From d9bb9f2bed41cd737f5f8b43c399ada9d80e3f61 Mon Sep 17 00:00:00 2001
From: Alexis King <lexi.lambda@gmail.com>
Date: Mon, 16 Dec 2019 03:07:03 -0600
Subject: [PATCH] Alter the type of hdb_catalog columns that reference SQL
 identifiers

As explained in the note included in the diff, this can lead to
dramatically better query plans, and it seems to be especially important
for versions of Postgres <12.
---
 server/src-rsr/initialise.sql          | 27 +++++++++++++++++++-------
 server/src-rsr/migrations/28_to_29.sql | 24 +++++++++++++++++++++++
 2 files changed, 44 insertions(+), 7 deletions(-)

diff --git a/server/src-rsr/initialise.sql b/server/src-rsr/initialise.sql
index 59c0ef2d28e..940fea21745 100644
--- a/server/src-rsr/initialise.sql
+++ b/server/src-rsr/initialise.sql
@@ -9,11 +9,24 @@ CREATE TABLE hdb_catalog.hdb_version (
 CREATE UNIQUE INDEX hdb_version_one_row
 ON hdb_catalog.hdb_version((version IS NOT NULL));
 
+/* Note [Reference system columns using type name]
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+While working on #3394, I (Alexis) discovered that Postgres seems to sometimes generate very bad
+query plans when joining against the system catalogs if we store things like table/schema names
+using type `text` rather than type `name`, the latter of which is used internally. The two types are
+compatible in the sense that Postgres will willingly widen `name` to `type` automatically, but
+`name`s are restricted to 64 bytes.
+
+Using `name` for ordinary user data would be a deep sin, but using it to store references to actual
+Postgres identifiers makes a lot of sense, so using `name` in those places is alright. And by doing
+so, we make Postgres much more likely to take advantage of certain indexes that can significantly
+improve query performance. */
+
 CREATE TABLE hdb_catalog.hdb_table
 (
-    table_schema TEXT,
-    table_name TEXT,
-    configuration JSONB,
+    table_schema name, -- See Note [Reference system columns using type name]
+    table_name name,
+    configuration jsonb,
     is_system_defined boolean default false,
     is_enum boolean NOT NULL DEFAULT false,
 
@@ -22,8 +35,8 @@ CREATE TABLE hdb_catalog.hdb_table
 
 CREATE TABLE hdb_catalog.hdb_relationship
 (
-    table_schema TEXT,
-    table_name TEXT,
+    table_schema name, -- See Note [Reference system columns using type name]
+    table_name name,
     rel_name   TEXT,
     rel_type   TEXT CHECK (rel_type IN ('object', 'array')),
     rel_def    JSONB NOT NULL,
@@ -36,8 +49,8 @@ CREATE TABLE hdb_catalog.hdb_relationship
 
 CREATE TABLE hdb_catalog.hdb_permission
 (
-    table_schema TEXT,
-    table_name TEXT,
+    table_schema name, -- See Note [Reference system columns using type name]
+    table_name name,
     role_name  TEXT,
     perm_type  TEXT CHECK(perm_type IN ('insert', 'select', 'update', 'delete')),
     perm_def   JSONB NOT NULL,
diff --git a/server/src-rsr/migrations/28_to_29.sql b/server/src-rsr/migrations/28_to_29.sql
index 76dcfd57919..3b7d4f44964 100644
--- a/server/src-rsr/migrations/28_to_29.sql
+++ b/server/src-rsr/migrations/28_to_29.sql
@@ -1,6 +1,30 @@
 DROP VIEW hdb_catalog.hdb_table_info_agg;
 DROP VIEW hdb_catalog.hdb_column;
 
+-- we need to drop this view temporarily so we can change the type of hdb_permission columns
+DROP VIEW hdb_catalog.hdb_permission_agg;
+
+-- See Note [Reference system columns using type name] in initialise.sql
+ALTER TABLE hdb_catalog.hdb_table
+  ALTER COLUMN table_schema TYPE name,
+  ALTER COLUMN table_name TYPE name;
+ALTER TABLE hdb_catalog.hdb_relationship
+  ALTER COLUMN table_schema TYPE name,
+  ALTER COLUMN table_name TYPE name;
+ALTER TABLE hdb_catalog.hdb_permission
+  ALTER COLUMN table_schema TYPE name,
+  ALTER COLUMN table_name TYPE name;
+
+-- this just recreates the view we dropped above with no actual changes
+CREATE VIEW hdb_catalog.hdb_permission_agg AS
+SELECT
+  table_schema,
+  table_name,
+  role_name,
+  json_object_agg(perm_type, perm_def) as permissions
+FROM hdb_catalog.hdb_permission
+GROUP BY table_schema, table_name, role_name;
+
 CREATE VIEW hdb_catalog.hdb_table_info_agg AS
   SELECT
     schema.nspname AS table_schema,