Merge branch 'sh-check-replication-secondary' into 'master'

Add a health check for the state of Geo replication lag in secondary

Closes #1929

See merge request !1610

app/assets/javascripts/geo_nodes.js

@@ -13,6 +13,7 @@ class GeoNodeStatus {
     this.$el = $(el);
     this.$icon = $('.js-geo-node-icon', this.$el);
     this.$loadingIcon = $('.js-geo-node-loading', this.$el);
+    this.$dbReplicationLag = $('.js-db-replication-lag', this.$status);
     this.$healthStatus = $('.js-health-status', this.$el);
     this.$status = $('.js-geo-node-status', this.$el);
     this.$repositoriesSynced = $('.js-repositories-synced', this.$status);
@@ -36,6 +37,15 @@ class GeoNodeStatus {
     $.getJSON(this.endpoint, (status) => {
       this.setStatusIcon(status.healthy);
       this.setHealthStatus(status.healthy);
+
+      // Replication lag can be nil if the secondary isn't actually streaming
+      if (status.db_replication_lag) {
+        const parsedTime = gl.utils.prettyTime.parseSeconds(status.db_replication_lag);
+        this.$dbReplicationLag.text(gl.utils.prettyTime.stringifyTime(parsedTime));
+      } else {
+        this.$dbReplicationLag.text('UNKNOWN');
+      }
+
       this.$repositoriesSynced.html(`${status.repositories_synced_count}/${status.repositories_count} (${status.repositories_synced_in_percentage})`);
       this.$repositoriesFailed.html(status.repositories_failed_count);
       this.$lfsObjectsSynced.html(`${status.lfs_objects_synced_count}/${status.lfs_objects_count} (${status.lfs_objects_synced_in_percentage})`);

app/models/geo_node_status.rb

@@ -14,6 +14,14 @@ class GeoNodeStatus
     health.blank?
   end
 
+  def db_replication_lag
+    @db_replication_lag ||= Gitlab::Geo::HealthCheck.db_replication_lag
+  end
+
+  def db_replication_lag=(value)
+    @db_replication_lag = value
+  end
+
   def repositories_count
     @repositories_count ||= repositories.count
   end

app/serializers/geo_node_status_entity.rb

@@ -14,6 +14,8 @@ class GeoNodeStatusEntity < Grape::Entity
     number_to_percentage(node.attachments_synced_in_percentage, precision: 2)
   end
 
+  expose :db_replication_lag
+
   expose :lfs_objects_count
   expose :lfs_objects_synced_count
   expose :lfs_objects_synced_in_percentage do |node|

app/services/geo/node_status_service.rb

@@ -5,6 +5,7 @@ module Geo
 
     KEYS = %w(
       health
+      db_replication_lag
       repositories_count
       repositories_synced_count
       repositories_failed_count

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

@@ -45,6 +45,10 @@
                     %span.help-block
                       Health Status:
                       %span.js-health-status
+                %p
+                  %span.help-block
+                    Database replication lag:
+                    %strong.node-info.js-db-replication-lag
                 %p
                   %span.help-block
                     Repositories synced:

lib/api/entities.rb

@@ -947,6 +947,7 @@ module API
 
     class GeoNodeStatus < Grape::Entity
       expose :id
+      expose :db_replication_lag
       expose :health
       expose :healthy?, as: :healthy
       expose :repositories_count

lib/gitlab/geo/health_check.rb

@@ -7,6 +7,7 @@ module Gitlab
         return '' unless Gitlab::Geo.secondary?
         return 'The Geo database configuration file is missing.' unless Gitlab::Geo.geo_database_configured?
         return 'The Geo node has a database that is not configured for streaming replication with the primary node.' unless self.database_secondary?
+        return 'The Geo node does not appear to be replicating data from the primary node.' unless self.db_replication_lag.present?
 
         database_version  = self.get_database_version.to_i
         migration_version = self.get_migration_version.to_i
@@ -54,12 +55,24 @@ module Gitlab
       end
 
       def self.database_secondary?
-        raise NotImplementedError unless Gitlab::Database.postgresql?
-
         ActiveRecord::Base.connection.execute('SELECT pg_is_in_recovery()')
           .first
           .fetch('pg_is_in_recovery') == 't'
       end
+
+      def self.db_replication_lag
+        # Obtain the replication lag in seconds
+        ActiveRecord::Base.connection.execute('
+          SELECT CASE
+                 WHEN pg_last_xlog_receive_location() = pg_last_xlog_replay_location()
+                  THEN 0
+                 ELSE
+                  EXTRACT (EPOCH FROM now() - pg_last_xact_replay_timestamp())::INTEGER
+                 END
+                 AS replication_lag')
+          .first
+          .fetch('replication_lag')
+      end
     end
   end
 end

spec/fixtures/api/schemas/geo_node_status.json

@@ -8,6 +8,7 @@
     "attachments_synced_count",
     "lfs_objects_count",
     "lfs_objects_synced_count",
+    "db_replication_lag",
     "repositories_count",
     "repositories_failed_count",
     "repositories_synced_count"
@@ -19,6 +20,7 @@
     "attachments_count": { "type": "integer" },
     "attachments_synced_count": { "type": "integer" },
     "attachments_synced_in_percentage": { "type": "string" },
+    "db_replication_lag": { "type": ["integer", "null"] },
     "lfs_objects_count": { "type": "integer" },
     "lfs_objects_synced_count": { "type": "integer" },
     "lfs_objects_synced_in_percentage": { "type": "string" },

spec/lib/gitlab/geo/health_check_spec.rb

@@ -12,6 +12,7 @@ describe Gitlab::Geo::HealthCheck, :postgresql do
       allow(described_class).to receive(:database_secondary?).and_return(true)
       allow(described_class).to receive(:get_database_version).and_return('20170101')
       allow(described_class).to receive(:get_migration_version).and_return('20170201')
+      allow(described_class).to receive(:db_replication_lag).and_return(0)
 
       message = subject.perform_checks
 
@@ -27,8 +28,18 @@ describe Gitlab::Geo::HealthCheck, :postgresql do
 
     it 'returns an error when database is not configured for streaming replication' do
       allow(Gitlab::Geo).to receive(:secondary?) { true }
+      allow(Gitlab::Geo).to receive(:configured?) { true }
       allow(Gitlab::Database).to receive(:postgresql?) { true }
-      allow(ActiveRecord::Base).to receive_message_chain(:connection, :execute, :first, :fetch) { 'f' }
+      allow(described_class).to receive(:database_secondary?) { false }
+
+      expect(subject.perform_checks).not_to be_blank
+    end
+
+    it 'returns an error when streaming replication is not working' do
+      allow(Gitlab::Geo).to receive(:secondary?) { true }
+      allow(Gitlab::Geo).to receive(:configured?) { true }
+      allow(Gitlab::Database).to receive(:postgresql?) { true }
+      allow(described_class).to receive(:database_secondary?) { false }
 
       expect(subject.perform_checks).to include('not configured for streaming replication')
     end
@@ -42,6 +53,7 @@ describe Gitlab::Geo::HealthCheck, :postgresql do
     it 'returns an error when Geo database version does not match the latest migration version' do
       allow(described_class).to receive(:database_secondary?).and_return(true)
       allow(subject).to receive(:get_database_version) { 1 }
+      allow(described_class).to receive(:db_replication_lag).and_return(0)
 
       expect(subject.perform_checks).to match(/Current Geo database version \([0-9]+\) does not match latest migration \([0-9]+\)/)
     end
@@ -49,9 +61,17 @@ describe Gitlab::Geo::HealthCheck, :postgresql do
     it 'returns an error when latest migration version does not match the Geo database version' do
       allow(described_class).to receive(:database_secondary?).and_return(true)
       allow(subject).to receive(:get_migration_version) { 1 }
+      allow(described_class).to receive(:db_replication_lag).and_return(0)
 
       expect(subject.perform_checks).to match(/Current Geo database version \([0-9]+\) does not match latest migration \([0-9]+\)/)
     end
+
+    it 'returns an error when replication lag is not present' do
+      allow(described_class).to receive(:database_secondary?).and_return(true)
+      allow(described_class).to receive(:db_replication_lag).and_return(nil)
+
+      expect(subject.perform_checks).to match(/The Geo node does not appear to be replicating data from the primary node/)
+    end
   end
 
   describe 'MySQL checks' do

spec/models/geo_node_status_spec.rb

@@ -98,6 +98,14 @@ describe GeoNodeStatus do
     end
   end
 
+  describe '#db_replication_lag' do
+    it 'returns the set replication lag' do
+      allow(Gitlab::Geo::HealthCheck).to receive(:db_replication_lag).and_return(1000)
+
+      expect(subject.db_replication_lag).to eq(1000)
+    end
+  end
+
   describe '#lfs_objects_synced_in_percentage' do
     let(:lfs_object_project) { create(:lfs_objects_project, project: project_1) }
 
@@ -164,6 +172,7 @@ describe GeoNodeStatus do
 
   context 'when no values are available' do
     it 'returns 0 for each attribute' do
+      allow(Gitlab::Geo::HealthCheck).to receive(:db_replication_lag).and_return(nil)
       subject.attachments_count = nil
       subject.attachments_synced_count = nil
       subject.lfs_objects_count = nil
@@ -172,6 +181,7 @@ describe GeoNodeStatus do
       subject.repositories_synced_count = nil
       subject.repositories_failed_count = nil
 
+      expect(subject.db_replication_lag).to be_nil
       expect(subject.repositories_count).to be_zero
       expect(subject.repositories_synced_count).to be_zero
       expect(subject.repositories_synced_in_percentage).to be_zero

spec/services/geo/node_status_service_spec.rb

@@ -25,6 +25,7 @@ describe Geo::NodeStatusService do
 
     it 'parses a 200 response' do
       data = { health: 'OK',
+               db_replication_lag: 0,
                repositories_count: 10,
                repositories_synced_count: 1,
                repositories_failed_count: 2,