Remove orphaned label links

On GitLab.com, there are over 2 million orphaned label links out of
a total of 13 million. These orphaned label links can cause
quiet failures, such as unexpected nil values in ExportCsvWorker.

Closes gitlab-org/gitlab-ee#7482

2 files changed, 44 insertions, 0 deletions

diff --git a/db/post_migrate/20180906051323_remove_orphaned_label_links.rb b/db/post_migrate/20180906051323_remove_orphaned_label_links.rb
new file mode 100644
index 00000000000..b56b74f483e
--- /dev/null
+++ b/db/post_migrate/20180906051323_remove_orphaned_label_links.rb
@@ -0,0 +1,43 @@
+# frozen_string_literal: true
+
+class RemoveOrphanedLabelLinks < ActiveRecord::Migration
+  include Gitlab::Database::MigrationHelpers
+
+  DOWNTIME = false
+
+  disable_ddl_transaction!
+
+  class LabelLinks < ActiveRecord::Base
+    self.table_name = 'label_links'
+    include EachBatch
+
+    def self.orphaned
+      where('NOT EXISTS ( SELECT 1 FROM labels WHERE labels.id = label_links.label_id )')
+    end
+  end
+
+  def up
+    # Some of these queries can take up to 10 seconds to run on GitLab.com,
+    # which is pretty close to our 15 second statement timeout. To ensure a
+    # smooth deployment procedure we disable the statement timeouts for this
+    # migration, just in case.
+    disable_statement_timeout do
+      # On GitLab.com there are over 2,000,000 orphaned label links. On
+      # staging, removing 100,000 rows generated a max replication lag of 6.7
+      # MB. In total, removing all these rows will only generate about 136 MB
+      # of data, so it should be safe to do this.
+      LabelLinks.orphaned.each_batch(of: 100_000) do |batch|
+        batch.delete_all
+      end
+    end
+
+    add_concurrent_foreign_key(:label_links, :labels, column: :label_id, on_delete: :cascade)
+  end
+
+  def down
+    # There is no way to restore orphaned label links.
+    if foreign_key_exists?(:label_links, column: :label_id)
+      remove_foreign_key(:label_links, column: :label_id)
+    end
+  end
+end
diff --git a/db/schema.rb b/db/schema.rb
index a417d0bc70a..98ae9f7b131 100644
--- a/db/schema.rb
+++ b/db/schema.rb
@@ -2357,6 +2357,7 @@ ActiveRecord::Schema.define(version: 20180906101639) do
   add_foreign_key "issues", "users", column: "author_id", name: "fk_05f1e72feb", on_delete: :nullify
   add_foreign_key "issues", "users", column: "closed_by_id", name: "fk_c63cbf6c25", on_delete: :nullify
   add_foreign_key "issues", "users", column: "updated_by_id", name: "fk_ffed080f01", on_delete: :nullify
+  add_foreign_key "label_links", "labels", name: "fk_d97dd08678", on_delete: :cascade
   add_foreign_key "label_priorities", "labels", on_delete: :cascade
   add_foreign_key "label_priorities", "projects", on_delete: :cascade
   add_foreign_key "labels", "namespaces", column: "group_id", on_delete: :cascade