_per_second #=> Float resp.snapshot.estimated_seconds_to_completion #=> Integer resp.snapshot.elapsed_time_in_seconds #=> Integer resp.snapshot.source_region #=> String resp.snapshot.tags #=> Array resp.snapshot.tags[0].key #=> String resp.snapshot.tags[0].value #=> String resp.snapshot.restorable_node_types #=> Array resp.snapshot.restorable_node_types[0] #=> String resp.snapshot.enhanced_vpc_routing #=> Boolean resp.snapshot.maintenance_track_name #=> String resp.snapshot.manual_snapshot_retention_period #=> Integer resp.snapshot.manual_snapshot_remaining_days #=> Integer resp.snapshot.snapshot_retention_start_time #=> Time
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/ModifyClusterSnapshot AWS API Documentation
@overload modify_cluster_snapshot
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 8278 def modify_cluster_snapshot(params = {}, options = {}) req = build_request(:modify_cluster_snapshot, params) req.send_request(options) end
Modifies a snapshot schedule for a cluster.
@option params [required, String] :cluster_identifier
A unique identifier for the cluster whose snapshot schedule you want to modify.
@option params [String] :schedule_identifier
A unique alphanumeric identifier for the schedule that you want to associate with the cluster.
@option params [Boolean] :disassociate_schedule
A boolean to indicate whether to remove the assoiciation between the cluster and the schedule.
@return [Struct] Returns an empty {Seahorse::Client::Response response}.
@example Request syntax with placeholder values
resp = client.modify_cluster_snapshot_schedule({ cluster_identifier: "String", # required schedule_identifier: "String", disassociate_schedule: false, })
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/ModifyClusterSnapshotSchedule AWS API Documentation
@overload modify_cluster_snapshot_schedule
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 8311 def modify_cluster_snapshot_schedule(params = {}, options = {}) req = build_request(:modify_cluster_snapshot_schedule, params) req.send_request(options) end
Modifies a cluster subnet group to include the specified list of VPC subnets. The operation replaces the existing list of subnets with the new list of subnets.
@option params [required, String] :cluster_subnet_group_name
The name of the subnet group to be modified.
@option params [String] :description
A text description of the subnet group to be modified.
@option params [required, Array<String>] :subnet_ids
An array of VPC subnet IDs. A maximum of 20 subnets can be modified in a single request.
@return [Types::ModifyClusterSubnetGroupResult] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::ModifyClusterSubnetGroupResult#cluster_subnet_group #cluster_subnet_group} => Types::ClusterSubnetGroup
@example Request syntax with placeholder values
resp = client.modify_cluster_subnet_group({ cluster_subnet_group_name: "String", # required description: "String", subnet_ids: ["String"], # required })
@example Response structure
resp.cluster_subnet_group.cluster_subnet_group_name #=> String resp.cluster_subnet_group.description #=> String resp.cluster_subnet_group.vpc_id #=> String resp.cluster_subnet_group.subnet_group_status #=> String resp.cluster_subnet_group.subnets #=> Array resp.cluster_subnet_group.subnets[0].subnet_identifier #=> String resp.cluster_subnet_group.subnets[0].subnet_availability_zone.name #=> String resp.cluster_subnet_group.subnets[0].subnet_availability_zone.supported_platforms #=> Array resp.cluster_subnet_group.subnets[0].subnet_availability_zone.supported_platforms[0].name #=> String resp.cluster_subnet_group.subnets[0].subnet_status #=> String resp.cluster_subnet_group.tags #=> Array resp.cluster_subnet_group.tags[0].key #=> String resp.cluster_subnet_group.tags[0].value #=> String
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/ModifyClusterSubnetGroup AWS API Documentation
@overload modify_cluster_subnet_group
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 8362 def modify_cluster_subnet_group(params = {}, options = {}) req = build_request(:modify_cluster_subnet_group, params) req.send_request(options) end
Modifies a Redshift-managed VPC endpoint.
@option params [required, String] :endpoint_name
The endpoint to be modified.
@option params [Array<String>] :vpc_security_group_ids
The complete list of VPC security groups associated with the endpoint after the endpoint is modified.
@return [Types::EndpointAccess] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::EndpointAccess#cluster_identifier #cluster_identifier} => String * {Types::EndpointAccess#resource_owner #resource_owner} => String * {Types::EndpointAccess#subnet_group_name #subnet_group_name} => String * {Types::EndpointAccess#endpoint_status #endpoint_status} => String * {Types::EndpointAccess#endpoint_name #endpoint_name} => String * {Types::EndpointAccess#endpoint_create_time #endpoint_create_time} => Time * {Types::EndpointAccess#port #port} => Integer * {Types::EndpointAccess#address #address} => String * {Types::EndpointAccess#vpc_security_groups #vpc_security_groups} => Array<Types::VpcSecurityGroupMembership> * {Types::EndpointAccess#vpc_endpoint #vpc_endpoint} => Types::VpcEndpoint
@example Request syntax with placeholder values
resp = client.modify_endpoint_access({ endpoint_name: "String", # required vpc_security_group_ids: ["String"], })
@example Response structure
resp.cluster_identifier #=> String resp.resource_owner #=> String resp.subnet_group_name #=> String resp.endpoint_status #=> String resp.endpoint_name #=> String resp.endpoint_create_time #=> Time resp.port #=> Integer resp.address #=> String resp.vpc_security_groups #=> Array resp.vpc_security_groups[0].vpc_security_group_id #=> String resp.vpc_security_groups[0].status #=> String resp.vpc_endpoint.vpc_endpoint_id #=> String resp.vpc_endpoint.vpc_id #=> String resp.vpc_endpoint.network_interfaces #=> Array resp.vpc_endpoint.network_interfaces[0].network_interface_id #=> String resp.vpc_endpoint.network_interfaces[0].subnet_id #=> String resp.vpc_endpoint.network_interfaces[0].private_ip_address #=> String resp.vpc_endpoint.network_interfaces[0].availability_zone #=> String
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/ModifyEndpointAccess AWS API Documentation
@overload modify_endpoint_access
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 8421 def modify_endpoint_access(params = {}, options = {}) req = build_request(:modify_endpoint_access, params) req.send_request(options) end
Modifies an existing Amazon Redshift
event notification subscription.
@option params [required, String] :subscription_name
The name of the modified Amazon Redshift event notification subscription.
@option params [String] :sns_topic_arn
The Amazon Resource Name (ARN) of the SNS topic to be used by the event notification subscription.
@option params [String] :source_type
The type of source that will be generating the events. For example, if you want to be notified of events generated by a cluster, you would set this parameter to cluster. If this value is not specified, events are returned for all Amazon Redshift objects in your Amazon Web Services account. You must specify a source type in order to specify source IDs. Valid values: cluster, cluster-parameter-group, cluster-security-group, cluster-snapshot, and scheduled-action.
@option params [Array<String>] :source_ids
A list of one or more identifiers of Amazon Redshift source objects. All of the objects must be of the same type as was specified in the source type parameter. The event subscription will return only events generated by the specified objects. If not specified, then events are returned for all objects within the source type specified. Example: my-cluster-1, my-cluster-2 Example: my-snapshot-20131010
@option params [Array<String>] :event_categories
Specifies the Amazon Redshift event categories to be published by the event notification subscription. Values: configuration, management, monitoring, security, pending
@option params [String] :severity
Specifies the Amazon Redshift event severity to be published by the event notification subscription. Values: ERROR, INFO
@option params [Boolean] :enabled
A Boolean value indicating if the subscription is enabled. `true` indicates the subscription is enabled
@return [Types::ModifyEventSubscriptionResult] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::ModifyEventSubscriptionResult#event_subscription #event_subscription} => Types::EventSubscription
@example Request syntax with placeholder values
resp = client.modify_event_subscription({ subscription_name: "String", # required sns_topic_arn: "String", source_type: "String", source_ids: ["String"], event_categories: ["String"], severity: "String", enabled: false, })
@example Response structure
resp.event_subscription.customer_aws_id #=> String resp.event_subscription.cust_subscription_id #=> String resp.event_subscription.sns_topic_arn #=> String resp.event_subscription.status #=> String resp.event_subscription.subscription_creation_time #=> Time resp.event_subscription.source_type #=> String resp.event_subscription.source_ids_list #=> Array resp.event_subscription.source_ids_list[0] #=> String resp.event_subscription.event_categories_list #=> Array resp.event_subscription.event_categories_list[0] #=> String resp.event_subscription.severity #=> String resp.event_subscription.enabled #=> Boolean resp.event_subscription.tags #=> Array resp.event_subscription.tags[0].key #=> String resp.event_subscription.tags[0].value #=> String
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/ModifyEventSubscription AWS API Documentation
@overload modify_event_subscription
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 8512 def modify_event_subscription(params = {}, options = {}) req = build_request(:modify_event_subscription, params) req.send_request(options) end
Modifies a scheduled action.
@option params [required, String] :scheduled_action_name
The name of the scheduled action to modify.
@option params [Types::ScheduledActionType] :target_action
A modified JSON format of the scheduled action. For more information about this parameter, see ScheduledAction.
@option params [String] :schedule
A modified schedule in either `at( )` or `cron( )` format. For more information about this parameter, see ScheduledAction.
@option params [String] :iam_role
A different IAM role to assume to run the target action. For more information about this parameter, see ScheduledAction.
@option params [String] :scheduled_action_description
A modified description of the scheduled action.
@option params [Time,DateTime,Date,Integer,String] :start_time
A modified start time of the scheduled action. For more information about this parameter, see ScheduledAction.
@option params [Time,DateTime,Date,Integer,String] :end_time
A modified end time of the scheduled action. For more information about this parameter, see ScheduledAction.
@option params [Boolean] :enable
A modified enable flag of the scheduled action. If true, the scheduled action is active. If false, the scheduled action is disabled.
@return [Types::ScheduledAction] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::ScheduledAction#scheduled_action_name #scheduled_action_name} => String * {Types::ScheduledAction#target_action #target_action} => Types::ScheduledActionType * {Types::ScheduledAction#schedule #schedule} => String * {Types::ScheduledAction#iam_role #iam_role} => String * {Types::ScheduledAction#scheduled_action_description #scheduled_action_description} => String * {Types::ScheduledAction#state #state} => String * {Types::ScheduledAction#next_invocations #next_invocations} => Array<Time> * {Types::ScheduledAction#start_time #start_time} => Time * {Types::ScheduledAction#end_time #end_time} => Time
@example Request syntax with placeholder values
resp = client.modify_scheduled_action({ scheduled_action_name: "String", # required target_action: { resize_cluster: { cluster_identifier: "String", # required cluster_type: "String", node_type: "String", number_of_nodes: 1, classic: false, }, pause_cluster: { cluster_identifier: "String", # required }, resume_cluster: { cluster_identifier: "String", # required }, }, schedule: "String", iam_role: "String", scheduled_action_description: "String", start_time: Time.now, end_time: Time.now, enable: false, })
@example Response structure
resp.scheduled_action_name #=> String resp.target_action.resize_cluster.cluster_identifier #=> String resp.target_action.resize_cluster.cluster_type #=> String resp.target_action.resize_cluster.node_type #=> String resp.target_action.resize_cluster.number_of_nodes #=> Integer resp.target_action.resize_cluster.classic #=> Boolean resp.target_action.pause_cluster.cluster_identifier #=> String resp.target_action.resume_cluster.cluster_identifier #=> String resp.schedule #=> String resp.iam_role #=> String resp.scheduled_action_description #=> String resp.state #=> String, one of "ACTIVE", "DISABLED" resp.next_invocations #=> Array resp.next_invocations[0] #=> Time resp.start_time #=> Time resp.end_time #=> Time
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/ModifyScheduledAction AWS API Documentation
@overload modify_scheduled_action
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 8611 def modify_scheduled_action(params = {}, options = {}) req = build_request(:modify_scheduled_action, params) req.send_request(options) end
Modifies the number of days to retain snapshots in the destination Amazon Web Services Region after they are copied from the source Amazon Web Services Region. By default, this operation only changes the retention period of copied automated snapshots. The retention periods for both new and existing copied automated snapshots are updated with the new retention period. You can set the manual option to change only the retention periods of copied manual snapshots. If you set this option, only newly copied manual snapshots have the new retention period.
@option params [required, String] :cluster_identifier
The unique identifier of the cluster for which you want to change the retention period for either automated or manual snapshots that are copied to a destination Amazon Web Services Region. Constraints: Must be the valid name of an existing cluster that has cross-region snapshot copy enabled.
@option params [required, Integer] :retention_period
The number of days to retain automated snapshots in the destination Amazon Web Services Region after they are copied from the source Amazon Web Services Region. By default, this only changes the retention period of copied automated snapshots. If you decrease the retention period for automated snapshots that are copied to a destination Amazon Web Services Region, Amazon Redshift deletes any existing automated snapshots that were copied to the destination Amazon Web Services Region and that fall outside of the new retention period. Constraints: Must be at least 1 and no more than 35 for automated snapshots. If you specify the `manual` option, only newly copied manual snapshots will have the new retention period. If you specify the value of -1 newly copied manual snapshots are retained indefinitely. Constraints: The number of days must be either -1 or an integer between 1 and 3,653 for manual snapshots.
@option params [Boolean] :manual
Indicates whether to apply the snapshot retention period to newly copied manual snapshots instead of automated snapshots.
@return [Types::ModifySnapshotCopyRetentionPeriodResult] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::ModifySnapshotCopyRetentionPeriodResult#cluster #cluster} => Types::Cluster
@example Request syntax with placeholder values
resp = client.modify_snapshot_copy_retention_period({ cluster_identifier: "String", # required retention_period: 1, # required manual: false, })
@example Response structure
resp.cluster.cluster_identifier #=> String resp.cluster.node_type #=> String resp.cluster.cluster_status #=> String resp.cluster.cluster_availability_status #=> String resp.cluster.modify_status #=> String resp.cluster.master_username #=> String resp.cluster.db_name #=> String resp.cluster.endpoint.address #=> String resp.cluster.endpoint.port #=> Integer resp.cluster.endpoint.vpc_endpoints #=> Array resp.cluster.endpoint.vpc_endpoints[0].vpc_endpoint_id #=> String resp.cluster.endpoint.vpc_endpoints[0].vpc_id #=> String resp.cluster.endpoint.vpc_endpoints[0].network_interfaces #=> Array resp.cluster.endpoint.vpc_endpoints[0].network_interfaces[0].network_interface_id #=> String resp.cluster.endpoint.vpc_endpoints[0].network_interfaces[0].subnet_id #=> String resp.cluster.endpoint.vpc_endpoints[0].network_interfaces[0].private_ip_address #=> String resp.cluster.endpoint.vpc_endpoints[0].network_interfaces[0].availability_zone #=> String resp.cluster.cluster_create_time #=> Time resp.cluster.automated_snapshot_retention_period #=> Integer resp.cluster.manual_snapshot_retention_period #=> Integer resp.cluster.cluster_security_groups #=> Array resp.cluster.cluster_security_groups[0].cluster_security_group_name #=> String resp.cluster.cluster_security_groups[0].status #=> String resp.cluster.vpc_security_groups #=> Array resp.cluster.vpc_security_groups[0].vpc_security_group_id #=> String resp.cluster.vpc_security_groups[0].status #=> String resp.cluster.cluster_parameter_groups #=> Array resp.cluster.cluster_parameter_groups[0].parameter_group_name #=> String resp.cluster.cluster_parameter_groups[0].parameter_apply_status #=> String resp.cluster.cluster_parameter_groups[0].cluster_parameter_status_list #=> Array resp.cluster.cluster_parameter_groups[0].cluster_parameter_status_list[0].parameter_name #=> String resp.cluster.cluster_parameter_groups[0].cluster_parameter_status_list[0].parameter_apply_status #=> String resp.cluster.cluster_parameter_groups[0].cluster_parameter_status_list[0].parameter_apply_error_description #=> String resp.cluster.cluster_subnet_group_name #=> String resp.cluster.vpc_id #=> String resp.cluster.availability_zone #=> String resp.cluster.preferred_maintenance_window #=> String resp.cluster.pending_modified_values.master_user_password #=> String resp.cluster.pending_modified_values.node_type #=> String resp.cluster.pending_modified_values.number_of_nodes #=> Integer resp.cluster.pending_modified_values.cluster_type #=> String resp.cluster.pending_modified_values.cluster_version #=> String resp.cluster.pending_modified_values.automated_snapshot_retention_period #=> Integer resp.cluster.pending_modified_values.cluster_identifier #=> String resp.cluster.pending_modified_values.publicly_accessible #=> Boolean resp.cluster.pending_modified_values.enhanced_vpc_routing #=> Boolean resp.cluster.pending_modified_values.maintenance_track_name #=> String resp.cluster.pending_modified_values.encryption_type #=> String resp.cluster.cluster_version #=> String resp.cluster.allow_version_upgrade #=> Boolean resp.cluster.number_of_nodes #=> Integer resp.cluster.publicly_accessible #=> Boolean resp.cluster.encrypted #=> Boolean resp.cluster.restore_status.status #=> String resp.cluster.restore_status.current_restore_rate_in_mega_bytes_per_second #=> Float resp.cluster.restore_status.snapshot_size_in_mega_bytes #=> Integer resp.cluster.restore_status.progress_in_mega_bytes #=> Integer resp.cluster.restore_status.elapsed_time_in_seconds #=> Integer resp.cluster.restore_status.estimated_time_to_completion_in_seconds #=> Integer resp.cluster.data_transfer_progress.status #=> String resp.cluster.data_transfer_progress.current_rate_in_mega_bytes_per_second #=> Float resp.cluster.data_transfer_progress.total_data_in_mega_bytes #=> Integer resp.cluster.data_transfer_progress.data_transferred_in_mega_bytes #=> Integer resp.cluster.data_transfer_progress.estimated_time_to_completion_in_seconds #=> Integer resp.cluster.data_transfer_progress.elapsed_time_in_seconds #=> Integer resp.cluster.hsm_status.hsm_client_certificate_identifier #=> String resp.cluster.hsm_status.hsm_configuration_identifier #=> String resp.cluster.hsm_status.status #=> String resp.cluster.cluster_snapshot_copy_status.destination_region #=> String resp.cluster.cluster_snapshot_copy_status.retention_period #=> Integer resp.cluster.cluster_snapshot_copy_status.manual_snapshot_retention_period #=> Integer resp.cluster.cluster_snapshot_copy_status.snapshot_copy_grant_name #=> String resp.cluster.cluster_public_key #=> String resp.cluster.cluster_nodes #=> Array resp.cluster.cluster_nodes[0].node_role #=> String resp.cluster.cluster_nodes[0].private_ip_address #=> String resp.cluster.cluster_nodes[0].public_ip_address #=> String resp.cluster.elastic_ip_status.elastic_ip #=> String resp.cluster.elastic_ip_status.status #=> String resp.cluster.cluster_revision_number #=> String resp.cluster.tags #=> Array resp.cluster.tags[0].key #=> String resp.cluster.tags[0].value #=> String resp.cluster.kms_key_id #=> String resp.cluster.enhanced_vpc_routing #=> Boolean resp.cluster.iam_roles #=> Array resp.cluster.iam_roles[0].iam_role_arn #=> String resp.cluster.iam_roles[0].apply_status #=> String resp.cluster.pending_actions #=> Array resp.cluster.pending_actions[0] #=> String resp.cluster.maintenance_track_name #=> String resp.cluster.elastic_resize_number_of_node_options #=> String resp.cluster.deferred_maintenance_windows #=> Array resp.cluster.deferred_maintenance_windows[0].defer_maintenance_identifier #=> String resp.cluster.deferred_maintenance_windows[0].defer_maintenance_start_time #=> Time resp.cluster.deferred_maintenance_windows[0].defer_maintenance_end_time #=> Time resp.cluster.snapshot_schedule_identifier #=> String resp.cluster.snapshot_schedule_state #=> String, one of "MODIFYING", "ACTIVE", "FAILED" resp.cluster.expected_next_snapshot_schedule_time #=> Time resp.cluster.expected_next_snapshot_schedule_time_status #=> String resp.cluster.next_maintenance_window_start_time #=> Time resp.cluster.resize_info.resize_type #=> String resp.cluster.resize_info.allow_cancel_resize #=> Boolean resp.cluster.availability_zone_relocation_status #=> String resp.cluster.cluster_namespace_arn #=> String resp.cluster.total_storage_capacity_in_mega_bytes #=> Integer resp.cluster.aqua_configuration.aqua_status #=> String, one of "enabled", "disabled", "applying" resp.cluster.aqua_configuration.aqua_configuration_status #=> String, one of "enabled", "disabled", "auto"
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/ModifySnapshotCopyRetentionPeriod AWS API Documentation
@overload modify_snapshot_copy_retention_period
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 8791 def modify_snapshot_copy_retention_period(params = {}, options = {}) req = build_request(:modify_snapshot_copy_retention_period, params) req.send_request(options) end
Modifies a snapshot schedule. Any schedule associated with a cluster is modified asynchronously.
@option params [required, String] :schedule_identifier
A unique alphanumeric identifier of the schedule to modify.
@option params [required, Array<String>] :schedule_definitions
An updated list of schedule definitions. A schedule definition is made up of schedule expressions, for example, "cron(30 12 *)" or "rate(12 hours)".
@return [Types::SnapshotSchedule] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::SnapshotSchedule#schedule_definitions #schedule_definitions} => Array<String> * {Types::SnapshotSchedule#schedule_identifier #schedule_identifier} => String * {Types::SnapshotSchedule#schedule_description #schedule_description} => String * {Types::SnapshotSchedule#tags #tags} => Array<Types::Tag> * {Types::SnapshotSchedule#next_invocations #next_invocations} => Array<Time> * {Types::SnapshotSchedule#associated_cluster_count #associated_cluster_count} => Integer * {Types::SnapshotSchedule#associated_clusters #associated_clusters} => Array<Types::ClusterAssociatedToSchedule>
@example Request syntax with placeholder values
resp = client.modify_snapshot_schedule({ schedule_identifier: "String", # required schedule_definitions: ["String"], # required })
@example Response structure
resp.schedule_definitions #=> Array resp.schedule_definitions[0] #=> String resp.schedule_identifier #=> String resp.schedule_description #=> String resp.tags #=> Array resp.tags[0].key #=> String resp.tags[0].value #=> String resp.next_invocations #=> Array resp.next_invocations[0] #=> Time resp.associated_cluster_count #=> Integer resp.associated_clusters #=> Array resp.associated_clusters[0].cluster_identifier #=> String resp.associated_clusters[0].schedule_association_state #=> String, one of "MODIFYING", "ACTIVE", "FAILED"
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/ModifySnapshotSchedule AWS API Documentation
@overload modify_snapshot_schedule
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 8844 def modify_snapshot_schedule(params = {}, options = {}) req = build_request(:modify_snapshot_schedule, params) req.send_request(options) end
Modifies a usage limit in a cluster. You can't modify the feature type or period of a usage limit.
@option params [required, String] :usage_limit_id
The identifier of the usage limit to modify.
@option params [Integer] :amount
The new limit amount. For more information about this parameter, see UsageLimit.
@option params [String] :breach_action
The new action that Amazon Redshift takes when the limit is reached. For more information about this parameter, see UsageLimit.
@return [Types::UsageLimit] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::UsageLimit#usage_limit_id #usage_limit_id} => String * {Types::UsageLimit#cluster_identifier #cluster_identifier} => String * {Types::UsageLimit#feature_type #feature_type} => String * {Types::UsageLimit#limit_type #limit_type} => String * {Types::UsageLimit#amount #amount} => Integer * {Types::UsageLimit#period #period} => String * {Types::UsageLimit#breach_action #breach_action} => String * {Types::UsageLimit#tags #tags} => Array<Types::Tag>
@example Request syntax with placeholder values
resp = client.modify_usage_limit({ usage_limit_id: "String", # required amount: 1, breach_action: "log", # accepts log, emit-metric, disable })
@example Response structure
resp.usage_limit_id #=> String resp.cluster_identifier #=> String resp.feature_type #=> String, one of "spectrum", "concurrency-scaling" resp.limit_type #=> String, one of "time", "data-scanned" resp.amount #=> Integer resp.period #=> String, one of "daily", "weekly", "monthly" resp.breach_action #=> String, one of "log", "emit-metric", "disable" resp.tags #=> Array resp.tags[0].key #=> String resp.tags[0].value #=> String
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/ModifyUsageLimit AWS API Documentation
@overload modify_usage_limit
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 8899 def modify_usage_limit(params = {}, options = {}) req = build_request(:modify_usage_limit, params) req.send_request(options) end
Pauses a cluster.
@option params [required, String] :cluster_identifier
The identifier of the cluster to be paused.
@return [Types::PauseClusterResult] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::PauseClusterResult#cluster #cluster} => Types::Cluster
@example Request syntax with placeholder values
resp = client.pause_cluster({ cluster_identifier: "String", # required })
@example Response structure
resp.cluster.cluster_identifier #=> String resp.cluster.node_type #=> String resp.cluster.cluster_status #=> String resp.cluster.cluster_availability_status #=> String resp.cluster.modify_status #=> String resp.cluster.master_username #=> String resp.cluster.db_name #=> String resp.cluster.endpoint.address #=> String resp.cluster.endpoint.port #=> Integer resp.cluster.endpoint.vpc_endpoints #=> Array resp.cluster.endpoint.vpc_endpoints[0].vpc_endpoint_id #=> String resp.cluster.endpoint.vpc_endpoints[0].vpc_id #=> String resp.cluster.endpoint.vpc_endpoints[0].network_interfaces #=> Array resp.cluster.endpoint.vpc_endpoints[0].network_interfaces[0].network_interface_id #=> String resp.cluster.endpoint.vpc_endpoints[0].network_interfaces[0].subnet_id #=> String resp.cluster.endpoint.vpc_endpoints[0].network_interfaces[0].private_ip_address #=> String resp.cluster.endpoint.vpc_endpoints[0].network_interfaces[0].availability_zone #=> String resp.cluster.cluster_create_time #=> Time resp.cluster.automated_snapshot_retention_period #=> Integer resp.cluster.manual_snapshot_retention_period #=> Integer resp.cluster.cluster_security_groups #=> Array resp.cluster.cluster_security_groups[0].cluster_security_group_name #=> String resp.cluster.cluster_security_groups[0].status #=> String resp.cluster.vpc_security_groups #=> Array resp.cluster.vpc_security_groups[0].vpc_security_group_id #=> String resp.cluster.vpc_security_groups[0].status #=> String resp.cluster.cluster_parameter_groups #=> Array resp.cluster.cluster_parameter_groups[0].parameter_group_name #=> String resp.cluster.cluster_parameter_groups[0].parameter_apply_status #=> String resp.cluster.cluster_parameter_groups[0].cluster_parameter_status_list #=> Array resp.cluster.cluster_parameter_groups[0].cluster_parameter_status_list[0].parameter_name #=> String resp.cluster.cluster_parameter_groups[0].cluster_parameter_status_list[0].parameter_apply_status #=> String resp.cluster.cluster_parameter_groups[0].cluster_parameter_status_list[0].parameter_apply_error_description #=> String resp.cluster.cluster_subnet_group_name #=> String resp.cluster.vpc_id #=> String resp.cluster.availability_zone #=> String resp.cluster.preferred_maintenance_window #=> String resp.cluster.pending_modified_values.master_user_password #=> String resp.cluster.pending_modified_values.node_type #=> String resp.cluster.pending_modified_values.number_of_nodes #=> Integer resp.cluster.pending_modified_values.cluster_type #=> String resp.cluster.pending_modified_values.cluster_version #=> String resp.cluster.pending_modified_values.automated_snapshot_retention_period #=> Integer resp.cluster.pending_modified_values.cluster_identifier #=> String resp.cluster.pending_modified_values.publicly_accessible #=> Boolean resp.cluster.pending_modified_values.enhanced_vpc_routing #=> Boolean resp.cluster.pending_modified_values.maintenance_track_name #=> String resp.cluster.pending_modified_values.encryption_type #=> String resp.cluster.cluster_version #=> String resp.cluster.allow_version_upgrade #=> Boolean resp.cluster.number_of_nodes #=> Integer resp.cluster.publicly_accessible #=> Boolean resp.cluster.encrypted #=> Boolean resp.cluster.restore_status.status #=> String resp.cluster.restore_status.current_restore_rate_in_mega_bytes_per_second #=> Float resp.cluster.restore_status.snapshot_size_in_mega_bytes #=> Integer resp.cluster.restore_status.progress_in_mega_bytes #=> Integer resp.cluster.restore_status.elapsed_time_in_seconds #=> Integer resp.cluster.restore_status.estimated_time_to_completion_in_seconds #=> Integer resp.cluster.data_transfer_progress.status #=> String resp.cluster.data_transfer_progress.current_rate_in_mega_bytes_per_second #=> Float resp.cluster.data_transfer_progress.total_data_in_mega_bytes #=> Integer resp.cluster.data_transfer_progress.data_transferred_in_mega_bytes #=> Integer resp.cluster.data_transfer_progress.estimated_time_to_completion_in_seconds #=> Integer resp.cluster.data_transfer_progress.elapsed_time_in_seconds #=> Integer resp.cluster.hsm_status.hsm_client_certificate_identifier #=> String resp.cluster.hsm_status.hsm_configuration_identifier #=> String resp.cluster.hsm_status.status #=> String resp.cluster.cluster_snapshot_copy_status.destination_region #=> String resp.cluster.cluster_snapshot_copy_status.retention_period #=> Integer resp.cluster.cluster_snapshot_copy_status.manual_snapshot_retention_period #=> Integer resp.cluster.cluster_snapshot_copy_status.snapshot_copy_grant_name #=> String resp.cluster.cluster_public_key #=> String resp.cluster.cluster_nodes #=> Array resp.cluster.cluster_nodes[0].node_role #=> String resp.cluster.cluster_nodes[0].private_ip_address #=> String resp.cluster.cluster_nodes[0].public_ip_address #=> String resp.cluster.elastic_ip_status.elastic_ip #=> String resp.cluster.elastic_ip_status.status #=> String resp.cluster.cluster_revision_number #=> String resp.cluster.tags #=> Array resp.cluster.tags[0].key #=> String resp.cluster.tags[0].value #=> String resp.cluster.kms_key_id #=> String resp.cluster.enhanced_vpc_routing #=> Boolean resp.cluster.iam_roles #=> Array resp.cluster.iam_roles[0].iam_role_arn #=> String resp.cluster.iam_roles[0].apply_status #=> String resp.cluster.pending_actions #=> Array resp.cluster.pending_actions[0] #=> String resp.cluster.maintenance_track_name #=> String resp.cluster.elastic_resize_number_of_node_options #=> String resp.cluster.deferred_maintenance_windows #=> Array resp.cluster.deferred_maintenance_windows[0].defer_maintenance_identifier #=> String resp.cluster.deferred_maintenance_windows[0].defer_maintenance_start_time #=> Time resp.cluster.deferred_maintenance_windows[0].defer_maintenance_end_time #=> Time resp.cluster.snapshot_schedule_identifier #=> String resp.cluster.snapshot_schedule_state #=> String, one of "MODIFYING", "ACTIVE", "FAILED" resp.cluster.expected_next_snapshot_schedule_time #=> Time resp.cluster.expected_next_snapshot_schedule_time_status #=> String resp.cluster.next_maintenance_window_start_time #=> Time resp.cluster.resize_info.resize_type #=> String resp.cluster.resize_info.allow_cancel_resize #=> Boolean resp.cluster.availability_zone_relocation_status #=> String resp.cluster.cluster_namespace_arn #=> String resp.cluster.total_storage_capacity_in_mega_bytes #=> Integer resp.cluster.aqua_configuration.aqua_status #=> String, one of "enabled", "disabled", "applying" resp.cluster.aqua_configuration.aqua_configuration_status #=> String, one of "enabled", "disabled", "auto"
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/PauseCluster AWS API Documentation
@overload pause_cluster
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 9034 def pause_cluster(params = {}, options = {}) req = build_request(:pause_cluster, params) req.send_request(options) end
Allows you to purchase reserved nodes. Amazon Redshift
offers a predefined set of reserved node offerings. You can purchase one or more of the offerings. You can call the DescribeReservedNodeOfferings API to obtain the available reserved node offerings. You can call this API by providing a specific reserved node offering and the number of nodes you want to reserve.
For more information about reserved node offerings, go to [Purchasing Reserved Nodes] in the *Amazon Redshift
Cluster Management Guide*.
[1]: docs.aws.amazon.com/redshift/latest/mgmt/purchase-reserved-node-instance.html
@option params [required, String] :reserved_node_offering_id
The unique identifier of the reserved node offering you want to purchase.
@option params [Integer] :node_count
The number of reserved nodes that you want to purchase. Default: `1`
@return [Types::PurchaseReservedNodeOfferingResult] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::PurchaseReservedNodeOfferingResult#reserved_node #reserved_node} => Types::ReservedNode
@example Request syntax with placeholder values
resp = client.purchase_reserved_node_offering({ reserved_node_offering_id: "String", # required node_count: 1, })
@example Response structure
resp.reserved_node.reserved_node_id #=> String resp.reserved_node.reserved_node_offering_id #=> String resp.reserved_node.node_type #=> String resp.reserved_node.start_time #=> Time resp.reserved_node.duration #=> Integer resp.reserved_node.fixed_price #=> Float resp.reserved_node.usage_price #=> Float resp.reserved_node.currency_code #=> String resp.reserved_node.node_count #=> Integer resp.reserved_node.state #=> String resp.reserved_node.offering_type #=> String resp.reserved_node.recurring_charges #=> Array resp.reserved_node.recurring_charges[0].recurring_charge_amount #=> Float resp.reserved_node.recurring_charges[0].recurring_charge_frequency #=> String resp.reserved_node.reserved_node_offering_type #=> String, one of "Regular", "Upgradable"
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/PurchaseReservedNodeOffering AWS API Documentation
@overload purchase_reserved_node_offering
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 9095 def purchase_reserved_node_offering(params = {}, options = {}) req = build_request(:purchase_reserved_node_offering, params) req.send_request(options) end
Reboots a cluster. This action is taken as soon as possible. It results in a momentary outage to the cluster, during which the cluster status is set to `rebooting`. A cluster event is created when the reboot is completed. Any pending cluster modifications (see ModifyCluster) are applied at this reboot. For more information about managing clusters, go to [Amazon Redshift
Clusters] in the *Amazon Redshift
Cluster Management Guide*.
[1]: docs.aws.amazon.com/redshift/latest/mgmt/working-with-clusters.html
@option params [required, String] :cluster_identifier
The cluster identifier.
@return [Types::RebootClusterResult] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::RebootClusterResult#cluster #cluster} => Types::Cluster
@example Request syntax with placeholder values
resp = client.reboot_cluster({ cluster_identifier: "String", # required })
@example Response structure
resp.cluster.cluster_identifier #=> String resp.cluster.node_type #=> String resp.cluster.cluster_status #=> String resp.cluster.cluster_availability_status #=> String resp.cluster.modify_status #=> String resp.cluster.master_username #=> String resp.cluster.db_name #=> String resp.cluster.endpoint.address #=> String resp.cluster.endpoint.port #=> Integer resp.cluster.endpoint.vpc_endpoints #=> Array resp.cluster.endpoint.vpc_endpoints[0].vpc_endpoint_id #=> String resp.cluster.endpoint.vpc_endpoints[0].vpc_id #=> String resp.cluster.endpoint.vpc_endpoints[0].network_interfaces #=> Array resp.cluster.endpoint.vpc_endpoints[0].network_interfaces[0].network_interface_id #=> String resp.cluster.endpoint.vpc_endpoints[0].network_interfaces[0].subnet_id #=> String resp.cluster.endpoint.vpc_endpoints[0].network_interfaces[0].private_ip_address #=> String resp.cluster.endpoint.vpc_endpoints[0].network_interfaces[0].availability_zone #=> String resp.cluster.cluster_create_time #=> Time resp.cluster.automated_snapshot_retention_period #=> Integer resp.cluster.manual_snapshot_retention_period #=> Integer resp.cluster.cluster_security_groups #=> Array resp.cluster.cluster_security_groups[0].cluster_security_group_name #=> String resp.cluster.cluster_security_groups[0].status #=> String resp.cluster.vpc_security_groups #=> Array resp.cluster.vpc_security_groups[0].vpc_security_group_id #=> String resp.cluster.vpc_security_groups[0].status #=> String resp.cluster.cluster_parameter_groups #=> Array resp.cluster.cluster_parameter_groups[0].parameter_group_name #=> String resp.cluster.cluster_parameter_groups[0].parameter_apply_status #=> String resp.cluster.cluster_parameter_groups[0].cluster_parameter_status_list #=> Array resp.cluster.cluster_parameter_groups[0].cluster_parameter_status_list[0].parameter_name #=> String resp.cluster.cluster_parameter_groups[0].cluster_parameter_status_list[0].parameter_apply_status #=> String resp.cluster.cluster_parameter_groups[0].cluster_parameter_status_list[0].parameter_apply_error_description #=> String resp.cluster.cluster_subnet_group_name #=> String resp.cluster.vpc_id #=> String resp.cluster.availability_zone #=> String resp.cluster.preferred_maintenance_window #=> String resp.cluster.pending_modified_values.master_user_password #=> String resp.cluster.pending_modified_values.node_type #=> String resp.cluster.pending_modified_values.number_of_nodes #=> Integer resp.cluster.pending_modified_values.cluster_type #=> String resp.cluster.pending_modified_values.cluster_version #=> String resp.cluster.pending_modified_values.automated_snapshot_retention_period #=> Integer resp.cluster.pending_modified_values.cluster_identifier #=> String resp.cluster.pending_modified_values.publicly_accessible #=> Boolean resp.cluster.pending_modified_values.enhanced_vpc_routing #=> Boolean resp.cluster.pending_modified_values.maintenance_track_name #=> String resp.cluster.pending_modified_values.encryption_type #=> String resp.cluster.cluster_version #=> String resp.cluster.allow_version_upgrade #=> Boolean resp.cluster.number_of_nodes #=> Integer resp.cluster.publicly_accessible #=> Boolean resp.cluster.encrypted #=> Boolean resp.cluster.restore_status.status #=> String resp.cluster.restore_status.current_restore_rate_in_mega_bytes_per_second #=> Float resp.cluster.restore_status.snapshot_size_in_mega_bytes #=> Integer resp.cluster.restore_status.progress_in_mega_bytes #=> Integer resp.cluster.restore_status.elapsed_time_in_seconds #=> Integer resp.cluster.restore_status.estimated_time_to_completion_in_seconds #=> Integer resp.cluster.data_transfer_progress.status #=> String resp.cluster.data_transfer_progress.current_rate_in_mega_bytes_per_second #=> Float resp.cluster.data_transfer_progress.total_data_in_mega_bytes #=> Integer resp.cluster.data_transfer_progress.data_transferred_in_mega_bytes #=> Integer resp.cluster.data_transfer_progress.estimated_time_to_completion_in_seconds #=> Integer resp.cluster.data_transfer_progress.elapsed_time_in_seconds #=> Integer resp.cluster.hsm_status.hsm_client_certificate_identifier #=> String resp.cluster.hsm_status.hsm_configuration_identifier #=> String resp.cluster.hsm_status.status #=> String resp.cluster.cluster_snapshot_copy_status.destination_region #=> String resp.cluster.cluster_snapshot_copy_status.retention_period #=> Integer resp.cluster.cluster_snapshot_copy_status.manual_snapshot_retention_period #=> Integer resp.cluster.cluster_snapshot_copy_status.snapshot_copy_grant_name #=> String resp.cluster.cluster_public_key #=> String resp.cluster.cluster_nodes #=> Array resp.cluster.cluster_nodes[0].node_role #=> String resp.cluster.cluster_nodes[0].private_ip_address #=> String resp.cluster.cluster_nodes[0].public_ip_address #=> String resp.cluster.elastic_ip_status.elastic_ip #=> String resp.cluster.elastic_ip_status.status #=> String resp.cluster.cluster_revision_number #=> String resp.cluster.tags #=> Array resp.cluster.tags[0].key #=> String resp.cluster.tags[0].value #=> String resp.cluster.kms_key_id #=> String resp.cluster.enhanced_vpc_routing #=> Boolean resp.cluster.iam_roles #=> Array resp.cluster.iam_roles[0].iam_role_arn #=> String resp.cluster.iam_roles[0].apply_status #=> String resp.cluster.pending_actions #=> Array resp.cluster.pending_actions[0] #=> String resp.cluster.maintenance_track_name #=> String resp.cluster.elastic_resize_number_of_node_options #=> String resp.cluster.deferred_maintenance_windows #=> Array resp.cluster.deferred_maintenance_windows[0].defer_maintenance_identifier #=> String resp.cluster.deferred_maintenance_windows[0].defer_maintenance_start_time #=> Time resp.cluster.deferred_maintenance_windows[0].defer_maintenance_end_time #=> Time resp.cluster.snapshot_schedule_identifier #=> String resp.cluster.snapshot_schedule_state #=> String, one of "MODIFYING", "ACTIVE", "FAILED" resp.cluster.expected_next_snapshot_schedule_time #=> Time resp.cluster.expected_next_snapshot_schedule_time_status #=> String resp.cluster.next_maintenance_window_start_time #=> Time resp.cluster.resize_info.resize_type #=> String resp.cluster.resize_info.allow_cancel_resize #=> Boolean resp.cluster.availability_zone_relocation_status #=> String resp.cluster.cluster_namespace_arn #=> String resp.cluster.total_storage_capacity_in_mega_bytes #=> Integer resp.cluster.aqua_configuration.aqua_status #=> String, one of "enabled", "disabled", "applying" resp.cluster.aqua_configuration.aqua_configuration_status #=> String, one of "enabled", "disabled", "auto"
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/RebootCluster AWS API Documentation
@overload reboot_cluster
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 9240 def reboot_cluster(params = {}, options = {}) req = build_request(:reboot_cluster, params) req.send_request(options) end
Sets one or more parameters of the specified parameter group to their default values and sets the source values of the parameters to “engine-default”. To reset the entire parameter group specify the ResetAllParameters parameter. For parameter changes to take effect you must reboot any associated clusters.
@option params [required, String] :parameter_group_name
The name of the cluster parameter group to be reset.
@option params [Boolean] :reset_all_parameters
If `true`, all parameters in the specified parameter group will be reset to their default values. Default: `true`
@option params [Array<Types::Parameter>] :parameters
An array of names of parameters to be reset. If *ResetAllParameters* option is not used, then at least one parameter name must be supplied. Constraints: A maximum of 20 parameters can be reset in a single request.
@return [Types::ClusterParameterGroupNameMessage] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::ClusterParameterGroupNameMessage#parameter_group_name #parameter_group_name} => String * {Types::ClusterParameterGroupNameMessage#parameter_group_status #parameter_group_status} => String
@example Request syntax with placeholder values
resp = client.reset_cluster_parameter_group({ parameter_group_name: "String", # required reset_all_parameters: false, parameters: [ { parameter_name: "String", parameter_value: "String", description: "String", source: "String", data_type: "String", allowed_values: "String", apply_type: "static", # accepts static, dynamic is_modifiable: false, minimum_engine_version: "String", }, ], })
@example Response structure
resp.parameter_group_name #=> String resp.parameter_group_status #=> String
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/ResetClusterParameterGroup AWS API Documentation
@overload reset_cluster_parameter_group
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 9339 def reset_cluster_parameter_group(params = {}, options = {}) req = build_request(:reset_cluster_parameter_group, params) req.send_request(options) end
Changes the size of the cluster. You can change the cluster's type, or change the number or type of nodes. The default behavior is to use the elastic resize method. With an elastic resize, your cluster is available for read and write operations more quickly than with the classic resize method.
Elastic resize operations have the following restrictions:
-
You can only resize clusters of the following types:
-
dc1.large (if your cluster is in a VPC)
-
dc1.8xlarge (if your cluster is in a VPC)
-
dc2.large
-
dc2.8xlarge
-
ds2.xlarge
-
ds2.8xlarge
-
ra3.xlplus
-
ra3.4xlarge
-
ra3.16xlarge
-
-
The type of nodes that you add must match the node type for the cluster.
@option params [required, String] :cluster_identifier
The unique identifier for the cluster to resize.
@option params [String] :cluster_type
The new cluster type for the specified cluster.
@option params [String] :node_type
The new node type for the nodes you are adding. If not specified, the cluster's current node type is used.
@option params [Integer] :number_of_nodes
The new number of nodes for the cluster. If not specified, the cluster's current number of nodes is used.
@option params [Boolean] :classic
A boolean value indicating whether the resize operation is using the classic resize process. If you don't provide this parameter or set the value to `false`, the resize type is elastic.
@return [Types::ResizeClusterResult] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::ResizeClusterResult#cluster #cluster} => Types::Cluster
@example Request syntax with placeholder values
resp = client.resize_cluster({ cluster_identifier: "String", # required cluster_type: "String", node_type: "String", number_of_nodes: 1, classic: false, })
@example Response structure
resp.cluster.cluster_identifier #=> String resp.cluster.node_type #=> String resp.cluster.cluster_status #=> String resp.cluster.cluster_availability_status #=> String resp.cluster.modify_status #=> String resp.cluster.master_username #=> String resp.cluster.db_name #=> String resp.cluster.endpoint.address #=> String resp.cluster.endpoint.port #=> Integer resp.cluster.endpoint.vpc_endpoints #=> Array resp.cluster.endpoint.vpc_endpoints[0].vpc_endpoint_id #=> String resp.cluster.endpoint.vpc_endpoints[0].vpc_id #=> String resp.cluster.endpoint.vpc_endpoints[0].network_interfaces #=> Array resp.cluster.endpoint.vpc_endpoints[0].network_interfaces[0].network_interface_id #=> String resp.cluster.endpoint.vpc_endpoints[0].network_interfaces[0].subnet_id #=> String resp.cluster.endpoint.vpc_endpoints[0].network_interfaces[0].private_ip_address #=> String resp.cluster.endpoint.vpc_endpoints[0].network_interfaces[0].availability_zone #=> String resp.cluster.cluster_create_time #=> Time resp.cluster.automated_snapshot_retention_period #=> Integer resp.cluster.manual_snapshot_retention_period #=> Integer resp.cluster.cluster_security_groups #=> Array resp.cluster.cluster_security_groups[0].cluster_security_group_name #=> String resp.cluster.cluster_security_groups[0].status #=> String resp.cluster.vpc_security_groups #=> Array resp.cluster.vpc_security_groups[0].vpc_security_group_id #=> String resp.cluster.vpc_security_groups[0].status #=> String resp.cluster.cluster_parameter_groups #=> Array resp.cluster.cluster_parameter_groups[0].parameter_group_name #=> String resp.cluster.cluster_parameter_groups[0].parameter_apply_status #=> String resp.cluster.cluster_parameter_groups[0].cluster_parameter_status_list #=> Array resp.cluster.cluster_parameter_groups[0].cluster_parameter_status_list[0].parameter_name #=> String resp.cluster.cluster_parameter_groups[0].cluster_parameter_status_list[0].parameter_apply_status #=> String resp.cluster.cluster_parameter_groups[0].cluster_parameter_status_list[0].parameter_apply_error_description #=> String resp.cluster.cluster_subnet_group_name #=> String resp.cluster.vpc_id #=> String resp.cluster.availability_zone #=> String resp.cluster.preferred_maintenance_window #=> String resp.cluster.pending_modified_values.master_user_password #=> String resp.cluster.pending_modified_values.node_type #=> String resp.cluster.pending_modified_values.number_of_nodes #=> Integer resp.cluster.pending_modified_values.cluster_type #=> String resp.cluster.pending_modified_values.cluster_version #=> String resp.cluster.pending_modified_values.automated_snapshot_retention_period #=> Integer resp.cluster.pending_modified_values.cluster_identifier #=> String resp.cluster.pending_modified_values.publicly_accessible #=> Boolean resp.cluster.pending_modified_values.enhanced_vpc_routing #=> Boolean resp.cluster.pending_modified_values.maintenance_track_name #=> String resp.cluster.pending_modified_values.encryption_type #=> String resp.cluster.cluster_version #=> String resp.cluster.allow_version_upgrade #=> Boolean resp.cluster.number_of_nodes #=> Integer resp.cluster.publicly_accessible #=> Boolean resp.cluster.encrypted #=> Boolean resp.cluster.restore_status.status #=> String resp.cluster.restore_status.current_restore_rate_in_mega_bytes_per_second #=> Float resp.cluster.restore_status.snapshot_size_in_mega_bytes #=> Integer resp.cluster.restore_status.progress_in_mega_bytes #=> Integer resp.cluster.restore_status.elapsed_time_in_seconds #=> Integer resp.cluster.restore_status.estimated_time_to_completion_in_seconds #=> Integer resp.cluster.data_transfer_progress.status #=> String resp.cluster.data_transfer_progress.current_rate_in_mega_bytes_per_second #=> Float resp.cluster.data_transfer_progress.total_data_in_mega_bytes #=> Integer resp.cluster.data_transfer_progress.data_transferred_in_mega_bytes #=> Integer resp.cluster.data_transfer_progress.estimated_time_to_completion_in_seconds #=> Integer resp.cluster.data_transfer_progress.elapsed_time_in_seconds #=> Integer resp.cluster.hsm_status.hsm_client_certificate_identifier #=> String resp.cluster.hsm_status.hsm_configuration_identifier #=> String resp.cluster.hsm_status.status #=> String resp.cluster.cluster_snapshot_copy_status.destination_region #=> String resp.cluster.cluster_snapshot_copy_status.retention_period #=> Integer resp.cluster.cluster_snapshot_copy_status.manual_snapshot_retention_period #=> Integer resp.cluster.cluster_snapshot_copy_status.snapshot_copy_grant_name #=> String resp.cluster.cluster_public_key #=> String resp.cluster.cluster_nodes #=> Array resp.cluster.cluster_nodes[0].node_role #=> String resp.cluster.cluster_nodes[0].private_ip_address #=> String resp.cluster.cluster_nodes[0].public_ip_address #=> String resp.cluster.elastic_ip_status.elastic_ip #=> String resp.cluster.elastic_ip_status.status #=> String resp.cluster.cluster_revision_number #=> String resp.cluster.tags #=> Array resp.cluster.tags[0].key #=> String resp.cluster.tags[0].value #=> String resp.cluster.kms_key_id #=> String resp.cluster.enhanced_vpc_routing #=> Boolean resp.cluster.iam_roles #=> Array resp.cluster.iam_roles[0].iam_role_arn #=> String resp.cluster.iam_roles[0].apply_status #=> String resp.cluster.pending_actions #=> Array resp.cluster.pending_actions[0] #=> String resp.cluster.maintenance_track_name #=> String resp.cluster.elastic_resize_number_of_node_options #=> String resp.cluster.deferred_maintenance_windows #=> Array resp.cluster.deferred_maintenance_windows[0].defer_maintenance_identifier #=> String resp.cluster.deferred_maintenance_windows[0].defer_maintenance_start_time #=> Time resp.cluster.deferred_maintenance_windows[0].defer_maintenance_end_time #=> Time resp.cluster.snapshot_schedule_identifier #=> String resp.cluster.snapshot_schedule_state #=> String, one of "MODIFYING", "ACTIVE", "FAILED" resp.cluster.expected_next_snapshot_schedule_time #=> Time resp.cluster.expected_next_snapshot_schedule_time_status #=> String resp.cluster.next_maintenance_window_start_time #=> Time resp.cluster.resize_info.resize_type #=> String resp.cluster.resize_info.allow_cancel_resize #=> Boolean resp.cluster.availability_zone_relocation_status #=> String resp.cluster.cluster_namespace_arn #=> String resp.cluster.total_storage_capacity_in_mega_bytes #=> Integer resp.cluster.aqua_configuration.aqua_status #=> String, one of "enabled", "disabled", "applying" resp.cluster.aqua_configuration.aqua_configuration_status #=> String, one of "enabled", "disabled", "auto"
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/ResizeCluster AWS API Documentation
@overload resize_cluster
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 9523 def resize_cluster(params = {}, options = {}) req = build_request(:resize_cluster, params) req.send_request(options) end
Creates a new cluster from a snapshot. By default, Amazon Redshift
creates the resulting cluster with the same configuration as the original cluster from which the snapshot was created, except that the new cluster is created with the default cluster security and parameter groups. After Amazon Redshift
creates the cluster, you can use the ModifyCluster API to associate a different security group and different parameter group with the restored cluster. If you are using a DS node type, you can also choose to change to another DS node type of the same size during restore.
If you restore a cluster into a VPC, you must provide a cluster subnet group where you want the cluster restored.
For more information about working with snapshots, go to [Amazon Redshift
Snapshots] in the *Amazon Redshift
Cluster Management Guide*.
[1]: docs.aws.amazon.com/redshift/latest/mgmt/working-with-snapshots.html
@option params [required, String] :cluster_identifier
The identifier of the cluster that will be created from restoring the snapshot. Constraints: * Must contain from 1 to 63 alphanumeric characters or hyphens. * Alphabetic characters must be lowercase. * First character must be a letter. * Cannot end with a hyphen or contain two consecutive hyphens. * Must be unique for all clusters within an Amazon Web Services account.
@option params [required, String] :snapshot_identifier
The name of the snapshot from which to create the new cluster. This parameter isn't case sensitive. Example: `my-snapshot-id`
@option params [String] :snapshot_cluster_identifier
The name of the cluster the source snapshot was created from. This parameter is required if your IAM user has a policy containing a snapshot resource element that specifies anything other than * for the cluster name.
@option params [Integer] :port
The port number on which the cluster accepts connections. Default: The same port as the original cluster. Constraints: Must be between `1115` and `65535`.
@option params [String] :availability_zone
The Amazon EC2 Availability Zone in which to restore the cluster. Default: A random, system-chosen Availability Zone. Example: `us-east-2a`
@option params [Boolean] :allow_version_upgrade
If `true`, major version upgrades can be applied during the maintenance window to the Amazon Redshift engine that is running on the cluster. Default: `true`
@option params [String] :cluster_subnet_group_name
The name of the subnet group where you want to cluster restored. A snapshot of cluster in VPC can be restored only in VPC. Therefore, you must provide subnet group name where you want the cluster restored.
@option params [Boolean] :publicly_accessible
If `true`, the cluster can be accessed from a public network.
@option params [String] :owner_account
The Amazon Web Services account used to create or copy the snapshot. Required if you are restoring a snapshot you do not own, optional if you own the snapshot.
@option params [String] :hsm_client_certificate_identifier
Specifies the name of the HSM client certificate the Amazon Redshift cluster uses to retrieve the data encryption keys stored in an HSM.
@option params [String] :hsm_configuration_identifier
Specifies the name of the HSM configuration that contains the information the Amazon Redshift cluster can use to retrieve and store keys in an HSM.
@option params [String] :elastic_ip
The elastic IP (EIP) address for the cluster.
@option params [String] :cluster_parameter_group_name
The name of the parameter group to be associated with this cluster. Default: The default Amazon Redshift cluster parameter group. For information about the default parameter group, go to [Working with Amazon Redshift Parameter Groups][1]. Constraints: * Must be 1 to 255 alphanumeric characters or hyphens. * First character must be a letter. * Cannot end with a hyphen or contain two consecutive hyphens. [1]: https://docs.aws.amazon.com/redshift/latest/mgmt/working-with-parameter-groups.html
@option params [Array<String>] :cluster_security_groups
A list of security groups to be associated with this cluster. Default: The default cluster security group for Amazon Redshift. Cluster security groups only apply to clusters outside of VPCs.
@option params [Array<String>] :vpc_security_group_ids
A list of Virtual Private Cloud (VPC) security groups to be associated with the cluster. Default: The default VPC security group is associated with the cluster. VPC security groups only apply to clusters in VPCs.
@option params [String] :preferred_maintenance_window
The weekly time range (in UTC) during which automated cluster maintenance can occur. Format: `ddd:hh24:mi-ddd:hh24:mi` Default: The value selected for the cluster from which the snapshot was taken. For more information about the time blocks for each region, see [Maintenance Windows][1] in Amazon Redshift Cluster Management Guide. Valid Days: Mon \| Tue \| Wed \| Thu \| Fri \| Sat \| Sun Constraints: Minimum 30-minute window. [1]: https://docs.aws.amazon.com/redshift/latest/mgmt/working-with-clusters.html#rs-maintenance-windows
@option params [Integer] :automated_snapshot_retention_period
The number of days that automated snapshots are retained. If the value is 0, automated snapshots are disabled. Even if automated snapshots are disabled, you can still create manual snapshots when you want with CreateClusterSnapshot. You can't disable automated snapshots for RA3 node types. Set the automated retention period from 1-35 days. Default: The value selected for the cluster from which the snapshot was taken. Constraints: Must be a value from 0 to 35.
@option params [Integer] :manual_snapshot_retention_period
The default number of days to retain a manual snapshot. If the value is -1, the snapshot is retained indefinitely. This setting doesn't change the retention period of existing snapshots. The value must be either -1 or an integer between 1 and 3,653.
@option params [String] :kms_key_id
The Key Management Service (KMS) key ID of the encryption key that you want to use to encrypt data in the cluster that you restore from a shared snapshot.
@option params [String] :node_type
The node type that the restored cluster will be provisioned with. Default: The node type of the cluster from which the snapshot was taken. You can modify this if you are using any DS node type. In that case, you can choose to restore into another DS node type of the same size. For example, you can restore ds1.8xlarge into ds2.8xlarge, or ds1.xlarge into ds2.xlarge. If you have a DC instance type, you must restore into that same instance type and size. In other words, you can only restore a dc1.large instance type into another dc1.large instance type or dc2.large instance type. You can't restore dc1.8xlarge to dc2.8xlarge. First restore to a dc1.8xlarge cluster, then resize to a dc2.8large cluster. For more information about node types, see [ About Clusters and Nodes][1] in the *Amazon Redshift Cluster Management Guide*. [1]: https://docs.aws.amazon.com/redshift/latest/mgmt/working-with-clusters.html#rs-about-clusters-and-nodes
@option params [Boolean] :enhanced_vpc_routing
An option that specifies whether to create the cluster with enhanced VPC routing enabled. To create a cluster that uses enhanced VPC routing, the cluster must be in a VPC. For more information, see [Enhanced VPC Routing][1] in the Amazon Redshift Cluster Management Guide. If this option is `true`, enhanced VPC routing is enabled. Default: false [1]: https://docs.aws.amazon.com/redshift/latest/mgmt/enhanced-vpc-routing.html
@option params [String] :additional_info
Reserved.
@option params [Array<String>] :iam_roles
A list of Identity and Access Management (IAM) roles that can be used by the cluster to access other Amazon Web Services services. You must supply the IAM roles in their Amazon Resource Name (ARN) format. You can supply up to 10 IAM roles in a single request. A cluster can have up to 10 IAM roles associated at any time.
@option params [String] :maintenance_track_name
The name of the maintenance track for the restored cluster. When you take a snapshot, the snapshot inherits the `MaintenanceTrack` value from the cluster. The snapshot might be on a different track than the cluster that was the source for the snapshot. For example, suppose that you take a snapshot of a cluster that is on the current track and then change the cluster to be on the trailing track. In this case, the snapshot and the source cluster are on different tracks.
@option params [String] :snapshot_schedule_identifier
A unique identifier for the snapshot schedule.
@option params [Integer] :number_of_nodes
The number of nodes specified when provisioning the restored cluster.
@option params [Boolean] :availability_zone_relocation
The option to enable relocation for an Amazon Redshift cluster between Availability Zones after the cluster is restored.
@option params [String] :aqua_configuration_status
The value represents how the cluster is configured to use AQUA (Advanced Query Accelerator) after the cluster is restored. Possible values include the following. * enabled - Use AQUA if it is available for the current Amazon Web Services Region and Amazon Redshift node type. * disabled - Don't use AQUA. * auto - Amazon Redshift determines whether to use AQUA.
@return [Types::RestoreFromClusterSnapshotResult] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::RestoreFromClusterSnapshotResult#cluster #cluster} => Types::Cluster
@example Request syntax with placeholder values
resp = client.restore_from_cluster_snapshot({ cluster_identifier: "String", # required snapshot_identifier: "String", # required snapshot_cluster_identifier: "String", port: 1, availability_zone: "String", allow_version_upgrade: false, cluster_subnet_group_name: "String", publicly_accessible: false, owner_account: "String", hsm_client_certificate_identifier: "String", hsm_configuration_identifier: "String", elastic_ip: "String", cluster_parameter_group_name: "String", cluster_security_groups: ["String"], vpc_security_group_ids: ["String"], preferred_maintenance_window: "String", automated_snapshot_retention_period: 1, manual_snapshot_retention_period: 1, kms_key_id: "String", node_type: "String", enhanced_vpc_routing: false, additional_info: "String", iam_roles: ["String"], maintenance_track_name: "String", snapshot_schedule_identifier: "String", number_of_nodes: 1, availability_zone_relocation: false, aqua_configuration_status: "enabled", # accepts enabled, disabled, auto })
@example Response structure
resp.cluster.cluster_identifier #=> String resp.cluster.node_type #=> String resp.cluster.cluster_status #=> String resp.cluster.cluster_availability_status #=> String resp.cluster.modify_status #=> String resp.cluster.master_username #=> String resp.cluster.db_name #=> String resp.cluster.endpoint.address #=> String resp.cluster.endpoint.port #=> Integer resp.cluster.endpoint.vpc_endpoints #=> Array resp.cluster.endpoint.vpc_endpoints[0].vpc_endpoint_id #=> String resp.cluster.endpoint.vpc_endpoints[0].vpc_id #=> String resp.cluster.endpoint.vpc_endpoints[0].network_interfaces #=> Array resp.cluster.endpoint.vpc_endpoints[0].network_interfaces[0].network_interface_id #=> String resp.cluster.endpoint.vpc_endpoints[0].network_interfaces[0].subnet_id #=> String resp.cluster.endpoint.vpc_endpoints[0].network_interfaces[0].private_ip_address #=> String resp.cluster.endpoint.vpc_endpoints[0].network_interfaces[0].availability_zone #=> String resp.cluster.cluster_create_time #=> Time resp.cluster.automated_snapshot_retention_period #=> Integer resp.cluster.manual_snapshot_retention_period #=> Integer resp.cluster.cluster_security_groups #=> Array resp.cluster.cluster_security_groups[0].cluster_security_group_name #=> String resp.cluster.cluster_security_groups[0].status #=> String resp.cluster.vpc_security_groups #=> Array resp.cluster.vpc_security_groups[0].vpc_security_group_id #=> String resp.cluster.vpc_security_groups[0].status #=> String resp.cluster.cluster_parameter_groups #=> Array resp.cluster.cluster_parameter_groups[0].parameter_group_name #=> String resp.cluster.cluster_parameter_groups[0].parameter_apply_status #=> String resp.cluster.cluster_parameter_groups[0].cluster_parameter_status_list #=> Array resp.cluster.cluster_parameter_groups[0].cluster_parameter_status_list[0].parameter_name #=> String resp.cluster.cluster_parameter_groups[0].cluster_parameter_status_list[0].parameter_apply_status #=> String resp.cluster.cluster_parameter_groups[0].cluster_parameter_status_list[0].parameter_apply_error_description #=> String resp.cluster.cluster_subnet_group_name #=> String resp.cluster.vpc_id #=> String resp.cluster.availability_zone #=> String resp.cluster.preferred_maintenance_window #=> String resp.cluster.pending_modified_values.master_user_password #=> String resp.cluster.pending_modified_values.node_type #=> String resp.cluster.pending_modified_values.number_of_nodes #=> Integer resp.cluster.pending_modified_values.cluster_type #=> String resp.cluster.pending_modified_values.cluster_version #=> String resp.cluster.pending_modified_values.automated_snapshot_retention_period #=> Integer resp.cluster.pending_modified_values.cluster_identifier #=> String resp.cluster.pending_modified_values.publicly_accessible #=> Boolean resp.cluster.pending_modified_values.enhanced_vpc_routing #=> Boolean resp.cluster.pending_modified_values.maintenance_track_name #=> String resp.cluster.pending_modified_values.encryption_type #=> String resp.cluster.cluster_version #=> String resp.cluster.allow_version_upgrade #=> Boolean resp.cluster.number_of_nodes #=> Integer resp.cluster.publicly_accessible #=> Boolean resp.cluster.encrypted #=> Boolean resp.cluster.restore_status.status #=> String resp.cluster.restore_status.current_restore_rate_in_mega_bytes_per_second #=> Float resp.cluster.restore_status.snapshot_size_in_mega_bytes #=> Integer resp.cluster.restore_status.progress_in_mega_bytes #=> Integer resp.cluster.restore_status.elapsed_time_in_seconds #=> Integer resp.cluster.restore_status.estimated_time_to_completion_in_seconds #=> Integer resp.cluster.data_transfer_progress.status #=> String resp.cluster.data_transfer_progress.current_rate_in_mega_bytes_per_second #=> Float resp.cluster.data_transfer_progress.total_data_in_mega_bytes #=> Integer resp.cluster.data_transfer_progress.data_transferred_in_mega_bytes #=> Integer resp.cluster.data_transfer_progress.estimated_time_to_completion_in_seconds #=> Integer resp.cluster.data_transfer_progress.elapsed_time_in_seconds #=> Integer resp.cluster.hsm_status.hsm_client_certificate_identifier #=> String resp.cluster.hsm_status.hsm_configuration_identifier #=> String resp.cluster.hsm_status.status #=> String resp.cluster.cluster_snapshot_copy_status.destination_region #=> String resp.cluster.cluster_snapshot_copy_status.retention_period #=> Integer resp.cluster.cluster_snapshot_copy_status.manual_snapshot_retention_period #=> Integer resp.cluster.cluster_snapshot_copy_status.snapshot_copy_grant_name #=> String resp.cluster.cluster_public_key #=> String resp.cluster.cluster_nodes #=> Array resp.cluster.cluster_nodes[0].node_role #=> String resp.cluster.cluster_nodes[0].private_ip_address #=> String resp.cluster.cluster_nodes[0].public_ip_address #=> String resp.cluster.elastic_ip_status.elastic_ip #=> String resp.cluster.elastic_ip_status.status #=> String resp.cluster.cluster_revision_number #=> String resp.cluster.tags #=> Array resp.cluster.tags[0].key #=> String resp.cluster.tags[0].value #=> String resp.cluster.kms_key_id #=> String resp.cluster.enhanced_vpc_routing #=> Boolean resp.cluster.iam_roles #=> Array resp.cluster.iam_roles[0].iam_role_arn #=> String resp.cluster.iam_roles[0].apply_status #=> String resp.cluster.pending_actions #=> Array resp.cluster.pending_actions[0] #=> String resp.cluster.maintenance_track_name #=> String resp.cluster.elastic_resize_number_of_node_options #=> String resp.cluster.deferred_maintenance_windows #=> Array resp.cluster.deferred_maintenance_windows[0].defer_maintenance_identifier #=> String resp.cluster.deferred_maintenance_windows[0].defer_maintenance_start_time #=> Time resp.cluster.deferred_maintenance_windows[0].defer_maintenance_end_time #=> Time resp.cluster.snapshot_schedule_identifier #=> String resp.cluster.snapshot_schedule_state #=> String, one of "MODIFYING", "ACTIVE", "FAILED" resp.cluster.expected_next_snapshot_schedule_time #=> Time resp.cluster.expected_next_snapshot_schedule_time_status #=> String resp.cluster.next_maintenance_window_start_time #=> Time resp.cluster.resize_info.resize_type #=> String resp.cluster.resize_info.allow_cancel_resize #=> Boolean resp.cluster.availability_zone_relocation_status #=> String resp.cluster.cluster_namespace_arn #=> String resp.cluster.total_storage_capacity_in_mega_bytes #=> Integer resp.cluster.aqua_configuration.aqua_status #=> String, one of "enabled", "disabled", "applying" resp.cluster.aqua_configuration.aqua_configuration_status #=> String, one of "enabled", "disabled", "auto"
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/RestoreFromClusterSnapshot AWS API Documentation
@overload restore_from_cluster_snapshot
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 9935 def restore_from_cluster_snapshot(params = {}, options = {}) req = build_request(:restore_from_cluster_snapshot, params) req.send_request(options) end
Creates a new table from a table in an Amazon Redshift
cluster snapshot. You must create the new table within the Amazon Redshift
cluster that the snapshot was taken from.
You cannot use `RestoreTableFromClusterSnapshot` to restore a table with the same name as an existing table in an Amazon Redshift
cluster. That is, you cannot overwrite an existing table in a cluster with a restored table. If you want to replace your original table with a new, restored table, then rename or drop your original table before you call `RestoreTableFromClusterSnapshot`. When you have renamed your original table, then you can pass the original name of the table as the `NewTableName` parameter value in the call to `RestoreTableFromClusterSnapshot`. This way, you can replace the original table with the table created from the snapshot.
@option params [required, String] :cluster_identifier
The identifier of the Amazon Redshift cluster to restore the table to.
@option params [required, String] :snapshot_identifier
The identifier of the snapshot to restore the table from. This snapshot must have been created from the Amazon Redshift cluster specified by the `ClusterIdentifier` parameter.
@option params [required, String] :source_database_name
The name of the source database that contains the table to restore from.
@option params [String] :source_schema_name
The name of the source schema that contains the table to restore from. If you do not specify a `SourceSchemaName` value, the default is `public`.
@option params [required, String] :source_table_name
The name of the source table to restore from.
@option params [String] :target_database_name
The name of the database to restore the table to.
@option params [String] :target_schema_name
The name of the schema to restore the table to.
@option params [required, String] :new_table_name
The name of the table to create as a result of the current request.
@option params [Boolean] :enable_case_sensitive_identifier
Indicates whether name identifiers for database, schema, and table are case sensitive. If `true`, the names are case sensitive. If `false` (default), the names are not case sensitive.
@return [Types::RestoreTableFromClusterSnapshotResult] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::RestoreTableFromClusterSnapshotResult#table_restore_status #table_restore_status} => Types::TableRestoreStatus
@example Request syntax with placeholder values
resp = client.restore_table_from_cluster_snapshot({ cluster_identifier: "String", # required snapshot_identifier: "String", # required source_database_name: "String", # required source_schema_name: "String", source_table_name: "String", # required target_database_name: "String", target_schema_name: "String", new_table_name: "String", # required enable_case_sensitive_identifier: false, })
@example Response structure
resp.table_restore_status.table_restore_request_id #=> String resp.table_restore_status.status #=> String, one of "PENDING", "IN_PROGRESS", "SUCCEEDED", "FAILED", "CANCELED" resp.table_restore_status.message #=> String resp.table_restore_status.request_time #=> Time resp.table_restore_status.progress_in_mega_bytes #=> Integer resp.table_restore_status.total_data_in_mega_bytes #=> Integer resp.table_restore_status.cluster_identifier #=> String resp.table_restore_status.snapshot_identifier #=> String resp.table_restore_status.source_database_name #=> String resp.table_restore_status.source_schema_name #=> String resp.table_restore_status.source_table_name #=> String resp.table_restore_status.target_database_name #=> String resp.table_restore_status.target_schema_name #=> String resp.table_restore_status.new_table_name #=> String
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/RestoreTableFromClusterSnapshot AWS API Documentation
@overload restore_table_from_cluster_snapshot
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 10028 def restore_table_from_cluster_snapshot(params = {}, options = {}) req = build_request(:restore_table_from_cluster_snapshot, params) req.send_request(options) end
Resumes a paused cluster.
@option params [required, String] :cluster_identifier
The identifier of the cluster to be resumed.
@return [Types::ResumeClusterResult] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::ResumeClusterResult#cluster #cluster} => Types::Cluster
@example Request syntax with placeholder values
resp = client.resume_cluster({ cluster_identifier: "String", # required })
@example Response structure
resp.cluster.cluster_identifier #=> String resp.cluster.node_type #=> String resp.cluster.cluster_status #=> String resp.cluster.cluster_availability_status #=> String resp.cluster.modify_status #=> String resp.cluster.master_username #=> String resp.cluster.db_name #=> String resp.cluster.endpoint.address #=> String resp.cluster.endpoint.port #=> Integer resp.cluster.endpoint.vpc_endpoints #=> Array resp.cluster.endpoint.vpc_endpoints[0].vpc_endpoint_id #=> String resp.cluster.endpoint.vpc_endpoints[0].vpc_id #=> String resp.cluster.endpoint.vpc_endpoints[0].network_interfaces #=> Array resp.cluster.endpoint.vpc_endpoints[0].network_interfaces[0].network_interface_id #=> String resp.cluster.endpoint.vpc_endpoints[0].network_interfaces[0].subnet_id #=> String resp.cluster.endpoint.vpc_endpoints[0].network_interfaces[0].private_ip_address #=> String resp.cluster.endpoint.vpc_endpoints[0].network_interfaces[0].availability_zone #=> String resp.cluster.cluster_create_time #=> Time resp.cluster.automated_snapshot_retention_period #=> Integer resp.cluster.manual_snapshot_retention_period #=> Integer resp.cluster.cluster_security_groups #=> Array resp.cluster.cluster_security_groups[0].cluster_security_group_name #=> String resp.cluster.cluster_security_groups[0].status #=> String resp.cluster.vpc_security_groups #=> Array resp.cluster.vpc_security_groups[0].vpc_security_group_id #=> String resp.cluster.vpc_security_groups[0].status #=> String resp.cluster.cluster_parameter_groups #=> Array resp.cluster.cluster_parameter_groups[0].parameter_group_name #=> String resp.cluster.cluster_parameter_groups[0].parameter_apply_status #=> String resp.cluster.cluster_parameter_groups[0].cluster_parameter_status_list #=> Array resp.cluster.cluster_parameter_groups[0].cluster_parameter_status_list[0].parameter_name #=> String resp.cluster.cluster_parameter_groups[0].cluster_parameter_status_list[0].parameter_apply_status #=> String resp.cluster.cluster_parameter_groups[0].cluster_parameter_status_list[0].parameter_apply_error_description #=> String resp.cluster.cluster_subnet_group_name #=> String resp.cluster.vpc_id #=> String resp.cluster.availability_zone #=> String resp.cluster.preferred_maintenance_window #=> String resp.cluster.pending_modified_values.master_user_password #=> String resp.cluster.pending_modified_values.node_type #=> String resp.cluster.pending_modified_values.number_of_nodes #=> Integer resp.cluster.pending_modified_values.cluster_type #=> String resp.cluster.pending_modified_values.cluster_version #=> String resp.cluster.pending_modified_values.automated_snapshot_retention_period #=> Integer resp.cluster.pending_modified_values.cluster_identifier #=> String resp.cluster.pending_modified_values.publicly_accessible #=> Boolean resp.cluster.pending_modified_values.enhanced_vpc_routing #=> Boolean resp.cluster.pending_modified_values.maintenance_track_name #=> String resp.cluster.pending_modified_values.encryption_type #=> String resp.cluster.cluster_version #=> String resp.cluster.allow_version_upgrade #=> Boolean resp.cluster.number_of_nodes #=> Integer resp.cluster.publicly_accessible #=> Boolean resp.cluster.encrypted #=> Boolean resp.cluster.restore_status.status #=> String resp.cluster.restore_status.current_restore_rate_in_mega_bytes_per_second #=> Float resp.cluster.restore_status.snapshot_size_in_mega_bytes #=> Integer resp.cluster.restore_status.progress_in_mega_bytes #=> Integer resp.cluster.restore_status.elapsed_time_in_seconds #=> Integer resp.cluster.restore_status.estimated_time_to_completion_in_seconds #=> Integer resp.cluster.data_transfer_progress.status #=> String resp.cluster.data_transfer_progress.current_rate_in_mega_bytes_per_second #=> Float resp.cluster.data_transfer_progress.total_data_in_mega_bytes #=> Integer resp.cluster.data_transfer_progress.data_transferred_in_mega_bytes #=> Integer resp.cluster.data_transfer_progress.estimated_time_to_completion_in_seconds #=> Integer resp.cluster.data_transfer_progress.elapsed_time_in_seconds #=> Integer resp.cluster.hsm_status.hsm_client_certificate_identifier #=> String resp.cluster.hsm_status.hsm_configuration_identifier #=> String resp.cluster.hsm_status.status #=> String resp.cluster.cluster_snapshot_copy_status.destination_region #=> String resp.cluster.cluster_snapshot_copy_status.retention_period #=> Integer resp.cluster.cluster_snapshot_copy_status.manual_snapshot_retention_period #=> Integer resp.cluster.cluster_snapshot_copy_status.snapshot_copy_grant_name #=> String resp.cluster.cluster_public_key #=> String resp.cluster.cluster_nodes #=> Array resp.cluster.cluster_nodes[0].node_role #=> String resp.cluster.cluster_nodes[0].private_ip_address #=> String resp.cluster.cluster_nodes[0].public_ip_address #=> String resp.cluster.elastic_ip_status.elastic_ip #=> String resp.cluster.elastic_ip_status.status #=> String resp.cluster.cluster_revision_number #=> String resp.cluster.tags #=> Array resp.cluster.tags[0].key #=> String resp.cluster.tags[0].value #=> String resp.cluster.kms_key_id #=> String resp.cluster.enhanced_vpc_routing #=> Boolean resp.cluster.iam_roles #=> Array resp.cluster.iam_roles[0].iam_role_arn #=> String resp.cluster.iam_roles[0].apply_status #=> String resp.cluster.pending_actions #=> Array resp.cluster.pending_actions[0] #=> String resp.cluster.maintenance_track_name #=> String resp.cluster.elastic_resize_number_of_node_options #=> String resp.cluster.deferred_maintenance_windows #=> Array resp.cluster.deferred_maintenance_windows[0].defer_maintenance_identifier #=> String resp.cluster.deferred_maintenance_windows[0].defer_maintenance_start_time #=> Time resp.cluster.deferred_maintenance_windows[0].defer_maintenance_end_time #=> Time resp.cluster.snapshot_schedule_identifier #=> String resp.cluster.snapshot_schedule_state #=> String, one of "MODIFYING", "ACTIVE", "FAILED" resp.cluster.expected_next_snapshot_schedule_time #=> Time resp.cluster.expected_next_snapshot_schedule_time_status #=> String resp.cluster.next_maintenance_window_start_time #=> Time resp.cluster.resize_info.resize_type #=> String resp.cluster.resize_info.allow_cancel_resize #=> Boolean resp.cluster.availability_zone_relocation_status #=> String resp.cluster.cluster_namespace_arn #=> String resp.cluster.total_storage_capacity_in_mega_bytes #=> Integer resp.cluster.aqua_configuration.aqua_status #=> String, one of "enabled", "disabled", "applying" resp.cluster.aqua_configuration.aqua_configuration_status #=> String, one of "enabled", "disabled", "auto"
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/ResumeCluster AWS API Documentation
@overload resume_cluster
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 10163 def resume_cluster(params = {}, options = {}) req = build_request(:resume_cluster, params) req.send_request(options) end
Revokes an ingress rule in an Amazon Redshift
security group for a previously authorized IP range or Amazon EC2 security group. To add an ingress rule, see AuthorizeClusterSecurityGroupIngress. For information about managing security groups, go to [Amazon Redshift
Cluster Security Groups] in the *Amazon Redshift
Cluster Management Guide*.
[1]: docs.aws.amazon.com/redshift/latest/mgmt/working-with-security-groups.html
@option params [required, String] :cluster_security_group_name
The name of the security Group from which to revoke the ingress rule.
@option params [String] :cidrip
The IP range for which to revoke access. This range must be a valid Classless Inter-Domain Routing (CIDR) block of IP addresses. If `CIDRIP` is specified, `EC2SecurityGroupName` and `EC2SecurityGroupOwnerId` cannot be provided.
@option params [String] :ec2_security_group_name
The name of the EC2 Security Group whose access is to be revoked. If `EC2SecurityGroupName` is specified, `EC2SecurityGroupOwnerId` must also be provided and `CIDRIP` cannot be provided.
@option params [String] :ec2_security_group_owner_id
The Amazon Web Services account number of the owner of the security group specified in the `EC2SecurityGroupName` parameter. The Amazon Web Services access key ID is not an acceptable value. If `EC2SecurityGroupOwnerId` is specified, `EC2SecurityGroupName` must also be provided. and `CIDRIP` cannot be provided. Example: `111122223333`
@return [Types::RevokeClusterSecurityGroupIngressResult] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::RevokeClusterSecurityGroupIngressResult#cluster_security_group #cluster_security_group} => Types::ClusterSecurityGroup
@example Request syntax with placeholder values
resp = client.revoke_cluster_security_group_ingress({ cluster_security_group_name: "String", # required cidrip: "String", ec2_security_group_name: "String", ec2_security_group_owner_id: "String", })
@example Response structure
resp.cluster_security_group.cluster_security_group_name #=> String resp.cluster_security_group.description #=> String resp.cluster_security_group.ec2_security_groups #=> Array resp.cluster_security_group.ec2_security_groups[0].status #=> String resp.cluster_security_group.ec2_security_groups[0].ec2_security_group_name #=> String resp.cluster_security_group.ec2_security_groups[0].ec2_security_group_owner_id #=> String resp.cluster_security_group.ec2_security_groups[0].tags #=> Array resp.cluster_security_group.ec2_security_groups[0].tags[0].key #=> String resp.cluster_security_group.ec2_security_groups[0].tags[0].value #=> String resp.cluster_security_group.ip_ranges #=> Array resp.cluster_security_group.ip_ranges[0].status #=> String resp.cluster_security_group.ip_ranges[0].cidrip #=> String resp.cluster_security_group.ip_ranges[0].tags #=> Array resp.cluster_security_group.ip_ranges[0].tags[0].key #=> String resp.cluster_security_group.ip_ranges[0].tags[0].value #=> String resp.cluster_security_group.tags #=> Array resp.cluster_security_group.tags[0].key #=> String resp.cluster_security_group.tags[0].value #=> String
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/RevokeClusterSecurityGroupIngress AWS API Documentation
@overload revoke_cluster_security_group_ingress
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 10240 def revoke_cluster_security_group_ingress(params = {}, options = {}) req = build_request(:revoke_cluster_security_group_ingress, params) req.send_request(options) end
Revokes access to a cluster.
@option params [String] :cluster_identifier
The cluster to revoke access from.
@option params [String] :account
The Amazon Web Services account ID whose access is to be revoked.
@option params [Array<String>] :vpc_ids
The virtual private cloud (VPC) identifiers for which access is to be revoked.
@option params [Boolean] :force
Indicates whether to force the revoke action. If true, the Redshift-managed VPC endpoints associated with the endpoint authorization are also deleted.
@return [Types::EndpointAuthorization] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::EndpointAuthorization#grantor #grantor} => String * {Types::EndpointAuthorization#grantee #grantee} => String * {Types::EndpointAuthorization#cluster_identifier #cluster_identifier} => String * {Types::EndpointAuthorization#authorize_time #authorize_time} => Time * {Types::EndpointAuthorization#cluster_status #cluster_status} => String * {Types::EndpointAuthorization#status #status} => String * {Types::EndpointAuthorization#allowed_all_vp_cs #allowed_all_vp_cs} => Boolean * {Types::EndpointAuthorization#allowed_vp_cs #allowed_vp_cs} => Array<String> * {Types::EndpointAuthorization#endpoint_count #endpoint_count} => Integer
@example Request syntax with placeholder values
resp = client.revoke_endpoint_access({ cluster_identifier: "String", account: "String", vpc_ids: ["String"], force: false, })
@example Response structure
resp.grantor #=> String resp.grantee #=> String resp.cluster_identifier #=> String resp.authorize_time #=> Time resp.cluster_status #=> String resp.status #=> String, one of "Authorized", "Revoking" resp.allowed_all_vp_cs #=> Boolean resp.allowed_vp_cs #=> Array resp.allowed_vp_cs[0] #=> String resp.endpoint_count #=> Integer
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/RevokeEndpointAccess AWS API Documentation
@overload revoke_endpoint_access
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 10300 def revoke_endpoint_access(params = {}, options = {}) req = build_request(:revoke_endpoint_access, params) req.send_request(options) end
Removes the ability of the specified Amazon Web Services account to restore the specified snapshot. If the account is currently restoring the snapshot, the restore will run to completion.
For more information about working with snapshots, go to [Amazon Redshift
Snapshots] in the *Amazon Redshift
Cluster Management Guide*.
[1]: docs.aws.amazon.com/redshift/latest/mgmt/working-with-snapshots.html
@option params [required, String] :snapshot_identifier
The identifier of the snapshot that the account can no longer access.
@option params [String] :snapshot_cluster_identifier
The identifier of the cluster the snapshot was created from. This parameter is required if your IAM user has a policy containing a snapshot resource element that specifies anything other than * for the cluster name.
@option params [required, String] :account_with_restore_access
The identifier of the Amazon Web Services account that can no longer restore the specified snapshot.
@return [Types::RevokeSnapshotAccessResult] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::RevokeSnapshotAccessResult#snapshot #snapshot} => Types::Snapshot
@example Request syntax with placeholder values
resp = client.revoke_snapshot_access({ snapshot_identifier: "String", # required snapshot_cluster_identifier: "String", account_with_restore_access: "String", # required })
@example Response structure
resp.snapshot.snapshot_identifier #=> String resp.snapshot.cluster_identifier #=> String resp.snapshot.snapshot_create_time #=> Time resp.snapshot.status #=> String resp.snapshot.port #=> Integer resp.snapshot.availability_zone #=> String resp.snapshot.cluster_create_time #=> Time resp.snapshot.master_username #=> String resp.snapshot.cluster_version #=> String resp.snapshot.engine_full_version #=> String resp.snapshot.snapshot_type #=> String resp.snapshot.node_type #=> String resp.snapshot.number_of_nodes #=> Integer resp.snapshot.db_name #=> String resp.snapshot.vpc_id #=> String resp.snapshot.encrypted #=> Boolean resp.snapshot.kms_key_id #=> String resp.snapshot.encrypted_with_hsm #=> Boolean resp.snapshot.accounts_with_restore_access #=> Array resp.snapshot.accounts_with_restore_access[0].account_id #=> String resp.snapshot.accounts_with_restore_access[0].account_alias #=> String resp.snapshot.owner_account #=> String resp.snapshot.total_backup_size_in_mega_bytes #=> Float resp.snapshot.actual_incremental_backup_size_in_mega_bytes #=> Float resp.snapshot.backup_progress_in_mega_bytes #=> Float resp.snapshot.current_backup_rate_in_mega_bytes_per_second #=> Float resp.snapshot.estimated_seconds_to_completion #=> Integer resp.snapshot.elapsed_time_in_seconds #=> Integer resp.snapshot.source_region #=> String resp.snapshot.tags #=> Array resp.snapshot.tags[0].key #=> String resp.snapshot.tags[0].value #=> String resp.snapshot.restorable_node_types #=> Array resp.snapshot.restorable_node_types[0] #=> String resp.snapshot.enhanced_vpc_routing #=> Boolean resp.snapshot.maintenance_track_name #=> String resp.snapshot.manual_snapshot_retention_period #=> Integer resp.snapshot.manual_snapshot_remaining_days #=> Integer resp.snapshot.snapshot_retention_start_time #=> Time
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/RevokeSnapshotAccess AWS API Documentation
@overload revoke_snapshot_access
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 10388 def revoke_snapshot_access(params = {}, options = {}) req = build_request(:revoke_snapshot_access, params) req.send_request(options) end
Rotates the encryption keys for a cluster.
@option params [required, String] :cluster_identifier
The unique identifier of the cluster that you want to rotate the encryption keys for. Constraints: Must be the name of valid cluster that has encryption enabled.
@return [Types::RotateEncryptionKeyResult] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::RotateEncryptionKeyResult#cluster #cluster} => Types::Cluster
@example Request syntax with placeholder values
resp = client.rotate_encryption_key({ cluster_identifier: "String", # required })
@example Response structure
resp.cluster.cluster_identifier #=> String resp.cluster.node_type #=> String resp.cluster.cluster_status #=> String resp.cluster.cluster_availability_status #=> String resp.cluster.modify_status #=> String resp.cluster.master_username #=> String resp.cluster.db_name #=> String resp.cluster.endpoint.address #=> String resp.cluster.endpoint.port #=> Integer resp.cluster.endpoint.vpc_endpoints #=> Array resp.cluster.endpoint.vpc_endpoints[0].vpc_endpoint_id #=> String resp.cluster.endpoint.vpc_endpoints[0].vpc_id #=> String resp.cluster.endpoint.vpc_endpoints[0].network_interfaces #=> Array resp.cluster.endpoint.vpc_endpoints[0].network_interfaces[0].network_interface_id #=> String resp.cluster.endpoint.vpc_endpoints[0].network_interfaces[0].subnet_id #=> String resp.cluster.endpoint.vpc_endpoints[0].network_interfaces[0].private_ip_address #=> String resp.cluster.endpoint.vpc_endpoints[0].network_interfaces[0].availability_zone #=> String resp.cluster.cluster_create_time #=> Time resp.cluster.automated_snapshot_retention_period #=> Integer resp.cluster.manual_snapshot_retention_period #=> Integer resp.cluster.cluster_security_groups #=> Array resp.cluster.cluster_security_groups[0].cluster_security_group_name #=> String resp.cluster.cluster_security_groups[0].status #=> String resp.cluster.vpc_security_groups #=> Array resp.cluster.vpc_security_groups[0].vpc_security_group_id #=> String resp.cluster.vpc_security_groups[0].status #=> String resp.cluster.cluster_parameter_groups #=> Array resp.cluster.cluster_parameter_groups[0].parameter_group_name #=> String resp.cluster.cluster_parameter_groups[0].parameter_apply_status #=> String resp.cluster.cluster_parameter_groups[0].cluster_parameter_status_list #=> Array resp.cluster.cluster_parameter_groups[0].cluster_parameter_status_list[0].parameter_name #=> String resp.cluster.cluster_parameter_groups[0].cluster_parameter_status_list[0].parameter_apply_status #=> String resp.cluster.cluster_parameter_groups[0].cluster_parameter_status_list[0].parameter_apply_error_description #=> String resp.cluster.cluster_subnet_group_name #=> String resp.cluster.vpc_id #=> String resp.cluster.availability_zone #=> String resp.cluster.preferred_maintenance_window #=> String resp.cluster.pending_modified_values.master_user_password #=> String resp.cluster.pending_modified_values.node_type #=> String resp.cluster.pending_modified_values.number_of_nodes #=> Integer resp.cluster.pending_modified_values.cluster_type #=> String resp.cluster.pending_modified_values.cluster_version #=> String resp.cluster.pending_modified_values.automated_snapshot_retention_period #=> Integer resp.cluster.pending_modified_values.cluster_identifier #=> String resp.cluster.pending_modified_values.publicly_accessible #=> Boolean resp.cluster.pending_modified_values.enhanced_vpc_routing #=> Boolean resp.cluster.pending_modified_values.maintenance_track_name #=> String resp.cluster.pending_modified_values.encryption_type #=> String resp.cluster.cluster_version #=> String resp.cluster.allow_version_upgrade #=> Boolean resp.cluster.number_of_nodes #=> Integer resp.cluster.publicly_accessible #=> Boolean resp.cluster.encrypted #=> Boolean resp.cluster.restore_status.status #=> String resp.cluster.restore_status.current_restore_rate_in_mega_bytes_per_second #=> Float resp.cluster.restore_status.snapshot_size_in_mega_bytes #=> Integer resp.cluster.restore_status.progress_in_mega_bytes #=> Integer resp.cluster.restore_status.elapsed_time_in_seconds #=> Integer resp.cluster.restore_status.estimated_time_to_completion_in_seconds #=> Integer resp.cluster.data_transfer_progress.status #=> String resp.cluster.data_transfer_progress.current_rate_in_mega_bytes_per_second #=> Float resp.cluster.data_transfer_progress.total_data_in_mega_bytes #=> Integer resp.cluster.data_transfer_progress.data_transferred_in_mega_bytes #=> Integer resp.cluster.data_transfer_progress.estimated_time_to_completion_in_seconds #=> Integer resp.cluster.data_transfer_progress.elapsed_time_in_seconds #=> Integer resp.cluster.hsm_status.hsm_client_certificate_identifier #=> String resp.cluster.hsm_status.hsm_configuration_identifier #=> String resp.cluster.hsm_status.status #=> String resp.cluster.cluster_snapshot_copy_status.destination_region #=> String resp.cluster.cluster_snapshot_copy_status.retention_period #=> Integer resp.cluster.cluster_snapshot_copy_status.manual_snapshot_retention_period #=> Integer resp.cluster.cluster_snapshot_copy_status.snapshot_copy_grant_name #=> String resp.cluster.cluster_public_key #=> String resp.cluster.cluster_nodes #=> Array resp.cluster.cluster_nodes[0].node_role #=> String resp.cluster.cluster_nodes[0].private_ip_address #=> String resp.cluster.cluster_nodes[0].public_ip_address #=> String resp.cluster.elastic_ip_status.elastic_ip #=> String resp.cluster.elastic_ip_status.status #=> String resp.cluster.cluster_revision_number #=> String resp.cluster.tags #=> Array resp.cluster.tags[0].key #=> String resp.cluster.tags[0].value #=> String resp.cluster.kms_key_id #=> String resp.cluster.enhanced_vpc_routing #=> Boolean resp.cluster.iam_roles #=> Array resp.cluster.iam_roles[0].iam_role_arn #=> String resp.cluster.iam_roles[0].apply_status #=> String resp.cluster.pending_actions #=> Array resp.cluster.pending_actions[0] #=> String resp.cluster.maintenance_track_name #=> String resp.cluster.elastic_resize_number_of_node_options #=> String resp.cluster.deferred_maintenance_windows #=> Array resp.cluster.deferred_maintenance_windows[0].defer_maintenance_identifier #=> String resp.cluster.deferred_maintenance_windows[0].defer_maintenance_start_time #=> Time resp.cluster.deferred_maintenance_windows[0].defer_maintenance_end_time #=> Time resp.cluster.snapshot_schedule_identifier #=> String resp.cluster.snapshot_schedule_state #=> String, one of "MODIFYING", "ACTIVE", "FAILED" resp.cluster.expected_next_snapshot_schedule_time #=> Time resp.cluster.expected_next_snapshot_schedule_time_status #=> String resp.cluster.next_maintenance_window_start_time #=> Time resp.cluster.resize_info.resize_type #=> String resp.cluster.resize_info.allow_cancel_resize #=> Boolean resp.cluster.availability_zone_relocation_status #=> String resp.cluster.cluster_namespace_arn #=> String resp.cluster.total_storage_capacity_in_mega_bytes #=> Integer resp.cluster.aqua_configuration.aqua_status #=> String, one of "enabled", "disabled", "applying" resp.cluster.aqua_configuration.aqua_configuration_status #=> String, one of "enabled", "disabled", "auto"
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/RotateEncryptionKey AWS API Documentation
@overload rotate_encryption_key
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 10527 def rotate_encryption_key(params = {}, options = {}) req = build_request(:rotate_encryption_key, params) req.send_request(options) end
Updates the status of a partner integration.
@option params [required, String] :account_id
The Amazon Web Services account ID that owns the cluster.
@option params [required, String] :cluster_identifier
The cluster identifier of the cluster whose partner integration status is being updated.
@option params [required, String] :database_name
The name of the database whose partner integration status is being updated.
@option params [required, String] :partner_name
The name of the partner whose integration status is being updated.
@option params [required, String] :status
The value of the updated status.
@option params [String] :status_message
The status message provided by the partner.
@return [Types::PartnerIntegrationOutputMessage] Returns a {Seahorse::Client::Response response} object which responds to the following methods:
* {Types::PartnerIntegrationOutputMessage#database_name #database_name} => String * {Types::PartnerIntegrationOutputMessage#partner_name #partner_name} => String
@example Request syntax with placeholder values
resp = client.update_partner_status({ account_id: "PartnerIntegrationAccountId", # required cluster_identifier: "PartnerIntegrationClusterIdentifier", # required database_name: "PartnerIntegrationDatabaseName", # required partner_name: "PartnerIntegrationPartnerName", # required status: "Active", # required, accepts Active, Inactive, RuntimeFailure, ConnectionFailure status_message: "PartnerIntegrationStatusMessage", })
@example Response structure
resp.database_name #=> String resp.partner_name #=> String
@see docs.aws.amazon.com/goto/WebAPI/redshift-2012-12-01/UpdatePartnerStatus AWS API Documentation
@overload update_partner_status
(params = {}) @param [Hash] params ({})
# File lib/aws-sdk-redshift/client.rb, line 10579 def update_partner_status(params = {}, options = {}) req = build_request(:update_partner_status, params) req.send_request(options) end
Polls an API operation until a resource enters a desired state.
## Basic Usage
A waiter will call an API operation until:
-
It is successful
-
It enters a terminal state
-
It makes the maximum number of attempts
In between attempts, the waiter will sleep.
# polls in a loop, sleeping between attempts client.wait_until(waiter_name, params)
## Configuration
You can configure the maximum number of polling attempts, and the delay (in seconds) between each polling attempt. You can pass configuration as the final arguments hash.
# poll for ~25 seconds client.wait_until(waiter_name, params, { max_attempts: 5, delay: 5, })
## Callbacks
You can be notified before each polling attempt and before each delay. If you throw `:success` or `:failure` from these callbacks, it will terminate the waiter.
started_at = Time.now client.wait_until(waiter_name, params, { # disable max attempts max_attempts: nil, # poll for 1 hour, instead of a number of attempts before_wait: -> (attempts, response) do throw :failure if Time.now - started_at > 3600 end })
## Handling Errors
When a waiter is unsuccessful, it will raise an error. All of the failure errors extend from {Aws::Waiters::Errors::WaiterFailed}.
begin client.wait_until(...) rescue Aws::Waiters::Errors::WaiterFailed # resource did not enter the desired state in time end
## Valid Waiters
The following table lists the valid waiter names, the operations they call, and the default `:delay` and `:max_attempts` values.
| waiter_name | params | :delay | :max_attempts | | —————— | ———————————– | ——– | ————- | | cluster_available | {Client#describe_clusters} | 60 | 30 | | cluster_deleted | {Client#describe_clusters} | 60 | 30 | | cluster_restored | {Client#describe_clusters} | 60 | 30 | | snapshot_available | {Client#describe_cluster_snapshots} | 15 | 20 |
@raise [Errors::FailureStateError] Raised when the waiter terminates
because the waiter has entered a state that it will not transition out of, preventing success.
@raise [Errors::TooManyAttemptsError] Raised when the configured
maximum number of attempts have been made, and the waiter is not yet successful.
@raise [Errors::UnexpectedError] Raised when an error is encounted
while polling for a resource that is not expected.
@raise [Errors::NoSuchWaiterError] Raised when you request to wait
for an unknown state.
@return [Boolean] Returns `true` if the waiter was successful. @param [Symbol] waiter_name @param [Hash] params ({}) @param [Hash] options ({}) @option options [Integer] :max_attempts @option options [Integer] :delay @option options [Proc] :before_attempt @option options [Proc] :before_wait
# File lib/aws-sdk-redshift/client.rb, line 10692 def wait_until(waiter_name, params = {}, options = {}) w = waiter(waiter_name, options) yield(w.waiter) if block_given? # deprecated w.wait(params) end
@api private @deprecated
# File lib/aws-sdk-redshift/client.rb, line 10700 def waiter_names waiters.keys end
Private Instance Methods
@param [Symbol] waiter_name @param [Hash] options ({})
# File lib/aws-sdk-redshift/client.rb, line 10708 def waiter(waiter_name, options = {}) waiter_class = waiters[waiter_name] if waiter_class waiter_class.new(options.merge(client: self)) else raise Aws::Waiters::Errors::NoSuchWaiterError.new(waiter_name, waiters.keys) end end
# File lib/aws-sdk-redshift/client.rb, line 10717 def waiters { cluster_available: Waiters::ClusterAvailable, cluster_deleted: Waiters::ClusterDeleted, cluster_restored: Waiters::ClusterRestored, snapshot_available: Waiters::SnapshotAvailable } end