Feature: Added centralized ActiveJob locking based on lock key. This prevents unnecessary executing/queuing of unique jobs multiple times.

app/jobs/application_job.rb

@@ -10,6 +10,10 @@ class ApplicationJob < ActiveJob::Base
   # This is a workaround to sync ActiveJob#executions to Delayed::Job#attempts
   # until we resolve this dependency.
   after_enqueue do |job|
+    # skip update of `attempts` attribute if job wasn't queued because of ActiveJobLock
+    #(another job with same lock key got queued before this job could be retried)
+    next if job.provider_job_id.blank?
+
     # update the column right away without loading Delayed::Job record
     # see: https://stackoverflow.com/a/34264580
     Delayed::Job.where(id: job.provider_job_id).update_all(attempts: job.executions) # rubocop:disable Rails/SkipsModelValidations

app/jobs/checks_kb_client_notification_job.rb

@@ -1,4 +1,11 @@
 class ChecksKbClientNotificationJob < ApplicationJob
+  include HasActiveJobLock
+
+  def lock_key
+    # "ChecksKbClientNotificationJob/KnowledgeBase::Answer/42/destroy"
+    "#{self.class.name}/#{arguments[0]}/#{arguments[1]}/#{arguments[2]}"
+  end
+
   def perform(klass_name, id, event)
     object = klass_name.constantize.find_by(id: id)
     return if object.blank?

app/jobs/concerns/has_active_job_lock.rb

@@ -0,0 +1,118 @@
+module HasActiveJobLock
+  extend ActiveSupport::Concern
+
+  included do
+    before_enqueue do |job| # rubocop:disable Style/SymbolProc
+      job.ensure_active_job_lock_for_enqueue!
+    end
+
+    around_perform do |job, block|
+      job.mark_active_job_lock_as_started
+
+      block.call
+    ensure
+      job.release_active_job_lock!
+    end
+  end
+
+  # Defines the lock key for the current job to prevent execution of jobs with the same key.
+  # This is by default the name of the ActiveJob class.
+  # If you're in the situation where you need to have a lock_key based on
+  # the given arguments you can overwrite this method in your job and access
+  # them via `arguments`. See ActiveJob::Core for more (e.g. queue).
+  #
+  # @example
+  #  # default
+  #  job = UniqueActiveJob.new
+  #  job.lock_key
+  #  # => "UniqueActiveJob"
+  #
+  # @example
+  #  # with lock_key: "#{self.class.name}/#{arguments[0]}/#{arguments[1]}"
+  #  job = SearchIndexJob.new('User', 42)
+  #  job.lock_key
+  #  # => "SearchIndexJob/User/42"
+  #
+  # return [String]
+  def lock_key
+    self.class.name
+  end
+
+  def mark_active_job_lock_as_started
+    release_active_job_lock_cache
+
+    in_active_job_lock_transaction do
+      # a perform_now job doesn't require any locking
+      return if active_job_lock.blank?
+      return if !active_job_lock.of?(self)
+
+      # a perform_later job started to perform and will be marked as such
+      active_job_lock.touch # rubocop:disable Rails/SkipsModelValidations
+    end
+  end
+
+  def ensure_active_job_lock_for_enqueue!
+    release_active_job_lock_cache
+
+    in_active_job_lock_transaction do
+      return if active_job_lock_for_enqueue!.present?
+
+      ActiveJobLock.create!(
+        lock_key:      lock_key,
+        active_job_id: job_id,
+      )
+    end
+  end
+
+  def release_active_job_lock!
+    # only delete lock if the current job is the one holding the lock
+    # perform_now jobs or perform_later jobs for which follow-up jobs were enqueued
+    # don't need to remove any locks
+    lock = ActiveJobLock.lock.find_by(lock_key: lock_key, active_job_id: job_id)
+
+    if !lock
+      logger.debug { "Found no ActiveJobLock for #{self.class.name} (Job ID: #{job_id}) with key '#{lock_key}'." }
+      return
+    end
+
+    logger.debug { "Deleting ActiveJobLock for #{self.class.name} (Job ID: #{job_id}) with key '#{lock_key}'." }
+    lock.destroy!
+  end
+
+  private
+
+  def in_active_job_lock_transaction
+    # re-use active DB transaction if present
+    return yield if ActiveRecord::Base.connection.open_transactions.nonzero?
+
+    # start own serializable DB transaction to prevent race conditions on DB level
+    ActiveJobLock.transaction(isolation: :serializable) do
+      yield
+    end
+  rescue ActiveRecord::RecordNotUnique
+    existing_active_job_lock!
+  end
+
+  def active_job_lock_for_enqueue!
+    return if active_job_lock.blank?
+
+    # don't enqueue perform_later jobs if a job with the same
+    # lock key exists that hasn't started to perform yet
+    existing_active_job_lock! if active_job_lock.peform_pending?
+
+    active_job_lock.tap { |lock| lock.transfer_to(self) }
+  end
+
+  def active_job_lock
+    @active_job_lock ||= ActiveJobLock.lock.find_by(lock_key: lock_key)
+  end
+
+  def release_active_job_lock_cache
+    @active_job_lock = nil
+  end
+
+  def existing_active_job_lock!
+    logger.info "Won't enqueue #{self.class.name} (Job ID: #{job_id}) because of already existing job with lock key '#{lock_key}'."
+    throw :abort
+  end
+end

app/jobs/scheduled_touch_job.rb

@@ -1,9 +1,16 @@
 class ScheduledTouchJob < ApplicationJob
-  def perform(klass_name, id)
+  include HasActiveJobLock
-    klass_name.constantize.find_by(id: id)&.touch # rubocop:disable Rails/SkipsModelValidations
+
+  def lock_key
+    # "ScheduledTouchJob/User/42"
+    "#{self.class.name}/#{arguments[0]}/#{arguments[1]}"
   end
 
   def self.touch_at(object, date)
     set(wait_until: date).perform_later(object.class.to_s, object.id)
   end
+
+  def perform(klass_name, id)
+    klass_name.constantize.find_by(id: id)&.touch # rubocop:disable Rails/SkipsModelValidations
+  end
 end

app/jobs/search_index_job.rb

@@ -1,9 +1,15 @@
 class SearchIndexJob < ApplicationJob
+  include HasActiveJobLock
 
   retry_on StandardError, attempts: 20, wait: lambda { |executions|
     executions * 10.seconds
   }
 
+  def lock_key
+    # "SearchIndexJob/User/42"
+    "#{self.class.name}/#{arguments[0]}/#{arguments[1]}"
+  end
+
   def perform(object, o_id)
     @object = object
     @o_id   = o_id

app/jobs/sla_ticket_rebuild_escalation_job.rb

@@ -1,4 +1,6 @@
 class SlaTicketRebuildEscalationJob < ApplicationJob
+  include HasActiveJobLock
+
   def perform
     Cache.delete('SLA::List::Active')
     Ticket::Escalation.rebuild_all

app/jobs/ticket_online_notification_seen_job.rb

@@ -1,4 +1,11 @@
 class TicketOnlineNotificationSeenJob < ApplicationJob
+  include HasActiveJobLock
+
+  def lock_key
+    # "TicketOnlineNotificationSeenJob/23/42"
+    "#{self.class.name}/#{arguments[0]}/#{arguments[1]}"
+  end
+
   def perform(ticket_id, user_id)
     user_id = user_id || 1
 

app/jobs/ticket_user_ticket_counter_job.rb

@@ -1,4 +1,11 @@
 class TicketUserTicketCounterJob < ApplicationJob
+  include HasActiveJobLock
+
+  def lock_key
+    # "TicketUserTicketCounterJob/23/42"
+    "#{self.class.name}/#{arguments[0]}/#{arguments[1]}"
+  end
+
   def perform(customer_id, updated_by_id)
 
     # check if update is needed

app/models/active_job_lock.rb

@@ -0,0 +1,21 @@
+class ActiveJobLock < ApplicationModel
+
+  def of?(active_job)
+    active_job.job_id == active_job_id
+  end
+
+  def peform_pending?
+    updated_at == created_at
+  end
+
+  def transfer_to(active_job)
+    logger.info "Transferring ActiveJobLock with id '#{id}' from active_job_id '#{active_job_id}' to active_job_id '#{active_job_id}'."
+
+    reset_time_stamp = Time.zone.now
+    update!(
+      active_job_id: active_job.job_id,
+      created_at:    reset_time_stamp,
+      updated_at:    reset_time_stamp
+    )
+  end
+end

db/migrate/20120101000001_create_base.rb

@@ -715,5 +715,14 @@ class CreateBase < ActiveRecord::Migration[4.2]
     add_index :http_logs, [:created_at]
     add_foreign_key :http_logs, :users, column: :created_by_id
     add_foreign_key :http_logs, :users, column: :updated_by_id
+
+    create_table :active_job_locks do |t|
+      t.string :lock_key
+      t.string :active_job_id
+
+      t.timestamps
+    end
+    add_index :active_job_locks, :lock_key, unique: true
+    add_index :active_job_locks, :active_job_id, unique: true
   end
 end

db/migrate/20191001090809_create_active_job_locks.rb

@@ -0,0 +1,15 @@
+class CreateActiveJobLocks < ActiveRecord::Migration[5.2]
+  def change
+    # return if it's a new setup
+    return if !Setting.find_by(name: 'system_init_done')
+
+    create_table :active_job_locks do |t|
+      t.string :lock_key
+      t.string :active_job_id
+
+      t.timestamps
+    end
+    add_index :active_job_locks, :lock_key, unique: true
+    add_index :active_job_locks, :active_job_id, unique: true
+  end
+end

spec/factories/active_job_lock.rb

@@ -0,0 +1,6 @@
+FactoryBot.define do
+  factory :active_job_lock do
+    lock_key { 'UniqueActiveJob' }
+    active_job_id { SecureRandom.uuid }
+  end
+end

spec/jobs/concerns/has_active_job_lock_spec.rb

@@ -0,0 +1,120 @@
+require 'rails_helper'
+
+RSpec.describe HasActiveJobLock, type: :job do
+
+  before do
+    stub_const job_class_namespace, job_class
+  end
+
+  let(:job_class_namespace) { 'UniqueActiveJob' }
+
+  let(:job_class) do
+    Class.new(ApplicationJob) do
+      include HasActiveJobLock
+
+      cattr_accessor :perform_counter, default: 0
+
+      def perform
+        self.class.perform_counter += 1
+      end
+    end
+  end
+
+  shared_examples 'handle locking of jobs' do
+    context 'performing job is present' do
+
+      before { create(:active_job_lock, lock_key: job_class.name, created_at: 1.minute.ago, updated_at: 1.second.ago) }
+
+      it 'allows enqueueing of perform_later jobs' do
+        expect { job_class.perform_later }.to have_enqueued_job(job_class).exactly(:once)
+      end
+
+      it 'allows execution of perform_now jobs' do
+        expect { job_class.perform_now }.to change(job_class, :perform_counter).by(1)
+      end
+    end
+
+    context 'enqueued job is present' do
+
+      before { job_class.perform_later }
+
+      it "won't enqueue perform_later jobs" do
+        expect { job_class.perform_later }.not_to have_enqueued_job(job_class)
+      end
+
+      it 'allows execution of perform_now jobs' do
+        expect { job_class.perform_now }.to change(job_class, :perform_counter).by(1)
+      end
+    end
+
+    context 'running perform_now job' do
+
+      let(:job_class) do
+        Class.new(super()) do
+
+          cattr_accessor :task_completed, default: false
+
+          def perform(long_running: false)
+
+            if long_running
+              sleep(0.1) until self.class.task_completed
+            end
+
+            # don't pass parameters to super method
+            super()
+          end
+        end
+      end
+
+      let!(:thread) { Thread.new { job_class.perform_now(long_running: true) } }
+
+      after do
+        job_class.task_completed = true
+        thread.join
+      end
+
+      it 'enqueues perform_later jobs' do
+        expect { job_class.perform_later }.to have_enqueued_job(job_class)
+      end
+
+      it 'allows execution of perform_now jobs' do
+        expect { job_class.perform_now }.to change(job_class, :perform_counter).by(1)
+      end
+    end
+
+    context 'dynamic lock key' do
+
+      let(:job_class) do
+        Class.new(super()) do
+
+          def lock_key
+            "#{super}/#{arguments[0]}/#{arguments[1]}"
+          end
+        end
+      end
+
+      it 'queues one job per lock key' do
+        expect do
+          2.times { job_class.perform_later('User', 23) }
+          job_class.perform_later('User', 42)
+        end.to have_enqueued_job(job_class).exactly(:twice)
+      end
+    end
+  end
+
+  include_examples 'handle locking of jobs'
+
+  context 'custom lock key' do
+
+    let(:job_class) do
+      Class.new(super()) do
+
+        def lock_key
+          'custom_lock_key'
+        end
+      end
+    end
+
+    include_examples 'handle locking of jobs'
+  end
+end

spec/requests/integration/monitoring_spec.rb

@@ -420,8 +420,8 @@ RSpec.describe 'Monitoring', type: :request do
 
       # health_check - scheduler job count
       travel 2.seconds
-      8001.times do
+      8001.times do |fake_ticket_id|
-        SearchIndexJob.perform_later('Ticket', 1)
+        SearchIndexJob.perform_later('Ticket', fake_ticket_id)
       end
       Scheduler.where(active: true).each do |local_scheduler|
         local_scheduler.last_run = Time.zone.now
@@ -579,12 +579,8 @@ RSpec.describe 'Monitoring', type: :request do
       expect('test4').to eq(json_response['name'])
       expect('Test 4').to eq(json_response['display'])
 
-      jobs = Delayed::Job.all
-
       4.times do
-        jobs.each do |job|
+        Delayed::Worker.new.work_off
-          Delayed::Worker.new.run(job)
-        end
       end
 
       # health_check
@@ -595,7 +591,7 @@ RSpec.describe 'Monitoring', type: :request do
       expect(json_response['message']).to be_truthy
       expect(json_response['issues']).to be_truthy
       expect(json_response['healthy']).to eq(false)
-      expect( json_response['message']).to eq("Failed to run background job #1 'SearchIndexJob' 4 time(s) with 4 attempt(s).")
+      expect( json_response['message']).to eq("Failed to run background job #1 'SearchIndexJob' 1 time(s) with 1 attempt(s).")
 
       # add another job
       manual_added = SearchIndexJob.perform_later('Ticket', 1)
@@ -609,10 +605,10 @@ RSpec.describe 'Monitoring', type: :request do
       expect(json_response['message']).to be_truthy
       expect(json_response['issues']).to be_truthy
       expect(json_response['healthy']).to eq(false)
-      expect( json_response['message']).to eq("Failed to run background job #1 'SearchIndexJob' 5 time(s) with 14 attempt(s).")
+      expect( json_response['message']).to eq("Failed to run background job #1 'SearchIndexJob' 2 time(s) with 11 attempt(s).")
 
       # add another job
-      dummy_class = Class.new do
+      dummy_class = Class.new(ApplicationJob) do
 
         def perform
           puts 'work work'
@@ -630,7 +626,7 @@ RSpec.describe 'Monitoring', type: :request do
       expect(json_response['message']).to be_truthy
       expect(json_response['issues']).to be_truthy
       expect(json_response['healthy']).to eq(false)
-      expect( json_response['message']).to eq("Failed to run background job #1 'Object' 1 time(s) with 5 attempt(s).;Failed to run background job #2 'SearchIndexJob' 5 time(s) with 14 attempt(s).")
+      expect( json_response['message']).to eq("Failed to run background job #1 'Object' 1 time(s) with 5 attempt(s).;Failed to run background job #2 'SearchIndexJob' 2 time(s) with 11 attempt(s).")
 
       # reset settings
       Setting.set('es_url', prev_es_config)
@@ -649,7 +645,7 @@ RSpec.describe 'Monitoring', type: :request do
       expect(json_response['message']).to be_truthy
       expect(json_response['issues']).to be_truthy
       expect(json_response['healthy']).to eq(false)
-      expect(json_response['message']).to eq("16 failing background jobs;Failed to run background job #1 'Object' 5 time(s) with 25 attempt(s).;Failed to run background job #2 'SearchIndexJob' 5 time(s) with 14 attempt(s).")
+      expect(json_response['message']).to eq("13 failing background jobs;Failed to run background job #1 'Object' 8 time(s) with 40 attempt(s).;Failed to run background job #2 'SearchIndexJob' 2 time(s) with 11 attempt(s).")
 
       # cleanup
       Delayed::Job.delete_all

spec/support/active_job.rb

@@ -9,6 +9,7 @@ module ZammadActiveJobHelper
   def clear_jobs
     enqueued_jobs.clear
     performed_jobs.clear
+    ActiveJobLock.destroy_all
   end
 end