From 33ec33c5b31072bd8a386a6f381d48cf0a07c4c7 Mon Sep 17 00:00:00 2001
From: Onder Kalaci <onder@citusdata.com>
Date: Fri, 15 Sep 2017 02:52:30 +0300
Subject: [PATCH] Ensure schema exists on reference table creation

If the schema doesn't exists on the workers, create it.
---
 .../commands/create_distributed_table.c       |  8 +++
 .../regress/expected/multi_create_table.out   | 55 +++++++++++++++++++
 src/test/regress/sql/multi_create_table.sql   | 36 ++++++++++++
 3 files changed, 99 insertions(+)

diff --git a/src/backend/distributed/commands/create_distributed_table.c b/src/backend/distributed/commands/create_distributed_table.c
index 247e41277..1e3d1da9f 100644
--- a/src/backend/distributed/commands/create_distributed_table.c
+++ b/src/backend/distributed/commands/create_distributed_table.c
@@ -236,6 +236,14 @@ create_reference_table(PG_FUNCTION_ARGS)
 	EnsureCoordinator();
 	CheckCitusVersion(ERROR);
 
+	/*
+	 * Ensure schema exists on each worker node. We can not run this function
+	 * transactionally, since we may create shards over separate sessions and
+	 * shard creation depends on the schema being present and visible from all
+	 * sessions.
+	 */
+	EnsureSchemaExistsOnAllNodes(relationId);
+
 	/*
 	 * Lock target relation with an exclusive lock - there's no way to make
 	 * sense of this table until we've committed, and we don't want multiple
diff --git a/src/test/regress/expected/multi_create_table.out b/src/test/regress/expected/multi_create_table.out
index bba39e136..f498325a5 100644
--- a/src/test/regress/expected/multi_create_table.out
+++ b/src/test/regress/expected/multi_create_table.out
@@ -873,5 +873,60 @@ SELECT count(*) FROM tt1;
      6
 (1 row)
 
+-- the goal of the following test is to make sure that
+-- both create_reference_table and create_distributed_table
+-- calls creates the schemas without leading to any deadlocks
+-- first create reference table, then hash distributed table
+BEGIN;
+CREATE SCHEMA sc;
+CREATE TABLE sc.ref(a int);
+insert into sc.ref SELECT s FROM generate_series(0, 100) s;
+SELECT create_reference_table('sc.ref');
+NOTICE:  Copying data from local table...
+ create_reference_table 
+------------------------
+ 
+(1 row)
+
+CREATE TABLE sc.hash(a int);
+insert into sc.hash SELECT s FROM generate_series(0, 100) s;
+SELECT create_distributed_table('sc.hash', 'a');
+NOTICE:  Copying data from local table...
+ create_distributed_table 
+--------------------------
+ 
+(1 row)
+
+COMMIT;
+-- first create hash distributed table, then reference table
+BEGIN;
+CREATE SCHEMA sc2;
+CREATE TABLE sc2.hash(a int);
+insert into sc2.hash SELECT s FROM generate_series(0, 100) s;
+SELECT create_distributed_table('sc2.hash', 'a');
+NOTICE:  Copying data from local table...
+ create_distributed_table 
+--------------------------
+ 
+(1 row)
+
+CREATE TABLE sc2.ref(a int);
+insert into sc2.ref SELECT s FROM generate_series(0, 100) s;
+SELECT create_reference_table('sc2.ref');
+NOTICE:  Copying data from local table...
+ create_reference_table 
+------------------------
+ 
+(1 row)
+
+COMMIT;
 DROP TABLE tt1;
 DROP TABLE tt2;
+DROP SCHEMA sc CASCADE;
+NOTICE:  drop cascades to 2 other objects
+DETAIL:  drop cascades to table sc.ref
+drop cascades to table sc.hash
+DROP SCHEMA sc2 CASCADE;
+NOTICE:  drop cascades to 2 other objects
+DETAIL:  drop cascades to table sc2.hash
+drop cascades to table sc2.ref
diff --git a/src/test/regress/sql/multi_create_table.sql b/src/test/regress/sql/multi_create_table.sql
index 531f337ea..d054eb314 100644
--- a/src/test/regress/sql/multi_create_table.sql
+++ b/src/test/regress/sql/multi_create_table.sql
@@ -482,5 +482,41 @@ END;
 
 SELECT count(*) FROM tt1;
 
+
+-- the goal of the following test is to make sure that
+-- both create_reference_table and create_distributed_table
+-- calls creates the schemas without leading to any deadlocks
+
+-- first create reference table, then hash distributed table
+BEGIN;
+
+CREATE SCHEMA sc;
+CREATE TABLE sc.ref(a int);
+insert into sc.ref SELECT s FROM generate_series(0, 100) s;
+SELECT create_reference_table('sc.ref');
+
+CREATE TABLE sc.hash(a int);
+insert into sc.hash SELECT s FROM generate_series(0, 100) s;
+SELECT create_distributed_table('sc.hash', 'a');
+
+COMMIT;
+
+
+-- first create hash distributed table, then reference table
+BEGIN;
+
+CREATE SCHEMA sc2;
+CREATE TABLE sc2.hash(a int);
+insert into sc2.hash SELECT s FROM generate_series(0, 100) s;
+SELECT create_distributed_table('sc2.hash', 'a');
+
+CREATE TABLE sc2.ref(a int);
+insert into sc2.ref SELECT s FROM generate_series(0, 100) s;
+SELECT create_reference_table('sc2.ref');
+
+COMMIT;
+
 DROP TABLE tt1;
 DROP TABLE tt2;
+DROP SCHEMA sc CASCADE;
+DROP SCHEMA sc2 CASCADE;