canvas-jobs 0.9.0

data/db/migrate/20140512213941_add_source_to_jobs.rb

@@ -0,0 +1,15 @@
+class AddSourceToJobs < ActiveRecord::Migration
+  def self.connection
+    Delayed::Job.connection
+  end
+
+  def self.up
+    add_column :delayed_jobs, :source, :string
+    add_column :failed_jobs, :source, :string
+  end
+
+  def self.down
+    remove_column :delayed_jobs, :source
+    remove_column :failed_jobs, :source
+  end
+end

data/lib/canvas-jobs.rb

@@ -0,0 +1 @@
+require 'delayed_job'

data/lib/delayed/backend/active_record.rb

@@ -0,0 +1,297 @@
+class ActiveRecord::Base
+  def self.load_for_delayed_job(id)
+    if id
+      find(id)
+    else
+      super
+    end
+  end
+end
+
+module Delayed
+  module Backend
+    module ActiveRecord
+      # A job object that is persisted to the database.
+      # Contains the work object as a YAML field.
+      class Job < ::ActiveRecord::Base
+        include Delayed::Backend::Base
+        self.table_name = :delayed_jobs
+
+        def self.reconnect!
+          clear_all_connections!
+        end
+
+        # be aware that some strand functionality is controlled by triggers on
+        # the database. see
+        # db/migrate/20110831210257_add_delayed_jobs_next_in_strand.rb
+        #
+        # next_in_strand defaults to true. if we insert a new job, and it has a
+        # strand, and it's not the next in the strand, we set it to false.
+        #
+        # if we delete a job, and it has a strand, mark the next job in that
+        # strand to be next_in_strand
+        # (this is safe even if we're not deleting the job that was currently
+        # next_in_strand)
+
+        # postgresql needs this lock to be taken before the before_insert
+        # trigger starts, or we risk deadlock inside of the trigger when trying
+        # to raise the lock level
+        before_create :lock_strand_on_create
+        def lock_strand_on_create
+          if strand.present?
+            self.class.connection.execute("SELECT pg_advisory_xact_lock(half_md5_as_bigint(#{self.class.sanitize(strand)}))")
+          end
+        end
+
+        def self.current
+          where("run_at<=?", db_time_now)
+        end
+
+        def self.future
+          where("run_at>?", db_time_now)
+        end
+
+        def self.failed
+          where("failed_at IS NOT NULL")
+        end
+
+        def self.running
+          where("locked_at IS NOT NULL AND locked_by<>'on hold'")
+        end
+
+        # a nice stress test:
+        # 10_000.times { |i| Kernel.send_later_enqueue_args(:system, { :strand => 's1', :run_at => (24.hours.ago + (rand(24.hours.to_i))) }, "echo #{i} >> test1.txt") }
+        # 500.times { |i| "ohai".send_later_enqueue_args(:reverse, { :run_at => (12.hours.ago + (rand(24.hours.to_i))) }) }
+        # then fire up your workers
+        # you can check out strand correctness: diff test1.txt <(sort -n test1.txt)
+         def self.ready_to_run
+           where("run_at<=? AND locked_at IS NULL AND next_in_strand=?", db_time_now, true)
+         end
+        def self.by_priority
+          order("priority ASC, run_at ASC")
+        end
+
+        # When a worker is exiting, make sure we don't have any locked jobs.
+        def self.clear_locks!(worker_name)
+          where(:locked_by => worker_name).update_all(:locked_by => nil, :locked_at => nil)
+        end
+
+        def self.strand_size(strand)
+          self.where(:strand => strand).count
+        end
+
+        def self.running_jobs()
+          self.running.order(:locked_at)
+        end
+
+        def self.scope_for_flavor(flavor, query)
+          scope = case flavor.to_s
+          when 'current'
+            self.current
+          when 'future'
+            self.future
+          when 'failed'
+            Delayed::Job::Failed
+          when 'strand'
+            self.where(:strand => query)
+          when 'tag'
+            self.where(:tag => query)
+          else
+            raise ArgumentError, "invalid flavor: #{flavor.inspect}"
+          end
+
+          if %w(current future).include?(flavor.to_s)
+            queue = query.presence || Delayed::Settings.queue
+            scope = scope.where(:queue => queue)
+          end
+
+          scope
+        end
+
+        # get a list of jobs of the given flavor in the given queue
+        # flavor is :current, :future, :failed, :strand or :tag
+        # depending on the flavor, query has a different meaning:
+        # for :current and :future, it's the queue name (defaults to Delayed::Settings.queue)
+        # for :strand it's the strand name
+        # for :tag it's the tag name
+        # for :failed it's ignored
+        def self.list_jobs(flavor,
+                           limit,
+                           offset = 0,
+                           query = nil)
+          scope = self.scope_for_flavor(flavor, query)
+          order = flavor.to_s == 'future' ? 'run_at' : 'id desc'
+          scope.order(order).limit(limit).offset(offset).to_a
+        end
+
+        # get the total job count for the given flavor
+        # see list_jobs for documentation on arguments
+        def self.jobs_count(flavor,
+                            query = nil)
+          scope = self.scope_for_flavor(flavor, query)
+          scope.count
+        end
+
+        # perform a bulk update of a set of jobs
+        # action is :hold, :unhold, or :destroy
+        # to specify the jobs to act on, either pass opts[:ids] = [list of job ids]
+        # or opts[:flavor] = <some flavor> to perform on all jobs of that flavor
+        def self.bulk_update(action, opts)
+          scope = if opts[:flavor]
+            raise("Can't bulk update failed jobs") if opts[:flavor].to_s == 'failed'
+            self.scope_for_flavor(opts[:flavor], opts[:query])
+          elsif opts[:ids]
+            self.where(:id => opts[:ids])
+          end
+
+          return 0 unless scope
+
+          case action.to_s
+          when 'hold'
+            scope.update_all(:locked_by => ON_HOLD_LOCKED_BY, :locked_at => db_time_now, :attempts => ON_HOLD_COUNT)
+          when 'unhold'
+            now = db_time_now
+            scope.update_all(["locked_by = NULL, locked_at = NULL, attempts = 0, run_at = (CASE WHEN run_at > ? THEN run_at ELSE ? END), failed_at = NULL", now, now])
+          when 'destroy'
+            scope.delete_all
+          end
+        end
+
+        # returns a list of hashes { :tag => tag_name, :count => current_count }
+        # in descending count order
+        # flavor is :current or :all
+        def self.tag_counts(flavor,
+                            limit,
+                            offset = 0)
+          raise(ArgumentError, "invalid flavor: #{flavor}") unless %w(current all).include?(flavor.to_s)
+          scope = case flavor.to_s
+            when 'current'
+              self.current
+            when 'all'
+              self
+            end
+
+          scope = scope.group(:tag).offset(offset).limit(limit)
+          scope.order("COUNT(tag) DESC").count.map { |t,c| { :tag => t, :count => c } }
+        end
+
+        def self.get_and_lock_next_available(worker_name,
+                                             queue = Delayed::Settings.queue,
+                                             min_priority = nil,
+                                             max_priority = nil)
+
+          check_queue(queue)
+          check_priorities(min_priority, max_priority)
+
+          loop do
+            jobs = find_available(Settings.fetch_batch_size, queue, min_priority, max_priority)
+            return nil if jobs.empty?
+            if Settings.select_random_from_batch
+              jobs = jobs.sort_by { rand }
+            end
+            job = jobs.detect do |job|
+              job.lock_exclusively!(worker_name)
+            end
+            return job if job
+          end
+        end
+
+        def self.find_available(limit,
+                                queue = Delayed::Settings.queue,
+                                min_priority = nil,
+                                max_priority = nil)
+          all_available(queue, min_priority, max_priority).limit(limit).to_a
+        end
+
+        def self.all_available(queue = Delayed::Settings.queue,
+                               min_priority = nil,
+                               max_priority = nil)
+          min_priority ||= Delayed::MIN_PRIORITY
+          max_priority ||= Delayed::MAX_PRIORITY
+
+          check_queue(queue)
+          check_priorities(min_priority, max_priority)
+
+          self.ready_to_run.
+              where(:priority => min_priority..max_priority, :queue => queue).
+              by_priority
+        end
+
+        # used internally by create_singleton to take the appropriate lock
+        # depending on the db driver
+        def self.transaction_for_singleton(strand)
+          self.transaction do
+            connection.execute(sanitize_sql(["SELECT pg_advisory_xact_lock(half_md5_as_bigint(?))", strand]))
+            yield
+          end
+        end
+
+        # Create the job on the specified strand, but only if there aren't any
+        # other non-running jobs on that strand.
+        # (in other words, the job will still be created if there's another job
+        # on the strand but it's already running)
+        def self.create_singleton(options)
+          strand = options[:strand]
+          transaction_for_singleton(strand) do
+            job = self.where(:strand => strand, :locked_at => nil).order(:id).first
+            job || self.create(options)
+          end
+        end
+
+        # Lock this job for this worker.
+        # Returns true if we have the lock, false otherwise.
+        #
+        # It's important to note that for performance reasons, this method does
+        # not re-check the strand constraints -- so you could manually lock a
+        # job using this method that isn't the next to run on its strand.
+        def lock_exclusively!(worker)
+          now = self.class.db_time_now
+          # We don't own this job so we will update the locked_by name and the locked_at
+          affected_rows = self.class.where("id=? AND locked_at IS NULL AND run_at<=?", self, now).update_all(:locked_at => now, :locked_by => worker)
+          if affected_rows == 1
+            mark_as_locked!(now, worker)
+            return true
+          else
+            return false
+          end
+        end
+
+        def mark_as_locked!(time, worker)
+          self.locked_at    = time
+          self.locked_by    = worker
+        end
+
+        def create_and_lock!(worker)
+          raise "job already exists" unless new_record?
+          self.locked_at = Delayed::Job.db_time_now
+          self.locked_by = worker
+          save!
+        end
+
+        def fail!
+          attrs = self.attributes
+          attrs['original_job_id'] = attrs.delete('id')
+          attrs['failed_at'] ||= self.class.db_time_now
+          attrs.delete('next_in_strand')
+          self.class.transaction do
+            failed_job = Failed.create(attrs)
+            self.destroy
+            failed_job
+          end
+        rescue
+          # we got an error while failing the job -- we need to at least get
+          # the job out of the queue
+          self.destroy
+          # re-raise so the worker logs the error, at least
+          raise
+        end
+
+        class Failed < Job
+          include Delayed::Backend::Base
+          self.table_name = :failed_jobs
+        end
+      end
+
+    end
+  end
+end

data/lib/delayed/backend/base.rb

@@ -0,0 +1,317 @@
+module Delayed
+  module Backend
+    class DeserializationError < StandardError
+    end
+
+    class RecordNotFound < DeserializationError
+    end
+
+    module Base
+      ON_HOLD_LOCKED_BY = 'on hold'
+      ON_HOLD_COUNT = 50
+
+      def self.included(base)
+        base.extend ClassMethods
+        base.default_priority = Delayed::NORMAL_PRIORITY
+        base.before_save :initialize_defaults
+      end
+
+      module ClassMethods
+        attr_accessor :batches
+        attr_accessor :batch_enqueue_args
+        attr_accessor :default_priority
+
+        # Add a job to the queue
+        # The first argument should be an object that respond_to?(:perform)
+        # The rest should be named arguments, these keys are expected:
+        # :priority, :run_at, :queue, :strand, :singleton
+        # Example: Delayed::Job.enqueue(object, :priority => 0, :run_at => time, :queue => queue)
+        def enqueue(*args)
+          object = args.shift
+          unless object.respond_to?(:perform)
+            raise ArgumentError, 'Cannot enqueue items which do not respond to perform'
+          end
+
+          options = Settings.default_job_options.merge(args.first || {})
+          options[:priority] ||= self.default_priority
+          options[:payload_object] = object
+          options[:queue] = Delayed::Settings.queue unless options.key?(:queue)
+          options[:max_attempts] ||= Delayed::Settings.max_attempts
+          options[:source] = Marginalia::Comment.construct_comment if defined?(Marginalia) && Marginalia::Comment.components
+
+          # If two parameters are given to n_strand, the first param is used
+          # as the strand name for looking up the Setting, while the second
+          # param is appended to make a unique set of strands.
+          #
+          # For instance, you can pass ["my_job_type", # root_account.global_id]
+          # to get a set of n strands per root account, and you can apply the
+          # same default to all.
+          if options[:n_strand]
+            strand_name, ext = options.delete(:n_strand)
+
+            if ext
+              full_strand_name = "#{strand_name}/#{ext}"
+              num_strands = Delayed::Settings.num_strands.call(full_strand_name)
+            else
+              full_strand_name = strand_name
+            end
+
+            num_strands ||= Delayed::Settings.num_strands.call(strand_name)
+            num_strands = num_strands ? num_strands.to_i : 1
+
+            strand_num = num_strands > 1 ? rand(num_strands) + 1 : 1
+            full_strand_name += ":#{strand_num}" if strand_num > 1
+            options[:strand] = full_strand_name
+          end
+
+          if options[:singleton]
+            options[:strand] = options.delete :singleton
+            job = self.create_singleton(options)
+          elsif batches && options.slice(:strand, :run_at).empty?
+            batch_enqueue_args = options.slice(*self.batch_enqueue_args)
+            batches[batch_enqueue_args] << options
+            return true
+          else
+            job = self.create(options)
+          end
+
+          JobTracking.job_created(job)
+
+          job
+        end
+
+        def in_delayed_job?
+          !!Thread.current[:in_delayed_job]
+        end
+
+        def in_delayed_job=(val)
+          Thread.current[:in_delayed_job] = val
+        end
+
+        def check_queue(queue)
+          raise(ArgumentError, "queue name can't be blank") if queue.blank?
+        end
+
+        def check_priorities(min_priority, max_priority)
+          if min_priority && min_priority < Delayed::MIN_PRIORITY
+            raise(ArgumentError, "min_priority #{min_priority} can't be less than #{Delayed::MIN_PRIORITY}")
+          end
+          if max_priority && max_priority > Delayed::MAX_PRIORITY
+            raise(ArgumentError, "max_priority #{max_priority} can't be greater than #{Delayed::MAX_PRIORITY}")
+          end
+        end
+
+        # Get the current time (UTC)
+        # Note: This does not ping the DB to get the time, so all your clients
+        # must have syncronized clocks.
+        def db_time_now
+          Time.zone.now
+        end
+
+        def unlock_orphaned_jobs(pid = nil, name = nil)
+          begin
+            name ||= Socket.gethostname
+          rescue
+            return 0
+          end
+          pid_regex = pid || '(\d+)'
+          regex = Regexp.new("^#{Regexp.escape(name)}:#{pid_regex}$")
+          unlocked_jobs = 0
+          running = false if pid
+          self.running_jobs.each do |job|
+            next unless job.locked_by =~ regex
+            unless pid
+              job_pid = $1.to_i
+              running = Process.kill(0, job_pid) rescue false
+            end
+            if !running
+              unlocked_jobs += 1
+              job.reschedule("process died")
+            end
+          end
+          unlocked_jobs
+        end
+      end
+
+      def failed?
+        failed_at
+      end
+      alias_method :failed, :failed?
+
+      # Reschedule the job in the future (when a job fails).
+      # Uses an exponential scale depending on the number of failed attempts.
+      def reschedule(error = nil, time = nil)
+        begin
+          obj = payload_object
+          obj.on_failure(error) if obj && obj.respond_to?(:on_failure)
+        rescue DeserializationError
+          # don't allow a failed deserialization to prevent rescheduling
+        end
+
+        self.attempts += 1
+        if self.attempts >= (self.max_attempts || Delayed::Settings.max_attempts)
+          permanent_failure error || "max attempts reached"
+        else
+          time ||= self.reschedule_at
+          self.run_at = time
+          self.unlock
+          self.save!
+        end
+      end
+
+      def permanent_failure(error)
+        begin
+          # notify the payload_object of a permanent failure
+          obj = payload_object
+          obj.on_permanent_failure(error) if obj && obj.respond_to?(:on_permanent_failure)
+        rescue DeserializationError
+          # don't allow a failed deserialization to prevent destroying the job
+        end
+
+        # optionally destroy the object
+        destroy_self = true
+        if Delayed::Worker.on_max_failures
+          destroy_self = Delayed::Worker.on_max_failures.call(self, error)
+        end
+
+        if destroy_self
+          self.destroy
+        else
+          self.fail!
+        end
+      end
+
+      def payload_object
+        @payload_object ||= deserialize(self['handler'].untaint)
+      end
+
+      def name
+        @name ||= begin
+          payload = payload_object
+          if payload.respond_to?(:display_name)
+            payload.display_name
+          else
+            payload.class.name
+          end
+        end
+      end
+
+      def full_name
+        obj = payload_object rescue nil
+        if obj && obj.respond_to?(:full_name)
+          obj.full_name
+        else
+          name
+        end
+      end
+
+      def payload_object=(object)
+        @payload_object = object
+        self['handler'] = object.to_yaml
+        self['tag'] = if object.respond_to?(:tag)
+          object.tag
+        elsif object.is_a?(Module)
+          "#{object}.perform"
+        else
+          "#{object.class}#perform"
+        end
+      end
+
+      # Moved into its own method so that new_relic can trace it.
+      def invoke_job
+        Delayed::Job.in_delayed_job = true
+        begin
+          payload_object.perform
+        ensure
+          Delayed::Job.in_delayed_job = false
+          ::ActiveRecord::Base.clear_active_connections! unless Rails.env.test?
+        end
+      end
+
+      def batch?
+        payload_object.is_a?(Delayed::Batch::PerformableBatch)
+      end
+
+      # Unlock this job (note: not saved to DB)
+      def unlock
+        self.locked_at    = nil
+        self.locked_by    = nil
+      end
+
+      def locked?
+        !!(self.locked_at || self.locked_by)
+      end
+
+      def reschedule_at
+        new_time = self.class.db_time_now + (attempts ** 4) + 5
+        begin
+          if payload_object.respond_to?(:reschedule_at)
+            new_time = payload_object.reschedule_at(
+                                        self.class.db_time_now, attempts)
+          end
+        rescue
+          # TODO: just swallow errors from reschedule_at ?
+        end
+        new_time
+      end
+
+      def hold!
+        self.locked_by = ON_HOLD_LOCKED_BY
+        self.locked_at = self.class.db_time_now
+        self.attempts = ON_HOLD_COUNT
+        self.save!
+      end
+
+      def unhold!
+        self.locked_by = nil
+        self.locked_at = nil
+        self.attempts = 0
+        self.run_at = [self.class.db_time_now, self.run_at].max
+        self.failed_at = nil
+        self.save!
+      end
+
+      def on_hold?
+        self.locked_by == 'on hold' && self.locked_at && self.attempts == ON_HOLD_COUNT
+      end
+
+    private
+
+      ParseObjectFromYaml = /\!ruby\/\w+\:([^\s]+)/
+
+      def deserialize(source)
+        handler = nil
+        begin
+          handler = _yaml_deserialize(source)
+        rescue TypeError
+          attempt_to_load_from_source(source)
+          handler = _yaml_deserialize(source)
+        end
+
+        return handler if handler.respond_to?(:perform)
+
+        raise DeserializationError,
+          'Job failed to load: Unknown handler. Try to manually require the appropriate file.'
+      rescue TypeError, LoadError, NameError => e
+        raise DeserializationError,
+          "Job failed to load: #{e.message}. Try to manually require the required file."
+      end
+
+      def _yaml_deserialize(source)
+        YAML.respond_to?(:unsafe_load) ? YAML.unsafe_load(source) : YAML.load(source)
+      end
+
+      def attempt_to_load_from_source(source)
+        if md = ParseObjectFromYaml.match(source)
+          md[1].constantize
+        end
+      end
+
+    public
+      def initialize_defaults
+        self.queue ||= Delayed::Settings.queue
+        self.run_at ||= self.class.db_time_now
+      end
+    end
+  end
+end