module Moped # The cluster represents a cluster of MongoDB server nodes, either a single # node, a replica set, or a mongos server. class Cluster # @attribute [r] options The cluster options. # @attribute [r] seeds The seeds the cluster was initialized with. attr_reader :options, :seeds # Get the authentication details for the cluster. # # @example Get the authentication details. # cluster.auth # # @return [ Hash ] the cached authentication credentials for this cluster. # # @since 1.0.0 def auth @auth ||= {} end # Disconnects all nodes in the cluster. This should only be used in cases # where you know you're not going to use the cluster on the thread anymore # and need to force the connections to close. # # @return [ true ] True if the disconnect succeeded. # # @since 1.2.0 def disconnect nodes(include_arbiters: true).each { |node| node.disconnect } and true end # Get the interval at which a node should be flagged as down before # retrying. # # @example Get the down interval, in seconds. # cluster.down_interval # # @return [ Integer ] The down interval. # # @since 1.2.7 def down_interval options[:down_interval] end # Get the number of times an operation should be retried before raising an # error. # # @example Get the maximum retries. # cluster.max_retries # # @return [ Integer ] The max retries. # # @since 1.2.7 def max_retries options[:max_retries] end # Get the interval in which the node list should be refreshed. # # @example Get the refresh interval, in seconds. # cluster.refresh_interval # # @return [ Integer ] The refresh interval. # # @since 1.2.7 def refresh_interval options[:refresh_interval] end # Get the operation retry interval - the time to wait before retrying a # single operation. # # @example Get the retry interval, in seconds. # cluster.retry_interval # # @return [ Integer ] The retry interval. # # @since 1.2.7 def retry_interval options[:retry_interval] end # Initialize the new cluster. # # @example Initialize the cluster. # Cluster.new([ "localhost:27017" ], down_interval: 20) # # @param [ Hash ] options The cluster options. # # @option options :down_interval number of seconds to wait before attempting # to reconnect to a down node. (30) # @option options :refresh_interval number of seconds to cache information # about a node. (300) # @option options [ Integer ] :timeout The time in seconds to wait for an # operation to timeout. (5) # # @since 1.0.0 def initialize(hosts, options) @seeds = hosts @nodes = hosts.map { |host| Node.new(host, options) } @options = { down_interval: 30, max_retries: 30, refresh_interval: 300, retry_interval: 1 }.merge(options) end # Returns the list of available nodes, refreshing 1) any nodes which were # down and ready to be checked again and 2) any nodes whose information is # out of date. Arbiter nodes are not returned. # # @example Get the available nodes. # cluster.nodes # # @return [ Array ] the list of available nodes. # # @since 1.0.0 def nodes(opts = {}) current_time = Time.new down_boundary = current_time - down_interval refresh_boundary = current_time - refresh_interval # Find the nodes that were down but are ready to be refreshed, or those # with stale connection information. needs_refresh, available = @nodes.partition do |node| (node.down? && node.down_at < down_boundary) || node.needs_refresh?(refresh_boundary) end # Refresh those nodes. available.concat refresh(needs_refresh) # Now return all the nodes that are available and participating in the # replica set. available.reject { |node| node.down? || (!opts[:include_arbiters] && node.arbiter?) } end # Refreshes information for each of the nodes provided. The node list # defaults to the list of all known nodes. # # If a node is successfully refreshed, any newly discovered peers will also # be refreshed. # # @example Refresh the nodes. # cluster.refresh # # @param [ Array ] nodes_to_refresh The nodes to refresh. # # @return [ Array ] the available nodes # # @since 1.0.0 def refresh(nodes_to_refresh = @nodes) refreshed_nodes = [] seen = {} # Set up a recursive lambda function for refreshing a node and it's peers. refresh_node = ->(node) do unless seen[node] seen[node] = true # Add the node to the global list of known nodes. @nodes << node unless @nodes.include?(node) begin node.refresh # This node is good, so add it to the list of nodes to return. refreshed_nodes << node unless refreshed_nodes.include?(node) # Now refresh any newly discovered peer nodes. (node.peers - @nodes).each(&refresh_node) if node.peers rescue Errors::ConnectionFailure # We couldn't connect to the node, so don't do anything with it. end end end nodes_to_refresh.each(&refresh_node) refreshed_nodes.to_a end # Yields the replica set's primary node to the provided block. This method # will retry the block in case of connection errors or replica set # reconfiguration. # # @example Yield the primary to the block. # cluster.with_primary do |node| # # ... # end # # @param [ Integer ] retries The number of times to retry. # # @raises [ ConnectionFailure ] When no primary node can be found # # @return [ Object ] The result of the yield. # # @since 1.0.0 def with_primary(retries = max_retries, &block) if node = nodes.find(&:primary?) begin node.ensure_primary do return yield node.apply_auth(auth) end rescue Errors::ConnectionFailure, Errors::ReplicaSetReconfigured # Fall through to the code below if our connection was dropped or the # node is no longer the primary. end end if retries > 0 # We couldn't find a primary node, so refresh the list and try again. sleep(retry_interval) refresh with_primary(retries - 1, &block) else raise( Errors::ConnectionFailure, "Could not connect to a primary node for replica set #{inspect}" ) end end # Yields a secondary node if available, otherwise the primary node. This # method will retry the block in case of connection errors. # # @example Yield the secondary to the block. # cluster.with_secondary do |node| # # ... # end # # @param [ Integer ] retries The number of times to retry. # # @raises [ ConnectionFailure ] When no primary node can be found # # @return [ Object ] The result of the yield. # # @since 1.0.0 def with_secondary(retries = max_retries, &block) available_nodes = nodes.shuffle!.partition(&:secondary?).flatten while node = available_nodes.shift begin return yield node.apply_auth(auth) rescue Errors::ConnectionFailure # That node's no good, so let's try the next one. next end end if retries > 0 # We couldn't find a secondary or primary node, so refresh the list and # try again. sleep(retry_interval) refresh with_secondary(retries - 1, &block) else raise( Errors::ConnectionFailure, "Could not connect to any secondary or primary nodes for replica set #{inspect}" ) end end def inspect "<#{self.class.name} nodes=#{@nodes.inspect}>" end private def initialize_copy(_) @nodes = @nodes.map(&:dup) end end end