Add partial index on projects for index-only scans.

This helps with queries that get project ids based on the - comparably
rare - visibility levels 10 and 20. For these, postgres can now leverage
the partial index for a index-only scan to improve performance.

Example queries:
SELECT id FROM projects WHERE visibility_level IN (10,20)
SELECT id FROM projects WHERE visibility_level IN (10)

For MySQL, this results in a full index on id because MySQL omits the
WHERE clause. That is, the index is a duplicate of the primary key
basically.

1 files changed, 21 insertions, 0 deletions

diff --git a/db/migrate/20180213131630_add_partial_index_to_projects_for_index_only_scans.rb b/db/migrate/20180213131630_add_partial_index_to_projects_for_index_only_scans.rb
new file mode 100644
index 00000000000..c8ebabbe627
--- /dev/null
+++ b/db/migrate/20180213131630_add_partial_index_to_projects_for_index_only_scans.rb
@@ -0,0 +1,21 @@
+class AddPartialIndexToProjectsForIndexOnlyScans < ActiveRecord::Migration
+  include Gitlab::Database::MigrationHelpers
+
+  DOWNTIME = false
+  INDEX_NAME = 'index_projects_on_id_partial_for_visibility'
+
+  disable_ddl_transaction!
+
+  # Adds a partial index to leverage index-only scans when looking up project ids
+  def up
+    unless index_exists?(:projects, :id, name: INDEX_NAME)
+      add_concurrent_index :projects, :id, name: INDEX_NAME, unique: true, where: 'visibility_level IN (10,20)'
+    end
+  end
+
+  def down
+    if index_exists?(:projects, :id, name: INDEX_NAME)
+      remove_concurrent_index :projects, :id, name: INDEX_NAME, unique: true, where: 'visibility_level IN (10,20)'
+    end
+  end
+end