Treat registry as SSOT for Job Artifacts

FileDownloadDispatchWorker looks only at the registry table
for things that have never been synced.

In order for the counts to be correct, we have to distinguish
between "never synced" and "failed" with "retry_count" being
"nil" for the former and "not nil" for the latter.

Feature flag is "geo_job_artifact_registry_ssot_sync".

ee/app/finders/geo/job_artifact_registry_finder.rb

@@ -31,53 +31,81 @@ module Geo
       Ci::JobArtifact.not_expired
     end
 
+    # Returns Geo::JobArtifactRegistry records that have never been synced.
+    #
+    # Does not care about selective sync, because it considers the Registry
+    # table to be the single source of truth. The contract is that other
+    # processes need to ensure that the table only contains records that should
+    # be synced.
+    #
+    # Any registries that have ever been synced that currently need to be
+    # resynced will be handled by other find methods (like
+    # #find_retryable_failed_registries)
+    #
+    # You can pass a list with `except_ids:` so you can exclude items you
+    # already scheduled but haven't finished and aren't persisted to the database yet
+    #
+    # @param [Integer] batch_size used to limit the results returned
+    # @param [Array<Integer>] except_ids ids that will be ignored from the query
+    # rubocop:disable CodeReuse/ActiveRecord
+    def find_never_synced_registries(batch_size:, except_ids: [])
+      Geo::JobArtifactRegistry
+        .never
+        .model_id_not_in(except_ids)
+        .limit(batch_size)
+    end
+    # rubocop:enable CodeReuse/ActiveRecord
+
+    # Deprecated in favor of the process using
+    # #find_missing_registry_ids and #find_never_synced_registries
+    #
     # Find limited amount of non replicated job artifacts.
     #
-    # You can pass a list with `except_artifact_ids:` so you can exclude items you
+    # You can pass a list with `except_ids:` so you can exclude items you
     # already scheduled but haven't finished and aren't persisted to the database yet
     #
     # TODO: Alternative here is to use some sort of window function with a cursor instead
     #       of simply limiting the query and passing a list of items we don't want
     #
     # @param [Integer] batch_size used to limit the results returned
-    # @param [Array<Integer>] except_artifact_ids ids that will be ignored from the query
+    # @param [Array<Integer>] except_ids ids that will be ignored from the query
     # rubocop: disable CodeReuse/ActiveRecord
-    def find_unsynced(batch_size:, except_artifact_ids: [])
+    def find_unsynced(batch_size:, except_ids: [])
       job_artifacts
         .not_expired
         .missing_job_artifact_registry
-        .id_not_in(except_artifact_ids)
+        .id_not_in(except_ids)
         .limit(batch_size)
     end
     # rubocop: enable CodeReuse/ActiveRecord
 
     # rubocop: disable CodeReuse/ActiveRecord
-    def find_migrated_local(batch_size:, except_artifact_ids: [])
+    def find_migrated_local(batch_size:, except_ids: [])
       all_job_artifacts
         .inner_join_job_artifact_registry
         .with_files_stored_remotely
-        .id_not_in(except_artifact_ids)
+        .id_not_in(except_ids)
         .limit(batch_size)
     end
     # rubocop: enable CodeReuse/ActiveRecord
 
     # rubocop: disable CodeReuse/ActiveRecord
-    def find_retryable_failed_registries(batch_size:, except_artifact_ids: [])
+    def find_retryable_failed_registries(batch_size:, except_ids: [])
       Geo::JobArtifactRegistry
         .failed
         .retry_due
-        .artifact_id_not_in(except_artifact_ids)
+        .model_id_not_in(except_ids)
         .limit(batch_size)
     end
     # rubocop: enable CodeReuse/ActiveRecord
 
     # rubocop: disable CodeReuse/ActiveRecord
-    def find_retryable_synced_missing_on_primary_registries(batch_size:, except_artifact_ids: [])
+    def find_retryable_synced_missing_on_primary_registries(batch_size:, except_ids: [])
       Geo::JobArtifactRegistry
         .synced
         .missing_on_primary
         .retry_due
-        .artifact_id_not_in(except_artifact_ids)
+        .model_id_not_in(except_ids)
         .limit(batch_size)
     end
     # rubocop: enable CodeReuse/ActiveRecord

ee/app/models/geo/job_artifact_registry.rb

@@ -3,15 +3,17 @@
 class Geo::JobArtifactRegistry < Geo::BaseRegistry
   include Geo::Syncable
 
-  def self.artifact_id_in(ids)
-    where(artifact_id: ids)
-  end
+  MODEL_CLASS = ::Ci::JobArtifact
+  MODEL_FOREIGN_KEY = :artifact_id
 
-  def self.artifact_id_not_in(ids)
-    where.not(artifact_id: ids)
-  end
+  scope :never, -> { where(success: false, retry_count: nil) }
 
-  def self.pluck_artifact_key
-    where(nil).pluck(:artifact_id)
+  def self.failed
+    if Feature.enabled?(:geo_job_artifact_registry_ssot_sync)
+      where(success: false).where.not(retry_count: nil)
+    else
+      # Would do `super` except it doesn't work with an included scope
+      where(success: false)
+    end
   end
 end

ee/app/workers/geo/file_download_dispatch_worker/job_artifact_job_finder.rb

@@ -4,12 +4,22 @@ module Geo
   class FileDownloadDispatchWorker # rubocop:disable Scalability/IdempotentWorker
     class JobArtifactJobFinder < JobFinder # rubocop:disable Scalability/IdempotentWorker
       RESOURCE_ID_KEY = :artifact_id
-      EXCEPT_RESOURCE_IDS_KEY = :except_artifact_ids
+      EXCEPT_RESOURCE_IDS_KEY = :except_ids
       FILE_SERVICE_OBJECT_TYPE = :job_artifact
 
       def registry_finder
         @registry_finder ||= Geo::JobArtifactRegistryFinder.new(current_node_id: Gitlab::Geo.current_node.id)
       end
+
+      def find_unsynced_jobs(batch_size:)
+        if Feature.enabled?(:geo_job_artifact_registry_ssot_sync)
+          convert_registry_relation_to_job_args(
+            registry_finder.find_never_synced_registries(find_batch_params(batch_size))
+          )
+        else
+          super
+        end
+      end
     end
   end
 end

ee/app/workers/geo/migrated_local_files_clean_up_worker.rb

@@ -68,7 +68,7 @@ module Geo
     def find_migrated_local_job_artifacts_ids(batch_size:)
       return [] unless job_artifacts_object_store_enabled?
 
-      job_artifacts_finder.find_migrated_local(batch_size: batch_size, except_artifact_ids: scheduled_file_ids(:job_artifact))
+      job_artifacts_finder.find_migrated_local(batch_size: batch_size, except_ids: scheduled_file_ids(:job_artifact))
                           .pluck(Geo::Fdw::Ci::JobArtifact.arel_table[:id])
                           .map { |id| ['job_artifact', id] }
     end

ee/lib/ee/gitlab/cleanup/orphan_job_artifact_files_batch.rb

@@ -29,7 +29,7 @@ module EE
         end
 
         def lost_and_found_geo_registries
-          @lost_and_found_geo_registries ||= ::Geo::JobArtifactRegistry.artifact_id_in(lost_and_found_ids)
+          @lost_and_found_geo_registries ||= ::Geo::JobArtifactRegistry.model_id_in(lost_and_found_ids)
         end
       end
     end

ee/spec/factories/geo/job_artifact_registry.rb

@@ -5,6 +5,16 @@ FactoryBot.define do
     sequence(:artifact_id)
     success { true }
 
+    trait :failed do
+      success { false }
+      retry_count { 1 }
+    end
+
+    trait :never_synced do
+      success { false }
+      retry_count { nil }
+    end
+
     trait :with_artifact do
       transient do
         artifact_type { nil } # e.g. :archive, :metadata, :trace ...

ee/spec/finders/geo/job_artifact_registry_finder_spec.rb

@@ -107,13 +107,13 @@ describe Geo::JobArtifactRegistryFinder, :geo_fdw do
 
     describe '#count_failed' do
       before do
-        create(:geo_job_artifact_registry, artifact_id: job_artifact_synced_project.id, success: false)
+        create(:geo_job_artifact_registry, :failed, artifact_id: job_artifact_synced_project.id)
         create(:geo_job_artifact_registry, artifact_id: job_artifact_unsynced_project.id)
-        create(:geo_job_artifact_registry, artifact_id: job_artifact_broken_storage_1.id, success: false)
-        create(:geo_job_artifact_registry, artifact_id: job_artifact_expired_synced_project.id, success: false)
-        create(:geo_job_artifact_registry, artifact_id: job_artifact_expired_broken_storage.id, success: false)
-        create(:geo_job_artifact_registry, artifact_id: job_artifact_remote_synced_project.id, success: false)
-        create(:geo_job_artifact_registry, artifact_id: job_artifact_remote_broken_storage.id, success: false)
+        create(:geo_job_artifact_registry, :failed, artifact_id: job_artifact_broken_storage_1.id)
+        create(:geo_job_artifact_registry, :failed, artifact_id: job_artifact_expired_synced_project.id)
+        create(:geo_job_artifact_registry, :failed, artifact_id: job_artifact_expired_broken_storage.id)
+        create(:geo_job_artifact_registry, :failed, artifact_id: job_artifact_remote_synced_project.id)
+        create(:geo_job_artifact_registry, :failed, artifact_id: job_artifact_remote_broken_storage.id)
       end
 
       context 'without selective sync' do
@@ -229,6 +229,21 @@ describe Geo::JobArtifactRegistryFinder, :geo_fdw do
   end
 
   context 'finds all the things' do
+    describe '#find_never_synced_registries' do
+      let!(:registry_job_artifact_1) { create(:geo_job_artifact_registry, :never_synced, artifact_id: job_artifact_synced_project.id) }
+      let!(:registry_job_artifact_2) { create(:geo_job_artifact_registry, :never_synced, artifact_id: job_artifact_unsynced_project.id) }
+      let!(:registry_job_artifact_3) { create(:geo_job_artifact_registry, artifact_id: job_artifact_broken_storage_1.id) }
+      let!(:registry_job_artifact_4) { create(:geo_job_artifact_registry, :failed, artifact_id: job_artifact_broken_storage_2.id) }
+      let!(:registry_job_artifact_remote_1) { create(:geo_job_artifact_registry, :never_synced, artifact_id: job_artifact_remote_synced_project.id) }
+
+      it 'returns registries for LFS objects that have never been synced' do
+        registries = subject.find_never_synced_registries(batch_size: 10)
+
+        expect(registries).to match_ids(registry_job_artifact_1, registry_job_artifact_2, registry_job_artifact_remote_1)
+      end
+    end
+
+
     describe '#find_unsynced' do
       before do
         create(:geo_job_artifact_registry, artifact_id: job_artifact_synced_project.id, success: false)
@@ -238,7 +253,7 @@ describe Geo::JobArtifactRegistryFinder, :geo_fdw do
 
       context 'without selective sync' do
         it 'returns job artifacts without an entry on the tracking database, ignoring expired ones' do
-          job_artifacts = subject.find_unsynced(batch_size: 10, except_artifact_ids: [job_artifact_unsynced_project.id])
+          job_artifacts = subject.find_unsynced(batch_size: 10, except_ids: [job_artifact_unsynced_project.id])
 
           expect(job_artifacts).to match_ids(job_artifact_remote_synced_project, job_artifact_remote_unsynced_project,
                                              job_artifact_broken_storage_2)
@@ -285,7 +300,7 @@ describe Geo::JobArtifactRegistryFinder, :geo_fdw do
       end
 
       it 'returns job artifacts excluding ones from the exception list' do
-        job_artifacts = subject.find_migrated_local(batch_size: 10, except_artifact_ids: [job_artifact_remote_synced_project.id])
+        job_artifacts = subject.find_migrated_local(batch_size: 10, except_ids: [job_artifact_remote_synced_project.id])
 
         expect(job_artifacts).to match_ids(job_artifact_remote_unsynced_project, job_artifact_remote_broken_storage)
       end
@@ -321,7 +336,7 @@ describe Geo::JobArtifactRegistryFinder, :geo_fdw do
         let(:secondary) { create(:geo_node, :local_storage_only) }
 
         it 'returns job artifacts excluding ones from the exception list' do
-          job_artifacts = subject.find_migrated_local(batch_size: 10, except_artifact_ids: [job_artifact_remote_synced_project.id])
+          job_artifacts = subject.find_migrated_local(batch_size: 10, except_ids: [job_artifact_remote_synced_project.id])
 
           expect(job_artifacts).to match_ids(job_artifact_remote_unsynced_project, job_artifact_remote_broken_storage)
         end

ee/spec/models/geo_node_status_spec.rb

@@ -342,16 +342,40 @@ describe GeoNodeStatus, :geo, :geo_fdw do
   end
 
   describe '#job_artifacts_failed_count' do
-    it 'counts failed job artifacts' do
-      # These should be ignored
-      create(:geo_upload_registry, :failed)
-      create(:geo_upload_registry, :avatar, :failed)
-      create(:geo_upload_registry, :attachment, :failed)
-      create(:geo_job_artifact_registry, :with_artifact, success: true)
+    context 'when geo_job_artifact_registry_ssot_sync is disabled' do
+      before do
+        stub_feature_flags(geo_job_artifact_registry_ssot_sync: false)
+      end
 
-      create(:geo_job_artifact_registry, :with_artifact, success: false)
+      it 'counts failed job artifacts' do
+        # These should be ignored
+        create(:geo_upload_registry, :failed)
+        create(:geo_upload_registry, :avatar, :failed)
+        create(:geo_upload_registry, :attachment, :failed)
+        create(:geo_job_artifact_registry, :with_artifact, success: true)
 
-      expect(subject.job_artifacts_failed_count).to eq(1)
+        create(:geo_job_artifact_registry, :with_artifact, success: false)
+
+        expect(subject.job_artifacts_failed_count).to eq(1)
+      end
+    end
+
+    context 'when geo_job_artifact_registry_ssot_sync is enabled' do
+      before do
+        stub_feature_flags(geo_job_artifact_registry_ssot_sync: true)
+      end
+
+      it 'counts failed job artifacts' do
+        # These should be ignored
+        create(:geo_upload_registry, :failed)
+        create(:geo_upload_registry, :avatar, :failed)
+        create(:geo_upload_registry, :attachment, :failed)
+        create(:geo_job_artifact_registry, :with_artifact, success: true)
+
+        create(:geo_job_artifact_registry, :with_artifact, :failed)
+
+        expect(subject.job_artifacts_failed_count).to eq(1)
+      end
     end
   end