require 'mongo' require 'resque/version' require 'resque/errors' require 'resque/failure' require 'resque/failure/base' require 'resque/helpers' require 'resque/stat' require 'resque/job' require 'resque/worker' require 'resque/plugin' module Resque include Helpers extend self @delayed_queues = [] # Set the queue database. Expects a Mongo::DB object. def mongo=(database) if database.is_a?(Mongo::DB) @mongo = database initialize_mongo else raise ArgumentError, "Resque.mongo= expects a Mongo::DB database, not a #{database.class}." end end # Returns the current Mongo::DB. If none has been created, it will # create a new one called 'resque'. def mongo return @mongo if @mongo self.mongo = Mongo::Connection.new.db("resque") @mongo end def initialize_mongo mongo_workers.create_index :worker mongo_stats.create_index :stat delayed_queues = mongo_stats.find_one(:stat => 'Delayed Queues') @delayed_queues = delayed_queues['value'] if delayed_queues end def mongo_workers mongo['resque.workers'] end def mongo_stats mongo['resque.metrics'] end def mongo_failures mongo['resque.failures'] end # The `before_first_fork` hook will be run in the **parent** process # only once, before forking to run the first job. Be careful- any # changes you make will be permanent for the lifespan of the # worker. # # Call with a block to set the hook. # Call with no arguments to return the hook. def before_first_fork(&block) block ? (@before_first_fork = block) : @before_first_fork end # Set a proc that will be called in the parent process before the # worker forks for the first time. def before_first_fork=(before_first_fork) @before_first_fork = before_first_fork end # The `before_fork` hook will be run in the **parent** process # before every job, so be careful- any changes you make will be # permanent for the lifespan of the worker. # # Call with a block to set the hook. # Call with no arguments to return the hook. def before_fork(&block) block ? (@before_fork = block) : @before_fork end # Set the before_fork proc. def before_fork=(before_fork) @before_fork = before_fork end # The `after_fork` hook will be run in the child process and is passed # the current job. Any changes you make, therefore, will only live as # long as the job currently being processed. # # Call with a block to set the hook. # Call with no arguments to return the hook. def after_fork(&block) block ? (@after_fork = block) : @after_fork end # Set the after_fork proc. def after_fork=(after_fork) @after_fork = after_fork end def to_s connection_info = mongo.connection.primary_pool "Resque Client connected to #{connection_info.host}:#{connection_info.port}/#{mongo.name}" end def delayed_job?(klass) klass.instance_variable_get(:@delayed) || (klass.respond_to?(:delayed) and klass.delayed) end def delayed_queue?(queue) @delayed_queues.include? namespace_queue(queue) end def enable_delay(queue) queue = namespace_queue(queue) unless delayed_queue? queue @delayed_queues << queue mongo_stats.update({:stat => 'Delayed Queues'}, {'$addToSet' => {'value' => queue}}, {:upsert => true}) end end # If 'inline' is true Resque will call #perform method inline # without queuing it into Redis and without any Resque callbacks. # The 'inline' is false Resque jobs will be put in queue regularly. def inline? @inline end alias_method :inline, :inline? def inline=(inline) @inline = inline end # # queue manipulation # # Pushes a job onto a queue. Queue name should be a string and the # item should be any JSON-able Ruby object. # # Resque works generally expect the `item` to be a hash with the following # keys: # # class - The String name of the job to run. # args - An Array of arguments to pass the job. Usually passed # via `class.to_class.perform(*args)`. # # Example # # Resque.push('archive', :class => 'Archive', :args => [ 35, 'tar' ]) # # Returns nothing def push(queue, item) queue = namespace_queue(queue) item[:resque_enqueue_timestamp] = Time.now mongo[queue] << item end # Pops a job off a queue. Queue name should be a string. # # Returns a Ruby object. def pop(queue) queue = namespace_queue(queue) query = {} query['delay_until'] = { '$lt' => Time.now } if delayed_queue?(queue) #sorting will result in significant performance penalties for large queues, you have been warned. item = mongo[queue].find_and_modify(:query => query, :remove => true, :sort => [[:_id, :asc]]) rescue Mongo::OperationFailure => e return nil if e.message =~ /No matching object/ raise e end # Returns an integer representing the size of a queue. # Queue name should be a string. def size(queue) queue = namespace_queue(queue) mongo[queue].count end def delayed_size(queue) queue = namespace_queue(queue) if delayed_queue? queue mongo[queue].find({'delay_until' => { '$gt' => Time.now }}).count else mongo[queue].count end end def ready_size(queue) queue = namespace_queue(queue) if delayed_queue? queue mongo[queue].find({'delay_until' => { '$lt' => Time.now }}).count else mongo[queue].count end end # Returns an array of items currently queued. Queue name should be # a string. # # start and count should be integer and can be used for pagination. # start is the item to begin, count is how many items to return. # # To get the 3rd page of a 30 item, paginatied list one would use: # Resque.peek('my_list', 59, 30) def peek(queue, start = 0, count = 1, mode = :ready) list_range(queue, start, count, mode) end # Does the dirty work of fetching a range of items from a Redis list # and converting them into Ruby objects. def list_range(key, start = 0, count = 1, mode = :ready) query = { } sort = [] if delayed_queue? key if mode == :ready query['delay_until'] = { '$not' => { '$gt' => Time.new}} elsif mode == :delayed query['delay_until'] = { '$gt' => Time.new} elsif mode == :delayed_sorted query['delay_until'] = { '$gt' => Time.new} sort << ['delay_until', 1] elsif mode == :all_sorted query = {} sort << ['delay_until', 1] end end queue = namespace_queue(key) items = mongo[queue].find(query, { :limit => count, :skip => start, :sort => sort}).to_a.map{ |i| i} count > 1 ? items : items.first end # Returns an array of all known Resque queues as strings. def queues mongo.collection_names. select { |name| name =~ /resque\.queues\./ }. collect { |name| name.split(".")[2..-1].join('.') } end # Returns the mongo collection for a given queue def collection_for_queue(queue) queue = namespace_queue(queue) mongo[queue] end # Given a queue name, completely deletes the queue. def remove_queue(queue) queue = namespace_queue(queue) mongo[queue].drop end # # job shortcuts # # This method can be used to conveniently add a job to a queue. # It assumes the class you're passing it is a real Ruby class (not # a string or reference) which either: # # a) has a @queue ivar set # b) responds to `queue` # # If either of those conditions are met, it will use the value obtained # from performing one of the above operations to determine the queue. # # If no queue can be inferred this method will raise a `Resque::NoQueueError` # # This method is considered part of the `stable` API. def enqueue(klass, *args) # Perform before_enqueue hooks. Don't perform enqueue if any hook returns false before_hooks = Plugin.before_enqueue_hooks(klass).collect do |hook| klass.send(hook, *args) end return if before_hooks.any? { |result| result == false } Job.create(queue_from_class(klass), klass, *args) Plugin.after_enqueue_hooks(klass).each do |hook| klass.send(hook, *args) end end def enqueue_delayed(klass, *args) end # This method can be used to conveniently remove a job from a queue. # It assumes the class you're passing it is a real Ruby class (not # a string or reference) which either: # # a) has a @queue ivar set # b) responds to `queue` # # If either of those conditions are met, it will use the value obtained # from performing one of the above operations to determine the queue. # # If no queue can be inferred this method will raise a `Resque::NoQueueError` # # If no args are given, this method will dequeue *all* jobs matching # the provided class. See `Resque::Job.destroy` for more # information. # # Returns the number of jobs destroyed. # # Example: # # # Removes all jobs of class `UpdateNetworkGraph` # Resque.dequeue(GitHub::Jobs::UpdateNetworkGraph) # # # Removes all jobs of class `UpdateNetworkGraph` with matching args. # Resque.dequeue(GitHub::Jobs::UpdateNetworkGraph, 'repo:135325') # # This method is considered part of the `stable` API. def dequeue(klass, *args) Job.destroy(queue_from_class(klass), klass, *args) end # Given a class, try to extrapolate an appropriate queue based on a # class instance variable or `queue` method. def queue_from_class(klass) klass.instance_variable_get(:@queue) || (klass.respond_to?(:queue) and klass.queue) end # This method will return a `Resque::Job` object or a non-true value # depending on whether a job can be obtained. You should pass it the # precise name of a queue: case matters. # # This method is considered part of the `stable` API. def reserve(queue) Job.reserve(queue) end # Validates if the given klass could be a valid Resque job # # If no queue can be inferred this method will raise a `Resque::NoQueueError` # # If given klass is nil this method will raise a `Resque::NoClassError` def validate(klass, queue = nil) queue ||= queue_from_class(klass) if !queue raise NoQueueError.new("Jobs must be placed onto a queue.") end if klass.to_s.empty? raise NoClassError.new("Jobs must be given a class.") end end # # worker shortcuts # # A shortcut to Worker.all def workers Worker.all end # A shortcut to Worker.working def working Worker.working end # A shortcut to unregister_worker # useful for command line tool def remove_worker(worker_id) worker = Resque::Worker.find(worker_id) worker.unregister_worker end # # stats # # Returns a hash, similar to redis-rb's #info, of interesting stats. def info return { :pending => queues.inject(0) { |m,k| m + size(k) }, :processed => Stat[:processed], :queues => queues.size, :workers => workers.size.to_i, :working => working.count, :failed => Stat[:failed], :servers => to_s, :environment => ENV['RAILS_ENV'] || ENV['RACK_ENV'] || 'development' } end # Returns an array of all known Resque keys in Redis. Redis' KEYS operation # is O(N) for the keyspace, so be careful - this can be slow for big databases. def keys names = mongo.collection_names end def drop mongo.collections.each{ |collection| collection.drop unless collection.name =~ /^system./ } @mongo = nil end private def namespace_queue(queue) queue = queue.to_s if queue.start_with?('resque.queues.') queue else "resque.queues.#{queue}" end end end