Convert migration to populate latest merge request ID into a background migration

This is to smear updates over a few hours to avoid causing excessive
replication lag as seen in https://gitlab.com/gitlab-com/infrastructure/issues/3235.

1 files changed, 29 insertions, 0 deletions

diff --git a/db/post_migrate/20171026082505_schedule_merge_request_latest_merge_request_diff_id_migrations.rb b/db/post_migrate/20171026082505_schedule_merge_request_latest_merge_request_diff_id_migrations.rb
new file mode 100644
index 00000000000..7a63382cc6d
--- /dev/null
+++ b/db/post_migrate/20171026082505_schedule_merge_request_latest_merge_request_diff_id_migrations.rb
@@ -0,0 +1,29 @@
+class ScheduleMergeRequestLatestMergeRequestDiffIdMigrations < ActiveRecord::Migration
+  include Gitlab::Database::MigrationHelpers
+
+  DOWNTIME = false
+  BATCH_SIZE = 50_000
+  MIGRATION = 'PopulateMergeRequestsLatestMergeRequestDiffId'
+
+  disable_ddl_transaction!
+
+  class MergeRequest < ActiveRecord::Base
+    self.table_name = 'merge_requests'
+
+    include ::EachBatch
+  end
+
+  # On GitLab.com, we saw that we generated about 500,000 dead tuples over 5 minutes.
+  # To keep replication lag from ballooning, we'll aim for 50,000 updates over 5 minutes.
+  #
+  # Assuming that there are 5 million rows affected (which is more than on
+  # GitLab.com), and that each batch of 50,000 rows takes up to 5 minutes, then
+  # we can migrate all the rows in 8.5 hours.
+  def up
+    MergeRequest.where(latest_merge_request_diff_id: nil).each_batch(of: BATCH_SIZE) do |relation, index|
+      range = relation.pluck('MIN(id)', 'MAX(id)').first
+
+      BackgroundMigrationWorker.perform_in(index * 5.minutes, MIGRATION, range)
+    end
+  end
+end