Merge branch 'sidekiq-job-throttling' into 'master'

Allow certain Sidekiq jobs to be throttled

## What does this MR do?

Allows certain slow running Sidekiq jobs to be throttled. It is disabled by default and can be enabled via the Application Settings. 

![Screen_Shot_2016-11-04_at_4.51.24_PM](/uploads/a1f1d24c693fcdb278602765cd404d94/Screen_Shot_2016-11-04_at_4.51.24_PM.png)

## Does this MR meet the acceptance criteria?

- [x] [CHANGELOG](https://gitlab.com/gitlab-org/gitlab-ce/blob/master/CHANGELOG.md) entry added
- [x] [Documentation created/updated](https://gitlab.com/gitlab-org/gitlab-ce/blob/master/doc/development/doc_styleguide.md)
- Tests
  - [x] Added for this feature/bug
  - [x] All builds are passing

## What are the relevant issue numbers?

Related to #23352

See merge request !7292

Gemfile

@@ -137,6 +137,7 @@ gem 'acts-as-taggable-on', '~> 4.0'
 gem 'sidekiq', '~> 4.2'
 gem 'sidekiq-cron', '~> 0.4.0'
 gem 'redis-namespace', '~> 1.5.2'
+gem 'sidekiq-limit_fetch', '~> 3.4'
 
 # HTTP requests
 gem 'httparty', '~> 0.13.3'

Gemfile.lock

@@ -685,6 +685,8 @@ GEM
       redis-namespace (>= 1.5.2)
       rufus-scheduler (>= 2.0.24)
       sidekiq (>= 4.0.0)
+    sidekiq-limit_fetch (3.4.0)
+      sidekiq (>= 4)
     simplecov (0.12.0)
       docile (~> 1.1.0)
       json (>= 1.8, < 3)
@@ -961,6 +963,7 @@ DEPENDENCIES
   shoulda-matchers (~> 2.8.0)
   sidekiq (~> 4.2)
   sidekiq-cron (~> 0.4.0)
+  sidekiq-limit_fetch (~> 3.4)
   simplecov (= 0.12.0)
   slack-notifier (~> 1.2.0)
   spinach-rails (~> 0.2.1)

app/controllers/admin/application_settings_controller.rb

@@ -117,6 +117,8 @@ class Admin::ApplicationSettingsController < Admin::ApplicationController
       :send_user_confirmation_email,
       :container_registry_token_expire_delay,
       :enabled_git_access_protocol,
+      :sidekiq_throttling_enabled,
+      :sidekiq_throttling_factor,
       :housekeeping_enabled,
       :housekeeping_bitmaps_enabled,
       :housekeeping_incremental_repack_period,
@@ -125,7 +127,8 @@ class Admin::ApplicationSettingsController < Admin::ApplicationController
       repository_storages: [],
       restricted_visibility_levels: [],
       import_sources: [],
-      disabled_oauth_sign_in_sources: []
+      disabled_oauth_sign_in_sources: [],
+      sidekiq_throttling_queues: []
     )
   end
 end

app/helpers/application_settings_helper.rb

@@ -100,4 +100,8 @@ module ApplicationSettingsHelper
 
     options_for_select(options, @application_setting.repository_storages)
   end
+
+  def sidekiq_queue_options_for_select
+    options_for_select(Sidekiq::Queue.all.map(&:name), @application_setting.sidekiq_throttling_queues)
+  end
 end

app/models/application_setting.rb

@@ -19,6 +19,7 @@ class ApplicationSetting < ActiveRecord::Base
   serialize :domain_whitelist, Array
   serialize :domain_blacklist, Array
   serialize :repository_storages
+  serialize :sidekiq_throttling_queues, Array
 
   cache_markdown_field :sign_in_text
   cache_markdown_field :help_page_text
@@ -85,6 +86,15 @@ class ApplicationSetting < ActiveRecord::Base
             presence: { message: 'Domain blacklist cannot be empty if Blacklist is enabled.' },
             if: :domain_blacklist_enabled?
 
+  validates :sidekiq_throttling_factor,
+            numericality: { greater_than: 0, less_than: 1 },
+            presence: { message: 'Throttling factor cannot be empty if Sidekiq Throttling is enabled.' },
+            if: :sidekiq_throttling_enabled?
+
+  validates :sidekiq_throttling_queues,
+            presence: { message: 'Queues to throttle cannot be empty if Sidekiq Throttling is enabled.' },
+            if: :sidekiq_throttling_enabled?
+
   validates :housekeeping_incremental_repack_period,
             presence: true,
             numericality: { only_integer: true, greater_than: 0 }
@@ -180,6 +190,7 @@ class ApplicationSetting < ActiveRecord::Base
       container_registry_token_expire_delay: 5,
       repository_storages: ['default'],
       user_default_external: false,
+      sidekiq_throttling_enabled: false,
       housekeeping_enabled: true,
       housekeeping_bitmaps_enabled: true,
       housekeeping_incremental_repack_period: 10,

app/views/admin/application_settings/_form.html.haml

@@ -283,6 +283,31 @@
           The amount of points to store in a single UDP packet. More points
           results in fewer but larger UDP packets being sent.
 
+  %fieldset
+    %legend Background Jobs
+    %p
+      These settings require a restart to take effect.
+    .form-group
+      .col-sm-offset-2.col-sm-10
+        .checkbox
+          = f.label :sidekiq_throttling_enabled do
+            = f.check_box :sidekiq_throttling_enabled
+            Enable Sidekiq Job Throttling
+          .help-block
+            Limit the amount of resources slow running jobs are assigned.
+    .form-group
+      = f.label :sidekiq_throttling_queues, 'Sidekiq queues to throttle', class: 'control-label col-sm-2'
+      .col-sm-10
+        = f.select :sidekiq_throttling_queues, sidekiq_queue_options_for_select, { include_hidden: false }, multiple: true, class: 'select2 select-wide', data: { field: 'sidekiq_throttling_queues' }
+        .help-block
+          Choose which queues you wish to throttle.
+    .form-group
+      = f.label :sidekiq_throttling_factor, 'Throttling Factor', class: 'control-label col-sm-2'
+      .col-sm-10
+        = f.number_field :sidekiq_throttling_factor, class: 'form-control', min: '0.01', max: '0.99', step: '0.01'
+        .help-block
+          The factor by which the queues should be throttled. A value between 0.0 and 1.0, exclusive.
+
   %fieldset
     %legend Spam and Anti-bot Protection
     .form-group

changelogs/unreleased/sidekiq-job-throttling.yml

@@ -0,0 +1,4 @@
+---
+title: Added ability to throttle Sidekiq Jobs
+merge_request: 7292
+author:

config/initializers/sidekiq.rb

@@ -29,6 +29,8 @@ Sidekiq.configure_server do |config|
   end
   Sidekiq::Cron::Job.load_from_hash! cron_jobs
 
+  Gitlab::SidekiqThrottler.execute!
+
   # Database pool should be at least `sidekiq_concurrency` + 2
   # For more info, see: https://github.com/mperham/sidekiq/blob/master/4.0-Upgrade.md
   config = ActiveRecord::Base.configurations[Rails.env] ||

db/migrate/20161103191444_add_sidekiq_throttling_to_application_settings.rb

@@ -0,0 +1,31 @@
+# See http://doc.gitlab.com/ce/development/migration_style_guide.html
+# for more information on how to write migrations for GitLab.
+
+class AddSidekiqThrottlingToApplicationSettings < ActiveRecord::Migration
+  include Gitlab::Database::MigrationHelpers
+
+  # Set this constant to true if this migration requires downtime.
+  DOWNTIME = false
+
+  # When a migration requires downtime you **must** uncomment the following
+  # constant and define a short and easy to understand explanation as to why the
+  # migration requires downtime.
+  # DOWNTIME_REASON = ''
+
+  # When using the methods "add_concurrent_index" or "add_column_with_default"
+  # you must disable the use of transactions as these methods can not run in an
+  # existing transaction. When using "add_concurrent_index" make sure that this
+  # method is the _only_ method called in the migration, any other changes
+  # should go in a separate migration. This ensures that upon failure _only_ the
+  # index creation fails and can be retried or reverted easily.
+  #
+  # To disable transactions uncomment the following line and remove these
+  # comments:
+  # disable_ddl_transaction!
+
+  def change
+    add_column :application_settings, :sidekiq_throttling_enabled, :boolean, default: false
+    add_column :application_settings, :sidekiq_throttling_queues, :string
+    add_column :application_settings, :sidekiq_throttling_factor, :decimal
+  end
+end

db/schema.rb

@@ -98,6 +98,9 @@ ActiveRecord::Schema.define(version: 20161109150329) do
     t.text "help_page_text_html"
     t.text "shared_runners_text_html"
     t.text "after_sign_up_text_html"
+    t.boolean "sidekiq_throttling_enabled", default: false
+    t.string "sidekiq_throttling_queues"
+    t.decimal "sidekiq_throttling_factor"
     t.boolean "housekeeping_enabled", default: true, null: false
     t.boolean "housekeeping_bitmaps_enabled", default: true, null: false
     t.integer "housekeeping_incremental_repack_period", default: 10, null: false

doc/administration/operations.md

@@ -1,6 +1,7 @@
 # GitLab operations
 
 - [Sidekiq MemoryKiller](operations/sidekiq_memory_killer.md)
+- [Sidekiq Job throttling](operations/sidekiq_job_throttling.md)
 - [Cleaning up Redis sessions](operations/cleaning_up_redis_sessions.md)
 - [Understanding Unicorn and unicorn-worker-killer](operations/unicorn.md)
 - [Moving repositories to a new location](operations/moving_repositories.md)

doc/administration/operations/sidekiq_job_throttling.md

@@ -0,0 +1,33 @@
+# Sidekiq Job throttling
+
+> Note: Introduced with GitLab 8.14
+
+When your GitLab installation needs to handle tens of thousands of background
+jobs, it can be convenient to throttle queues that do not need to be executed
+immediately, e.g. long running jobs like Pipelines, thus allowing jobs that do
+need to be executed immediately to have access to more resources.
+
+In order to accomplish this, you can limit the amount of workers that certain
+slow running queues can have available. This is what we call Sidekiq Job
+Throttling. Depending on your infrastructure, you might have different slow
+running queues, which is why you can choose which queues you want to throttle 
+and by how much you want to throttle them.
+
+These settings are available in the Application Settings of your GitLab
+installation.
+
+![Sidekiq Job Throttling](img/sidekiq_job_throttling.png)
+
+The throttle factor determines the maximum number of workers a queue can run on.
+This value gets multiplied by `:concurrency` value set in the Sidekiq settings
+and rounded up to the closest full integer.
+
+So, for example, you set the `:concurrency` to 25 and the `Throttling factor` to
+0.1, the maximum workers assigned to the selected queues would be 3.
+
+```ruby
+queue_limit = (factor * Sidekiq.options[:concurrency]).ceil
+```
+
+After enabling the job throttling, you will need to restart your GitLab
+instance, in order for the changes to take effect.

lib/gitlab/current_settings.rb

@@ -23,6 +23,10 @@ module Gitlab
       settings || fake_application_settings
     end
 
+    def sidekiq_throttling_enabled?
+      current_application_settings.sidekiq_throttling_enabled
+    end
+
     def fake_application_settings
       OpenStruct.new(
         default_projects_limit: Settings.gitlab['default_projects_limit'],
@@ -50,6 +54,7 @@ module Gitlab
         repository_checks_enabled: true,
         container_registry_token_expire_delay: 5,
         user_default_external: false,
+        sidekiq_throttling_enabled: false,
       )
     end
 

lib/gitlab/sidekiq_throttler.rb

@@ -0,0 +1,23 @@
+module Gitlab
+  class SidekiqThrottler
+    class << self
+      def execute!
+        if Gitlab::CurrentSettings.sidekiq_throttling_enabled?
+          Gitlab::CurrentSettings.current_application_settings.sidekiq_throttling_queues.each do |queue|
+            Sidekiq::Queue[queue].limit = queue_limit
+          end
+        end
+      end
+
+      private
+
+      def queue_limit
+        @queue_limit ||=
+          begin
+            factor = Gitlab::CurrentSettings.current_application_settings.sidekiq_throttling_factor
+            (factor * Sidekiq.options[:concurrency]).ceil
+          end
+      end
+    end
+  end
+end

spec/lib/gitlab/sidekiq_throttler_spec.rb

@@ -0,0 +1,28 @@
+require 'spec_helper'
+
+describe Gitlab::SidekiqThrottler do
+  before do
+    Sidekiq.options[:concurrency] = 35
+
+    stub_application_setting(
+      sidekiq_throttling_enabled: true,
+      sidekiq_throttling_factor: 0.1,
+      sidekiq_throttling_queues: %w[build project_cache]
+    )
+  end
+
+  describe '#execute!' do
+    it 'sets limits on the selected queues' do
+      Gitlab::SidekiqThrottler.execute!
+
+      expect(Sidekiq::Queue['build'].limit).to eq 4
+      expect(Sidekiq::Queue['project_cache'].limit).to eq 4
+    end
+
+    it 'does not set limits on other queues' do
+      Gitlab::SidekiqThrottler.execute!
+
+      expect(Sidekiq::Queue['merge'].limit).to be_nil
+    end
+  end
+end