Geo: Don't retry repositories or files until everything has been backfilled

app/workers/geo/file_download_dispatch_worker.rb

@@ -13,23 +13,27 @@ module Geo
     end
 
     def load_pending_resources
-      unsynced = find_unsynced_objects
-      failed = find_failed_objects
+      resources = find_unsynced_objects(batch_size: db_retrieve_batch_size)
+      remaining_capacity = db_retrieve_batch_size - resources.count
 
-      interleave(unsynced, failed)
+      if remaining_capacity.zero?
+        resources
+      else
+        resources + find_failed_objects(batch_size: remaining_capacity)
+      end
     end
 
-    def find_unsynced_objects
-      lfs_object_ids = find_lfs_object_ids
-      upload_objects_ids = find_upload_object_ids
+    def find_unsynced_objects(batch_size:)
+      lfs_object_ids = find_lfs_object_ids(batch_size: batch_size)
+      upload_objects_ids = find_upload_object_ids(batch_size: batch_size)
 
       interleave(lfs_object_ids, upload_objects_ids)
     end
 
-    def find_failed_objects
+    def find_failed_objects(batch_size:)
       Geo::FileRegistry
         .failed
-        .limit(db_retrieve_batch_size)
+        .limit(batch_size)
         .pluck(:file_id, :file_type)
     end
 
@@ -37,7 +41,7 @@ module Geo
       current_node.restricted_project_ids
     end
 
-    def find_lfs_object_ids
+    def find_lfs_object_ids(batch_size:)
       # Selective project replication adds a wrinkle to FDW queries, so
       # we fallback to the legacy version for now.
       relation =
@@ -48,12 +52,12 @@ module Geo
         end
 
       relation
-        .limit(db_retrieve_batch_size)
+        .limit(batch_size)
         .pluck(:id)
         .map { |id| [id, :lfs] }
     end
 
-    def find_upload_object_ids
+    def find_upload_object_ids(batch_size:)
       # Selective project replication adds a wrinkle to FDW queries, so
       # we fallback to the legacy version for now.
       relation =
@@ -64,7 +68,7 @@ module Geo
         end
 
       relation
-        .limit(db_retrieve_batch_size)
+        .limit(batch_size)
         .pluck(:id, :uploader)
         .map { |id, uploader| [id, uploader.sub(/Uploader\z/, '').underscore] }
     end
@@ -84,7 +88,6 @@ module Geo
 
       Geo::Fdw::Upload.joins("LEFT OUTER JOIN file_registry ON file_registry.file_id = #{fdw_table}.id AND file_registry.file_type IN (#{obj_types})")
         .where('file_registry.file_id IS NULL')
-        .order(created_at: :desc)
     end
 
     def legacy_find_upload_object_ids

app/workers/geo/repository_sync_worker.rb

@@ -13,24 +13,28 @@ module Geo
     end
 
     def load_pending_resources
-      project_ids_not_synced = find_project_ids_not_synced
-      project_ids_updated_recently = find_project_ids_updated_recently
+      resources = find_project_ids_not_synced(batch_size: db_retrieve_batch_size)
+      remaining_capacity = db_retrieve_batch_size - resources.size
 
-      interleave(project_ids_not_synced, project_ids_updated_recently)
+      if remaining_capacity.zero?
+        resources
+      else
+        resources + find_project_ids_updated_recently(batch_size: remaining_capacity)
+      end
     end
 
-    def find_project_ids_not_synced
+    def find_project_ids_not_synced(batch_size:)
       healthy_shards_restriction(current_node.unsynced_projects)
         .reorder(last_repository_updated_at: :desc)
-        .limit(db_retrieve_batch_size)
+        .limit(batch_size)
         .pluck(:id)
     end
 
-    def find_project_ids_updated_recently
+    def find_project_ids_updated_recently(batch_size:)
       current_node.project_registries
                   .dirty
                   .order(Gitlab::Database.nulls_first_order(:last_repository_synced_at, :desc))
-                  .limit(db_retrieve_batch_size)
+                  .limit(batch_size)
                   .pluck(:project_id)
     end
 

changelogs/unreleased-ee/3810-geo-disable-retries.yml

+---
+title: 'Geo: Don''t retry repositories or files until everything has been backfilled'
+merge_request: 3182
+author:
+type: changed

spec/workers/geo/repository_sync_worker_spec.rb

@@ -101,10 +101,14 @@ describe Geo::RepositorySyncWorker, :postgresql do
       let!(:project_list) { create_list(:project, 4, :random_last_repository_updated_at) }
 
       before do
+        # Neither of these are needed for this spec
+        unsynced_project.destroy
+        project_in_synced_group.destroy
+
         allow_any_instance_of(described_class).to receive(:db_retrieve_batch_size).and_return(2) # Must be >1 because of the Geo::BaseSchedulerWorker#interleave
         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::ProjectRegistry).to receive(:wiki_sync_due?).and_return(false)
         allow_any_instance_of(Geo::RepositorySyncService).to receive(:expire_repository_caches)
       end