Replace events index with partial one

This improves the index size by removing all records with `group_id` from the index. On GitLab.com this means reducing the size from 9 GB to 8kb (as long as the column stays mostly null). Relates to https://gitlab.com/gitlab-org/gitlab-ce/merge_requests/32826#note_214690051

Replace events index with partial one
This improves the index size by removing all records with `group_id` from the index. On GitLab.com this means reducing the size from 9 GB to 8kb (as long as the column stays mostly null). Relates to https://gitlab.com/gitlab-org/gitlab-ce/merge_requests/32826#note_214690051
c98e5d92 · Andreas Brandl · 0844ec02 · c98e5d92 · c98e5d92 · c98e5d92
--- a/changelogs/unreleased/ab-partial-events-index.yml
+++ b/changelogs/unreleased/ab-partial-events-index.yml
+---
+title: Replace events index with partial one
+merge_request: 32874
+author:
+type: performance
--- a/db/migrate/20190910103144_replace_events_index_on_group_id_with_partial_index.rb
+++ b/db/migrate/20190910103144_replace_events_index_on_group_id_with_partial_index.rb
+# frozen_string_literal: true
+
+class ReplaceEventsIndexOnGroupIdWithPartialIndex < ActiveRecord::Migration[5.2]
+  include Gitlab::Database::MigrationHelpers
+
+  DOWNTIME = false
+
+  disable_ddl_transaction!
+
+  def up
+    add_concurrent_index(:events, :group_id, where: 'group_id IS NOT NULL', name: 'index_events_on_group_id_partial')
+    remove_concurrent_index_by_name(:events, 'index_events_on_group_id')
+  end
+
+  def down
+    add_concurrent_index(:events, :group_id, name: 'index_events_on_group_id')
+    remove_concurrent_index_by_name(:events, 'index_events_on_group_id_partial')
+  end
+end
--- a/db/schema.rb
+++ b/db/schema.rb
@@ -1321,7 +1321,7 @@
    t.index ["action"], name: "index_events_on_action"
    t.index ["author_id", "project_id"], name: "index_events_on_author_id_and_project_id"
    t.index ["created_at", "author_id"], name: "analytics_index_events_on_created_at_and_author_id"
-    t.index ["group_id"], name: "index_events_on_group_id"
+    t.index ["group_id"], name: "index_events_on_group_id_partial", where: "(group_id IS NOT NULL)"
    t.index ["project_id", "created_at"], name: "index_events_on_project_id_and_created_at"
    t.index ["project_id", "id"], name: "index_events_on_project_id_and_id"
    t.index ["target_type", "target_id"], name: "index_events_on_target_type_and_target_id"