GitLab Next

In b89457d9 we refactored some code and lost one important bit from the LFS query: b89457d9

When

def fdw_find_unsynced(except_file_ids:)
      fdw_all.joins("LEFT OUTER JOIN file_registry
                                          ON file_registry.file_id = #{fdw_table}.id
                                         AND file_registry.file_type = 'lfs'")
        .geo_syncable
        .where(file_registry: { id: nil })
        .where.not(id: except_file_ids)

become:

    def lfs_objects_unsynced(except_file_ids:)
      fdw_geo_node
        .lfs_objects
        .syncable
        .missing_file_registry
        .id_not_in(except_file_ids)

The query lost the join condition that matches file_type = 'lfs'. This makes the LFS sync stop adding new items to be synchronized.

I'm flagging this as a regression from %12.1 but there is a chance this affected even previous versions (last 11.x). In %12.1 we removed the legacy queries, which made this affect any customer from that version ahead.

---

current code from lfs_objects_unsynced results in the following query:

SELECT "gitlab_secondary"."lfs_objects"."id"
FROM "gitlab_secondary"."lfs_objects"
         LEFT OUTER JOIN "file_registry" ON "gitlab_secondary"."lfs_objects"."id" = "file_registry"."file_id"
WHERE ("gitlab_secondary"."lfs_objects"."file_store" = 1 OR "gitlab_secondary"."lfs_objects"."file_store" IS NULL)
  AND "file_registry"."id" IS NULL
LIMIT 25;

what we should be doing instead:

SELECT "gitlab_secondary"."lfs_objects"."id"
FROM "gitlab_secondary"."lfs_objects"
         LEFT OUTER JOIN "file_registry" ON ("gitlab_secondary"."lfs_objects"."id" = "file_registry"."file_id" AND
                                             file_registry.file_type = 'lfs')
WHERE ("gitlab_secondary"."lfs_objects"."file_store" = 1 OR "gitlab_secondary"."lfs_objects"."file_store" IS NULL)
  AND "file_registry"."id" IS NULL
LIMIT 25;