class Sequel::ShardedTimedQueueConnectionPool
A connection pool allowing multi-threaded access to a sharded pool of connections, using a timed queue (only available in Ruby 3.2+).
Attributes
The maximum number of connections this pool will create per shard.
Public Class Methods
The following additional options are respected:
- :max_connections
-
The maximum number of connections the connection pool will open (default 4)
- :pool_timeout
-
The amount of seconds to wait to acquire a connection before raising a PoolTimeout (default 5)
- :servers
-
A hash of servers to use. Keys should be symbols. If not present, will use a single :default server.
- :servers_hash
-
The base hash to use for the servers. By default, Sequel uses Hash.new(:default). You can use a hash with a default proc that raises an error if you want to catch all cases where a nonexistent server is used.
# File lib/sequel/connection_pool/sharded_timed_queue.rb, line 23 def initialize(db, opts = OPTS) super @max_size = Integer(opts[:max_connections] || 4) raise(Sequel::Error, ':max_connections must be positive') if @max_size < 1 @mutex = Mutex.new @timeout = Float(opts[:pool_timeout] || 5) @allocated = {} @sizes = {} @queues = {} @servers = opts.fetch(:servers_hash, Hash.new(:default)) add_servers([:default]) add_servers(opts[:servers].keys) if opts[:servers] end
Public Instance Methods
Adds new servers to the connection pool. Allows for dynamic expansion of
the potential replicas/shards at runtime. servers
argument
should be an array of symbols.
# File lib/sequel/connection_pool/sharded_timed_queue.rb, line 42 def add_servers(servers) sync do servers.each do |server| next if @servers.has_key?(server) @servers[server] = server @sizes[server] = 0 @queues[server] = Queue.new (@allocated[server] = {}).compare_by_identity end end nil end
Yield all of the available connections, and the one currently allocated to this thread (if one is allocated). This will not yield connections currently allocated to other threads, as it is not safe to operate on them.
# File lib/sequel/connection_pool/sharded_timed_queue.rb, line 59 def all_connections thread = Sequel.current sync{@queues.to_a}.each do |server, queue| if conn = owned_connection(thread, server) yield conn end # Use a hash to record all connections already seen. As soon as we # come across a connection we've already seen, we stop the loop. conns = {} conns.compare_by_identity while true conn = nil begin break unless (conn = queue.pop(timeout: 0)) && !conns[conn] conns[conn] = true yield conn ensure queue.push(conn) if conn end end end nil end
Removes all connections currently in the pool's queue. This method has the effect of disconnecting from the database, assuming that no connections are currently being used.
Once a connection is requested using hold, the connection pool creates new connections to the database.
If the :server option is provided, it should be a symbol or array of symbols, and then the method will only disconnect connectsion from those specified shards.
# File lib/sequel/connection_pool/sharded_timed_queue.rb, line 94 def disconnect(opts=OPTS) (opts[:server] ? Array(opts[:server]) : sync{@servers.keys}).each do |server| raise Sequel::Error, "invalid server" unless queue = sync{@queues[server]} while conn = queue.pop(timeout: 0) disconnect_pool_connection(conn, server) end fill_queue(server) end nil end
Chooses the first available connection for the given server, or if none are available, creates a new connection. Passes the connection to the supplied block:
pool.hold(:server1) {|conn| conn.execute('DROP TABLE posts')}
Pool#hold is re-entrant, meaning it can be called recursively in the same thread without blocking.
If no connection is immediately available and the pool is already using the maximum number of connections, Pool#hold will block until a connection is available or the timeout expires. If the timeout expires before a connection can be acquired, a Sequel::PoolTimeout is raised.
# File lib/sequel/connection_pool/sharded_timed_queue.rb, line 118 def hold(server=:default) server = pick_server(server) t = Sequel.current if conn = owned_connection(t, server) return yield(conn) end begin conn = acquire(t, server) yield conn rescue Sequel::DatabaseDisconnectError, *@error_classes => e if disconnect_error?(e) oconn = conn conn = nil disconnect_pool_connection(oconn, server) if oconn sync{@allocated[server].delete(t)} fill_queue(server) end raise ensure release(t, conn, server) if conn end end
# File lib/sequel/connection_pool/sharded_timed_queue.rb, line 189 def pool_type :sharded_timed_queue end
Remove servers from the connection pool. Similar to disconnecting from all given servers, except that after it is used, future requests for the servers will use the :default server instead.
Note that an error will be raised if there are any connections currently checked out for the given servers.
# File lib/sequel/connection_pool/sharded_timed_queue.rb, line 153 def remove_servers(servers) conns = [] raise(Sequel::Error, "cannot remove default server") if servers.include?(:default) sync do servers.each do |server| next unless @servers.has_key?(server) queue = @queues[server] while conn = queue.pop(timeout: 0) @sizes[server] -= 1 conns << conn end unless @sizes[server] == 0 raise Sequel::Error, "cannot remove server #{server} as it has allocated connections" end @servers.delete(server) @sizes.delete(server) @queues.delete(server) @allocated.delete(server) end end nil ensure disconnect_connections(conns) end
Return an array of symbols for servers in the connection pool.
# File lib/sequel/connection_pool/sharded_timed_queue.rb, line 185 def servers sync{@servers.keys} end
The total number of connections in the pool. Using a non-existant server will return nil.
# File lib/sequel/connection_pool/sharded_timed_queue.rb, line 143 def size(server=:default) sync{@sizes[server]} end
Private Instance Methods
Assigns a connection to the supplied thread, if one is available.
This should return a connection if one is available within the timeout, or raise PoolTimeout if a connection could not be acquired within the timeout.
Calling code should not have the mutex when calling this.
# File lib/sequel/connection_pool/sharded_timed_queue.rb, line 305 def acquire(thread, server) queue = sync{@queues[server]} if conn = queue.pop(timeout: 0) || try_make_new(server) || queue.pop(timeout: @timeout) sync{@allocated[server][thread] = conn} else name = db.opts[:name] raise ::Sequel::PoolTimeout, "timeout: #{@timeout}, server: #{server}#{", database name: #{name}" if name}" end end
Whether the given size is less than the maximum size of the pool. In that case, the pool's current size is incremented. If this method returns true, space in the pool for the connection is preallocated, and #preallocated_make_new should be called to create the connection.
Calling code should have the mutex when calling this.
# File lib/sequel/connection_pool/sharded_timed_queue.rb, line 258 def can_make_new?(server, current_size) if @max_size > current_size @sizes[server] += 1 end end
Adds a connection to the queue of available connections, returns the connection.
# File lib/sequel/connection_pool/sharded_timed_queue.rb, line 362 def checkin_connection(conn, server) sync{@queues[server]}.push(conn) conn end
Disconnect all available connections immediately, and schedule currently allocated connections for disconnection as soon as they are returned to the pool. The calling code should NOT have the mutex before calling this.
# File lib/sequel/connection_pool/sharded_timed_queue.rb, line 217 def disconnect_connections(conns) conns.each{|conn| disconnect_connection(conn)} end
Decrement the current size of the pool for the server when disconnecting connections.
Calling code should not have the mutex when calling this.
# File lib/sequel/connection_pool/sharded_timed_queue.rb, line 224 def disconnect_pool_connection(conn, server) sync{@sizes[server] -= 1} disconnect_connection(conn) end
If there are any threads waiting on the queue, try to create new connections in a separate thread if the pool is not yet at the maximum size.
The reason for this method is to handle cases where acquire could not retrieve a connection immediately, and the pool was already at the maximum size. In that case, the acquire will wait on the queue until the timeout. This method is called after disconnecting to potentially add new connections to the pool, so the threads that are currently waiting for connections do not timeout after the pool is no longer full.
# File lib/sequel/connection_pool/sharded_timed_queue.rb, line 240 def fill_queue(server) queue = sync{@queues[server]} if queue.num_waiting > 0 Thread.new do while queue.num_waiting > 0 && (conn = try_make_new(server)) queue.push(conn) end end end end
Returns the connection owned by the supplied thread for the given server, if any. The calling code should NOT already have the mutex before calling this.
# File lib/sequel/connection_pool/sharded_timed_queue.rb, line 317 def owned_connection(thread, server) sync{@allocated[server][thread]} end
If the server given is in the hash, return it, otherwise, return the default server.
# File lib/sequel/connection_pool/sharded_timed_queue.rb, line 322 def pick_server(server) sync{@servers[server]} end
Create a new connection, after the pool's current size has already been updated to account for the new connection. If there is an exception when creating the connection, decrement the current size.
This should only be called after can_make_new?. If there is an exception between when can_make_new? is called and when #preallocated_make_new is called, it has the effect of reducing the maximum size of the connection pool by 1, since the current size of the pool will show a higher number than the number of connections allocated or in the queue.
Calling code should not have the mutex when calling this.
# File lib/sequel/connection_pool/sharded_timed_queue.rb, line 207 def preallocated_make_new(server) make_new(server) rescue Exception sync{@sizes[server] -= 1} raise end
Create the maximum number of connections immediately. This should not be called with a true argument unless no code is currently operating on the database.
Calling code should not have the mutex when calling this.
# File lib/sequel/connection_pool/sharded_timed_queue.rb, line 330 def preconnect(concurrent = false) conn_servers = sync{@servers.keys}.map!{|s| Array.new(@max_size - @sizes[s], s)}.flatten! if concurrent conn_servers.map! do |server| queue = sync{@queues[server]} Thread.new do if conn = try_make_new(server) queue.push(conn) end end end.each(&:value) else conn_servers.each do |server| if conn = try_make_new(server) sync{@queues[server]}.push(conn) end end end nil end
Releases the connection assigned to the supplied thread back to the pool.
Calling code should not have the mutex when calling this.
# File lib/sequel/connection_pool/sharded_timed_queue.rb, line 356 def release(thread, _, server) checkin_connection(sync{@allocated[server].delete(thread)}, server) nil end
Yield to the block while inside the mutex.
Calling code should not have the mutex when calling this.
# File lib/sequel/connection_pool/sharded_timed_queue.rb, line 370 def sync @mutex.synchronize{yield} end
Try to make a new connection if there is space in the pool. If the pool is already full, look for dead threads/fibers and disconnect the related connections.
Calling code should not have the mutex when calling this.
# File lib/sequel/connection_pool/sharded_timed_queue.rb, line 269 def try_make_new(server) return preallocated_make_new(server) if sync{can_make_new?(server, @sizes[server])} to_disconnect = nil do_make_new = false sync do current_size = @sizes[server] alloc = @allocated[server] alloc.keys.each do |t| unless t.alive? (to_disconnect ||= []) << alloc.delete(t) current_size -= 1 end end do_make_new = true if can_make_new?(server, current_size) end begin preallocated_make_new(server) if do_make_new ensure if to_disconnect to_disconnect.each{|conn| disconnect_pool_connection(conn, server)} fill_queue(server) end end end