Merge branch '3532-configurable-geo-max-capacity' into 'master'

Make the maximum capacity of Geo backfill operations configurable

Closes #3532

See merge request gitlab-org/gitlab-ee!3107

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

@@ -29,3 +29,17 @@
     = hidden_field_tag "#{form.object_name}[namespace_ids]", geo_node.namespace_ids.join(","), class: 'js-geo-node-namespaces', data: { selected: node_namespaces_options(geo_node.namespaces).to_json }
     .help-block
       #{ s_("Choose which groups you wish to replicate to this secondary node. Leave blank to replicate all.") }
+
+.form-group.js-hide-if-geo-primary{ class: ('hidden' unless geo_node.secondary?) }
+  = form.label :repos_max_capacity, s_('Geo|Repository sync capacity'), class: 'control-label'
+  .col-sm-10
+    = form.number_field :repos_max_capacity, class: 'form-control', min: 0
+    .help-block
+      #{ s_('Control the maximum concurrency of repository backfill for this secondary node') }
+
+.form-group.js-hide-if-geo-primary{ class: ('hidden' unless geo_node.secondary?) }
+  = form.label :files_max_capacity, s_('Geo|File sync capacity'), class: 'control-label'
+  .col-sm-10
+    = form.number_field :files_max_capacity, class: 'form-control', min: 0
+    .help-block
+      #{ s_('Control the maximum concurrency of LFS/attachment backfill for this secondary node') }

app/workers/geo/base_scheduler_worker.rb

@@ -5,7 +5,6 @@ module Geo
 
     DB_RETRIEVE_BATCH_SIZE = 1000
     LEASE_TIMEOUT = 60.minutes
-    MAX_CAPACITY = 10
     RUN_TIME = 60.minutes.to_i
 
     attr_reader :pending_resources, :scheduled_jobs, :start_time, :loops
@@ -18,7 +17,7 @@ module Geo
     # The scheduling works as the following:
     #
     # 1. Load a batch of IDs that we need to schedule (DB_RETRIEVE_BATCH_SIZE) into a pending list.
-    # 2. Schedule them so that at most MAX_CAPACITY are running at once.
+    # 2. Schedule them so that at most `max_capacity` are running at once.
     # 3. When a slot frees, schedule another job.
     # 4. When we have drained the pending list, load another batch into memory, and schedule the
     #    remaining jobs, excluding ones in progress.
@@ -82,7 +81,7 @@ module Geo
     end
 
     def max_capacity
-      MAX_CAPACITY
+      raise NotImplementedError
     end
 
     def run_time

app/workers/geo/file_download_dispatch_worker.rb

@@ -2,6 +2,10 @@ module Geo
   class FileDownloadDispatchWorker < Geo::BaseSchedulerWorker
     private
 
+    def max_capacity
+      current_node.files_max_capacity
+    end
+
     def schedule_job(object_db_id, object_type)
       job_id = GeoFileDownloadWorker.perform_async(object_type, object_db_id)
 

app/workers/geo/repository_sync_worker.rb

 module Geo
   class RepositorySyncWorker < Geo::BaseSchedulerWorker
-    MAX_CAPACITY = 25
-
     private
 
     def max_capacity
-      MAX_CAPACITY
+      current_node.repos_max_capacity
     end
 
     def schedule_job(project_id)

changelogs/unreleased-ee/3532-configurable-geo-max-capacity.yml

+---
+title: Make the maximum capacity of Geo backfill operations configurable
+merge_request: 3107
+author:
+type: added

db/migrate/20171010140746_geo_configurable_max_capacities.rb

+class GeoConfigurableMaxCapacities < ActiveRecord::Migration
+  include Gitlab::Database::MigrationHelpers
+
+  DOWNTIME = false
+
+  disable_ddl_transaction!
+
+  def up
+    add_column_with_default :geo_nodes, :files_max_capacity, :integer, allow_null: false, default: 10
+    add_column_with_default :geo_nodes, :repos_max_capacity, :integer, allow_null: false, default: 25
+  end
+
+  def down
+    remove_column :geo_nodes, :files_max_capacity, :integer
+    remove_column :geo_nodes, :repos_max_capacity, :integer
+  end
+end

db/schema.rb

@@ -11,7 +11,7 @@
 #
 # It's strongly recommended that you check this file into your version control system.
 
-ActiveRecord::Schema.define(version: 20171006091000) do
+ActiveRecord::Schema.define(version: 20171010140746) do
 
   # These are extensions that must be enabled in order to support this database
   enable_extension "plpgsql"
@@ -758,6 +758,8 @@ ActiveRecord::Schema.define(version: 20171006091000) do
     t.string "encrypted_secret_access_key"
     t.string "encrypted_secret_access_key_iv"
     t.string "clone_url_prefix"
+    t.integer "files_max_capacity", default: 10, null: false
+    t.integer "repos_max_capacity", default: 25, null: false
   end
 
   add_index "geo_nodes", ["access_key"], name: "index_geo_nodes_on_access_key", using: :btree

doc/user/admin_area/geo_nodes.md

@@ -5,13 +5,44 @@ For more information about setting up GitLab Geo, read the
 
 When you're done, you can navigate to **Admin area ➔ Geo nodes** (`/admin/geo_nodes`).
 
-In the following table you can see what all these settings mean:
+## Common settings
+
+All Geo nodes have the following settings:
 
 | Setting | Description |
-| --------- | ----------- |
+| --------| ----------- |
 | Primary | This marks a Geo Node as primary. There can be only one primary, make sure that you first add the primary node and then all the others. |
-| URL       | Your instance's full URL, in the same way it is configured in  `/etc/gitlab/gitlab.rb` (Omnibus GitLab installations) or `gitlab.yml` (source based installations). |
+| URL     | The instance's full URL, in the same way it is configured in  `/etc/gitlab/gitlab.rb` (Omnibus GitLab installations) or `gitlab.yml` (source based installations). |
+
+The node you're reading from is indicated with a green `Current node` label, and
+the primary is given a blue `Primary` label. Remember that you can only make
+changes on the primary!
+
+## Secondary node settings
+
+Secondaries have a number of additional settings available:
+
+| Setting                  | Description|
+|--------------------------|------------|
 | Public Key               | The SSH public key of the user that your GitLab instance runs on (unless changed, should be the user `git`). |
+| Groups to replicate      | Enable Geo selective sync for this secondary - only the selected groups will be synchronized. |
+| Repository sync capacity | Number of concurrent requests this secondary will make to the primary when backfilling repositories. |
+| File sync capacity       | Number of concurrent requests this secondary will make to the primary when backfilling files. |
+
+## Geo backfill
+
+Secondaries are notified of changes to repositories and files by the primary,
+and will always attempt to synchronize those changes as quickly as possible.
+
+Backfill is the act of populating the secondary with repositories and files that
+existed *before* the secondary was added to the database. Since there may be
+extremely large numbers of repositories and files, it's infeasible to attempt to
+download them all at once, so GitLab places an upper limit on the concurrency of
+these operations.
 
-A primary node will have a star right next to it to distinguish from the
-secondaries.
+How long the backfill takes is a function of the maximum concurrency, but higher
+values place more strain on the primary node. From [GitLab 10.2](https://gitlab.com/gitlab-org/gitlab-ee/merge_requests/3107),
+the limits are configurable - if your primary node has lots of surplus capacity,
+you can increase the values to complete backfill in a shorter time. If it's
+under heavy load and backfill is reducing its availability for normal requests,
+you can decrease them.

spec/models/geo_node_spec.rb

 require 'spec_helper'
 
 describe GeoNode, type: :model do
+  using RSpec::Parameterized::TableSyntax
   include ::EE::GeoHelpers
 
   let(:new_node) { create(:geo_node, schema: 'https', host: 'localhost', port: 3000, relative_url_root: 'gitlab') }
@@ -29,28 +30,22 @@ describe GeoNode, type: :model do
   context 'default values' do
     let(:gitlab_host) { 'gitlabhost' }
 
-    before do
-      allow(Gitlab.config.gitlab).to receive(:host) { gitlab_host }
-    end
-
-    it 'defines a default schema' do
-      expect(empty_node.schema).to eq('http')
-    end
-
-    it 'defines a default host' do
-      expect(empty_node.host).to eq(gitlab_host)
-    end
-
-    it 'defines a default port' do
-      expect(empty_node.port).to eq(80)
+    where(:attribute, :value) do
+      :schema             | 'http'
+      :host               | 'gitlabhost'
+      :port               | 80
+      :relative_url_root  | ''
+      :primary            | false
+      :repos_max_capacity | 25
+      :files_max_capacity | 10
     end
 
-    it 'defines a default relative_url_root' do
-      expect(empty_node.relative_url_root).to eq('')
+    with_them do
+      before do
+        allow(Gitlab.config.gitlab).to receive(:host) { gitlab_host }
       end
 
-    it 'defines a default primary flag' do
-      expect(empty_node.primary).to eq(false)
+      it { expect(empty_node[attribute]).to eq(value) }
     end
   end
 

spec/workers/geo/file_download_dispatch_worker_spec.rb

@@ -61,7 +61,7 @@ describe Geo::FileDownloadDispatchWorker, :geo do
     # 2. We send 2, wait for 1 to finish, and then send again.
     it 'attempts to load a new batch without pending downloads' do
       stub_const('Geo::BaseSchedulerWorker::DB_RETRIEVE_BATCH_SIZE', 5)
-      stub_const('Geo::BaseSchedulerWorker::MAX_CAPACITY', 2)
+      secondary.update!(files_max_capacity: 2)
       allow_any_instance_of(::Gitlab::Geo::Transfer).to receive(:download_from_primary).and_return(100)
 
       avatar = fixture_file_upload(Rails.root.join('spec/fixtures/dk.png'))

spec/workers/geo/repository_sync_worker_spec.rb

@@ -102,7 +102,7 @@ describe Geo::RepositorySyncWorker, :postgresql do
 
       before do
         allow_any_instance_of(described_class).to receive(:db_retrieve_batch_size).and_return(2) # Must be >1 because of the Geo::BaseSchedulerWorker#interleave
-        allow_any_instance_of(described_class).to receive(:max_capacity).and_return(3) # Must be more than db_retrieve_batch_size
+        secondary.update!(repos_max_capacity: 3) # Must be more than db_retrieve_batch_size
         allow_any_instance_of(Project).to receive(:ensure_repository).and_raise(Gitlab::Shell::Error.new('foo'))
         allow_any_instance_of(Geo::ProjectSyncWorker).to receive(:sync_wiki?).and_return(false)
         allow_any_instance_of(Geo::RepositorySyncService).to receive(:expire_repository_caches)