Fix: Container repository geo syncs stuck in started state

We add Container Registry class to SyncTimeoutCronWorker
as all the newer SSF registries.

Changelog: fixed
EE: true

ee/app/models/geo/container_repository_registry.rb

@@ -13,6 +13,7 @@ class Geo::ContainerRepositoryRegistry < Geo::BaseRegistry
   scope :never_attempted_sync, -> { with_state(:pending).where(last_synced_at: nil) }
   scope :retry_due, -> { where(arel_table[:retry_at].eq(nil).or(arel_table[:retry_at].lt(Time.current))) }
   scope :synced, -> { with_state(:synced) }
+  scope :sync_timed_out, -> { with_state(:started).where("last_synced_at < ?", Geo::ContainerRepositorySyncService::LEASE_TIMEOUT.ago) }
 
   state_machine :state, initial: :pending do
     state :started
@@ -38,24 +39,42 @@ class Geo::ContainerRepositoryRegistry < Geo::BaseRegistry
     end
   end
 
-  def self.find_registries_needs_sync_again(batch_size:, except_ids: [])
+  class << self
+    include Delay
+
+    def find_registries_needs_sync_again(batch_size:, except_ids: [])
       super.order(Gitlab::Database.nulls_first_order(:last_synced_at))
     end
 
-  def self.delete_for_model_ids(container_repository_ids)
+    def delete_for_model_ids(container_repository_ids)
       where(container_repository_id: container_repository_ids).delete_all
 
       container_repository_ids
     end
 
-  def self.pluck_container_repository_key
+    def pluck_container_repository_key
       where(nil).pluck(:container_repository_id)
     end
 
-  def self.replication_enabled?
+    def replication_enabled?
       Gitlab.config.geo.registry_replication.enabled
     end
 
+    # Fail syncs for records which started syncing a long time ago
+    def fail_sync_timeouts
+      attrs = {
+        state: :failed,
+        last_sync_failure: "Sync timed out after #{Geo::ContainerRepositorySyncService::LEASE_TIMEOUT} hours",
+        retry_count: 1,
+        retry_at: next_retry_time(1)
+      }
+
+      sync_timed_out.all.each_batch do |relation|
+        relation.update_all(attrs)
+      end
+    end
+  end
+
   def fail_sync!(message, error)
     new_retry_count = retry_count + 1
 

ee/app/workers/geo/sync_timeout_cron_worker.rb

@@ -20,6 +20,8 @@ module Geo
       Gitlab::Geo.enabled_replicator_classes.each do |replicator_class|
         replicator_class.fail_sync_timeouts
       end
+
+      ::Geo::ContainerRepositoryRegistry.fail_sync_timeouts
     end
   end
 end

ee/spec/models/geo/container_repository_registry_spec.rb

@@ -223,4 +223,16 @@ RSpec.describe Geo::ContainerRepositoryRegistry, :geo do
       expect(Geo::ContainerRepositoryRegistry.replication_enabled?).to be_falsey
     end
   end
+
+  describe '.fail_sync_timeouts' do
+    it 'marks started records as failed if they are expired' do
+      record1 = create(:container_repository_registry, :sync_started, last_synced_at: 9.hours.ago)
+      record2 = create(:container_repository_registry, :sync_started, last_synced_at: 1.hour.ago) # not yet expired
+
+      described_class.fail_sync_timeouts
+
+      expect(record1.reload.state).to eq "failed"
+      expect(record2.reload.state).to eq "started"
+    end
+  end
 end

ee/spec/workers/geo/sync_timeout_cron_worker_spec.rb

@@ -4,11 +4,12 @@ require 'spec_helper'
 
 RSpec.describe Geo::SyncTimeoutCronWorker, :geo do
   describe '#perform' do
-    it 'calls fail_sync_timeouts on enabled Replicators' do
+    it 'calls fail_sync_timeouts' do
       replicator = double('replicator')
 
       expect(replicator).to receive(:fail_sync_timeouts)
       expect(Gitlab::Geo).to receive(:enabled_replicator_classes).and_return([replicator])
+      expect(Geo::ContainerRepositoryRegistry).to receive(:fail_sync_timeouts)
 
       described_class.new.perform
     end