-- -- MULTI_COLOCATED_SHARD_REBALANCE -- ALTER SEQUENCE pg_catalog.pg_dist_shardid_seq RESTART 13000000; SET citus.shard_count TO 6; SET citus.shard_replication_factor TO 1; -- create distributed tables CREATE TABLE table1_group1 ( id int PRIMARY KEY); SELECT create_distributed_table('table1_group1', 'id', 'hash'); create_distributed_table --------------------------------------------------------------------- (1 row) CREATE TABLE table2_group1 ( id int ); SELECT create_distributed_table('table2_group1', 'id', 'hash'); create_distributed_table --------------------------------------------------------------------- (1 row) SET citus.shard_count TO 8; CREATE TABLE table5_groupX ( id int ); SELECT create_distributed_table('table5_groupX', 'id', 'hash'); create_distributed_table --------------------------------------------------------------------- (1 row) CREATE TABLE table6_append ( id int ); SELECT master_create_distributed_table('table6_append', 'id', 'append'); master_create_distributed_table --------------------------------------------------------------------- (1 row) SELECT master_create_empty_shard('table6_append'); master_create_empty_shard --------------------------------------------------------------------- 13000020 (1 row) SELECT master_create_empty_shard('table6_append'); master_create_empty_shard --------------------------------------------------------------------- 13000021 (1 row) -- Mark tables as non-mx tables, in order to be able to test master_copy_shard_placement UPDATE pg_dist_partition SET repmodel='c' WHERE logicalrelid IN ('table1_group1'::regclass, 'table2_group1'::regclass, 'table5_groupX'::regclass); -- test copy -- test copying colocated shards -- status before shard copy SELECT s.shardid, s.logicalrelid::regclass, sp.nodeport FROM pg_dist_partition p, pg_dist_shard s, pg_dist_shard_placement sp WHERE p.logicalrelid = s.logicalrelid AND s.shardid = sp.shardid AND colocationid = (SELECT colocationid FROM pg_dist_partition WHERE logicalrelid = 'table1_group1'::regclass) ORDER BY s.shardid, sp.nodeport; shardid | logicalrelid | nodeport --------------------------------------------------------------------- 13000000 | table1_group1 | 57637 13000001 | table1_group1 | 57638 13000002 | table1_group1 | 57637 13000003 | table1_group1 | 57638 13000004 | table1_group1 | 57637 13000005 | table1_group1 | 57638 13000006 | table2_group1 | 57637 13000007 | table2_group1 | 57638 13000008 | table2_group1 | 57637 13000009 | table2_group1 | 57638 13000010 | table2_group1 | 57637 13000011 | table2_group1 | 57638 (12 rows) -- copy colocated shards SELECT master_copy_shard_placement(13000000, 'localhost', :worker_1_port, 'localhost', :worker_2_port, false); master_copy_shard_placement --------------------------------------------------------------------- (1 row) -- status after shard copy SELECT s.shardid, s.logicalrelid::regclass, sp.nodeport FROM pg_dist_partition p, pg_dist_shard s, pg_dist_shard_placement sp WHERE p.logicalrelid = s.logicalrelid AND s.shardid = sp.shardid AND colocationid = (SELECT colocationid FROM pg_dist_partition WHERE logicalrelid = 'table1_group1'::regclass) ORDER BY s.shardid, sp.nodeport; shardid | logicalrelid | nodeport --------------------------------------------------------------------- 13000000 | table1_group1 | 57637 13000000 | table1_group1 | 57638 13000001 | table1_group1 | 57638 13000002 | table1_group1 | 57637 13000003 | table1_group1 | 57638 13000004 | table1_group1 | 57637 13000005 | table1_group1 | 57638 13000006 | table2_group1 | 57637 13000006 | table2_group1 | 57638 13000007 | table2_group1 | 57638 13000008 | table2_group1 | 57637 13000009 | table2_group1 | 57638 13000010 | table2_group1 | 57637 13000011 | table2_group1 | 57638 (14 rows) -- also connect worker to verify we successfully copied given shard (and other colocated shards) \c - - - :worker_2_port SELECT "Column", "Type", "Modifiers" FROM table_desc WHERE relid='public.table1_group1_13000000'::regclass; Column | Type | Modifiers --------------------------------------------------------------------- id | integer | not null (1 row) SELECT "Column", "Type", "Modifiers" FROM table_desc WHERE relid='public.table2_group1_13000006'::regclass; Column | Type | Modifiers --------------------------------------------------------------------- id | integer | (1 row) \c - - - :master_port -- copy colocated shards again to see error message SELECT master_copy_shard_placement(13000000, 'localhost', :worker_1_port, 'localhost', :worker_2_port, false, 'force_logical'); ERROR: the force_logical transfer mode is currently unsupported -- test copying NOT colocated shard -- status before shard copy SELECT s.shardid, s.logicalrelid::regclass, sp.nodeport FROM pg_dist_partition p, pg_dist_shard s, pg_dist_shard_placement sp WHERE p.logicalrelid = s.logicalrelid AND s.shardid = sp.shardid AND p.logicalrelid = 'table5_groupX'::regclass ORDER BY s.shardid, sp.nodeport; shardid | logicalrelid | nodeport --------------------------------------------------------------------- 13000012 | table5_groupx | 57637 13000013 | table5_groupx | 57638 13000014 | table5_groupx | 57637 13000015 | table5_groupx | 57638 13000016 | table5_groupx | 57637 13000017 | table5_groupx | 57638 13000018 | table5_groupx | 57637 13000019 | table5_groupx | 57638 (8 rows) -- copy NOT colocated shard SELECT master_copy_shard_placement(13000012, 'localhost', :worker_1_port, 'localhost', :worker_2_port, false); master_copy_shard_placement --------------------------------------------------------------------- (1 row) -- status after shard copy SELECT s.shardid, s.logicalrelid::regclass, sp.nodeport FROM pg_dist_partition p, pg_dist_shard s, pg_dist_shard_placement sp WHERE p.logicalrelid = s.logicalrelid AND s.shardid = sp.shardid AND p.logicalrelid = 'table5_groupX'::regclass ORDER BY s.shardid, sp.nodeport; shardid | logicalrelid | nodeport --------------------------------------------------------------------- 13000012 | table5_groupx | 57637 13000012 | table5_groupx | 57638 13000013 | table5_groupx | 57638 13000014 | table5_groupx | 57637 13000015 | table5_groupx | 57638 13000016 | table5_groupx | 57637 13000017 | table5_groupx | 57638 13000018 | table5_groupx | 57637 13000019 | table5_groupx | 57638 (9 rows) -- test copying shard in append distributed table -- status before shard copy SELECT s.shardid, s.logicalrelid::regclass, sp.nodeport FROM pg_dist_partition p, pg_dist_shard s, pg_dist_shard_placement sp WHERE p.logicalrelid = s.logicalrelid AND s.shardid = sp.shardid AND p.logicalrelid = 'table6_append'::regclass ORDER BY s.shardid, sp.nodeport; shardid | logicalrelid | nodeport --------------------------------------------------------------------- 13000020 | table6_append | 57638 13000021 | table6_append | 57637 (2 rows) -- copy shard in append distributed table SELECT master_copy_shard_placement(13000020, 'localhost', :worker_2_port, 'localhost', :worker_1_port, false, 'force_logical'); ERROR: the force_logical transfer mode is currently unsupported -- status after shard copy SELECT s.shardid, s.logicalrelid::regclass, sp.nodeport FROM pg_dist_partition p, pg_dist_shard s, pg_dist_shard_placement sp WHERE p.logicalrelid = s.logicalrelid AND s.shardid = sp.shardid AND p.logicalrelid = 'table6_append'::regclass ORDER BY s.shardid, sp.nodeport; shardid | logicalrelid | nodeport --------------------------------------------------------------------- 13000020 | table6_append | 57638 13000021 | table6_append | 57637 (2 rows) -- test move -- test moving colocated shards -- status before shard move SELECT s.shardid, s.logicalrelid::regclass, sp.nodeport FROM pg_dist_partition p, pg_dist_shard s, pg_dist_shard_placement sp WHERE p.logicalrelid = s.logicalrelid AND s.shardid = sp.shardid AND colocationid = (SELECT colocationid FROM pg_dist_partition WHERE logicalrelid = 'table1_group1'::regclass) ORDER BY s.shardid, sp.nodeport; shardid | logicalrelid | nodeport --------------------------------------------------------------------- 13000000 | table1_group1 | 57637 13000000 | table1_group1 | 57638 13000001 | table1_group1 | 57638 13000002 | table1_group1 | 57637 13000003 | table1_group1 | 57638 13000004 | table1_group1 | 57637 13000005 | table1_group1 | 57638 13000006 | table2_group1 | 57637 13000006 | table2_group1 | 57638 13000007 | table2_group1 | 57638 13000008 | table2_group1 | 57637 13000009 | table2_group1 | 57638 13000010 | table2_group1 | 57637 13000011 | table2_group1 | 57638 (14 rows) -- try force_logical SELECT master_move_shard_placement(13000001, 'localhost', :worker_2_port, 'localhost', :worker_1_port, 'force_logical'); ERROR: the force_logical transfer mode is currently unsupported -- move colocated shards SELECT master_move_shard_placement(13000001, 'localhost', :worker_2_port, 'localhost', :worker_1_port); master_move_shard_placement --------------------------------------------------------------------- (1 row) -- status after shard move SELECT s.shardid, s.logicalrelid::regclass, sp.nodeport FROM pg_dist_partition p, pg_dist_shard s, pg_dist_shard_placement sp WHERE p.logicalrelid = s.logicalrelid AND s.shardid = sp.shardid AND colocationid = (SELECT colocationid FROM pg_dist_partition WHERE logicalrelid = 'table1_group1'::regclass) AND sp.shardstate != 4 ORDER BY s.shardid, sp.nodeport; shardid | logicalrelid | nodeport --------------------------------------------------------------------- 13000000 | table1_group1 | 57637 13000000 | table1_group1 | 57638 13000001 | table1_group1 | 57637 13000002 | table1_group1 | 57637 13000003 | table1_group1 | 57638 13000004 | table1_group1 | 57637 13000005 | table1_group1 | 57638 13000006 | table2_group1 | 57637 13000006 | table2_group1 | 57638 13000007 | table2_group1 | 57637 13000008 | table2_group1 | 57637 13000009 | table2_group1 | 57638 13000010 | table2_group1 | 57637 13000011 | table2_group1 | 57638 (14 rows) -- also connect worker to verify we successfully moved given shard (and other colocated shards) \c - - - :worker_1_port SELECT "Column", "Type", "Modifiers" FROM table_desc WHERE relid='public.table1_group1_13000001'::regclass; Column | Type | Modifiers --------------------------------------------------------------------- id | integer | not null (1 row) SELECT "Column", "Type", "Modifiers" FROM table_desc WHERE relid='public.table2_group1_13000007'::regclass; Column | Type | Modifiers --------------------------------------------------------------------- id | integer | (1 row) \c - - - :master_port -- test moving NOT colocated shard -- status before shard move SELECT s.shardid, s.logicalrelid::regclass, sp.nodeport FROM pg_dist_partition p, pg_dist_shard s, pg_dist_shard_placement sp WHERE p.logicalrelid = s.logicalrelid AND s.shardid = sp.shardid AND p.logicalrelid = 'table5_groupX'::regclass AND sp.shardstate != 4 ORDER BY s.shardid, sp.nodeport; shardid | logicalrelid | nodeport --------------------------------------------------------------------- 13000012 | table5_groupx | 57637 13000012 | table5_groupx | 57638 13000013 | table5_groupx | 57638 13000014 | table5_groupx | 57637 13000015 | table5_groupx | 57638 13000016 | table5_groupx | 57637 13000017 | table5_groupx | 57638 13000018 | table5_groupx | 57637 13000019 | table5_groupx | 57638 (9 rows) -- move NOT colocated shard SELECT master_move_shard_placement(13000013, 'localhost', :worker_2_port, 'localhost', :worker_1_port); master_move_shard_placement --------------------------------------------------------------------- (1 row) -- status after shard move SELECT s.shardid, s.logicalrelid::regclass, sp.nodeport FROM pg_dist_partition p, pg_dist_shard s, pg_dist_shard_placement sp WHERE p.logicalrelid = s.logicalrelid AND s.shardid = sp.shardid AND p.logicalrelid = 'table5_groupX'::regclass AND sp.shardstate != 4 ORDER BY s.shardid, sp.nodeport; shardid | logicalrelid | nodeport --------------------------------------------------------------------- 13000012 | table5_groupx | 57637 13000012 | table5_groupx | 57638 13000013 | table5_groupx | 57637 13000014 | table5_groupx | 57637 13000015 | table5_groupx | 57638 13000016 | table5_groupx | 57637 13000017 | table5_groupx | 57638 13000018 | table5_groupx | 57637 13000019 | table5_groupx | 57638 (9 rows) -- test moving shard in append distributed table -- status before shard move SELECT s.shardid, s.logicalrelid::regclass, sp.nodeport FROM pg_dist_partition p, pg_dist_shard s, pg_dist_shard_placement sp WHERE p.logicalrelid = s.logicalrelid AND s.shardid = sp.shardid AND p.logicalrelid = 'table6_append'::regclass AND sp.shardstate != 4 ORDER BY s.shardid, sp.nodeport; shardid | logicalrelid | nodeport --------------------------------------------------------------------- 13000020 | table6_append | 57638 13000021 | table6_append | 57637 (2 rows) -- move shard in append distributed table SELECT master_move_shard_placement(13000021, 'localhost', :worker_1_port, 'localhost', :worker_2_port); master_move_shard_placement --------------------------------------------------------------------- (1 row) -- status after shard move SELECT s.shardid, s.logicalrelid::regclass, sp.nodeport FROM pg_dist_partition p, pg_dist_shard s, pg_dist_shard_placement sp WHERE p.logicalrelid = s.logicalrelid AND s.shardid = sp.shardid AND p.logicalrelid = 'table6_append'::regclass AND sp.shardstate != 4 ORDER BY s.shardid, sp.nodeport; shardid | logicalrelid | nodeport --------------------------------------------------------------------- 13000020 | table6_append | 57638 13000021 | table6_append | 57638 (2 rows) -- try to move shard from wrong node SELECT master_move_shard_placement(13000021, 'localhost', :worker_1_port, 'localhost', :worker_2_port); ERROR: source placement must be in active state -- test shard move with foreign constraints DROP TABLE IF EXISTS table1_group1, table2_group1; SET citus.shard_count TO 6; SET citus.shard_replication_factor TO 1; -- create distributed tables CREATE TABLE table1_group1 ( id int PRIMARY KEY); SELECT create_distributed_table('table1_group1', 'id', 'hash'); create_distributed_table --------------------------------------------------------------------- (1 row) CREATE TABLE table2_group1 ( id int, table1_id int, FOREIGN KEY(table1_id) REFERENCES table1_group1(id)); SELECT create_distributed_table('table2_group1', 'table1_id', 'hash'); create_distributed_table --------------------------------------------------------------------- (1 row) -- Mark the tables as non-mx tables UPDATE pg_dist_partition SET repmodel='c' WHERE logicalrelid IN ('table1_group1'::regclass, 'table2_group1'::regclass); -- status before shard rebalance SELECT s.shardid, s.logicalrelid::regclass, sp.nodeport FROM pg_dist_partition p, pg_dist_shard s, pg_dist_shard_placement sp WHERE p.logicalrelid = s.logicalrelid AND s.shardid = sp.shardid AND colocationid = (SELECT colocationid FROM pg_dist_partition WHERE logicalrelid = 'table1_group1'::regclass) AND sp.shardstate != 4 ORDER BY s.shardid, sp.nodeport; shardid | logicalrelid | nodeport --------------------------------------------------------------------- 13000022 | table1_group1 | 57637 13000023 | table1_group1 | 57638 13000024 | table1_group1 | 57637 13000025 | table1_group1 | 57638 13000026 | table1_group1 | 57637 13000027 | table1_group1 | 57638 13000028 | table2_group1 | 57637 13000029 | table2_group1 | 57638 13000030 | table2_group1 | 57637 13000031 | table2_group1 | 57638 13000032 | table2_group1 | 57637 13000033 | table2_group1 | 57638 (12 rows) SELECT master_move_shard_placement(13000022, 'localhost', :worker_1_port, 'localhost', :worker_2_port, 'block_writes'); master_move_shard_placement --------------------------------------------------------------------- (1 row) -- status after shard rebalance SELECT s.shardid, s.logicalrelid::regclass, sp.nodeport FROM pg_dist_partition p, pg_dist_shard s, pg_dist_shard_placement sp WHERE p.logicalrelid = s.logicalrelid AND s.shardid = sp.shardid AND colocationid = (SELECT colocationid FROM pg_dist_partition WHERE logicalrelid = 'table1_group1'::regclass) AND sp.shardstate != 4 ORDER BY s.shardid, sp.nodeport; shardid | logicalrelid | nodeport --------------------------------------------------------------------- 13000022 | table1_group1 | 57638 13000023 | table1_group1 | 57638 13000024 | table1_group1 | 57637 13000025 | table1_group1 | 57638 13000026 | table1_group1 | 57637 13000027 | table1_group1 | 57638 13000028 | table2_group1 | 57638 13000029 | table2_group1 | 57638 13000030 | table2_group1 | 57637 13000031 | table2_group1 | 57638 13000032 | table2_group1 | 57637 13000033 | table2_group1 | 57638 (12 rows) -- also connect worker to verify we successfully moved given shard (and other colocated shards) \c - - - :worker_2_port SELECT "Column", "Type", "Modifiers" FROM table_desc WHERE relid='public.table1_group1_13000022'::regclass; Column | Type | Modifiers --------------------------------------------------------------------- id | integer | not null (1 row) SELECT "Column", "Type", "Modifiers" FROM table_desc WHERE relid='public.table2_group1_13000028'::regclass; Column | Type | Modifiers --------------------------------------------------------------------- id | integer | table1_id | integer | (2 rows) -- make sure that we've created the foreign keys SELECT "Constraint", "Definition" FROM table_fkeys WHERE "Constraint" LIKE 'table2_group%' OR "Constraint" LIKE 'table1_group%'; Constraint | Definition --------------------------------------------------------------------- table2_group1_table1_id_fkey_13000028 | FOREIGN KEY (table1_id) REFERENCES table1_group1_13000022(id) table2_group1_table1_id_fkey_13000029 | FOREIGN KEY (table1_id) REFERENCES table1_group1_13000023(id) table2_group1_table1_id_fkey_13000031 | FOREIGN KEY (table1_id) REFERENCES table1_group1_13000025(id) table2_group1_table1_id_fkey_13000033 | FOREIGN KEY (table1_id) REFERENCES table1_group1_13000027(id) (4 rows) \c - - - :master_port -- test shard copy with foreign constraints -- we expect it to error out because we do not support foreign constraints with replication factor > 1 SELECT master_copy_shard_placement(13000022, 'localhost', :worker_2_port, 'localhost', :worker_1_port, false); ERROR: cannot replicate shards with foreign keys -- lets also test that master_move_shard_placement doesn't break serials CREATE TABLE serial_move_test (key int, other_val serial); SET citus.shard_replication_factor TO 1; SELECT create_distributed_table('serial_move_test', 'key'); create_distributed_table --------------------------------------------------------------------- (1 row) -- key 15 goes to shard xxxxx INSERT INTO serial_move_test (key) VALUES (15) RETURNING *; key | other_val --------------------------------------------------------------------- 15 | 1 (1 row) INSERT INTO serial_move_test (key) VALUES (15) RETURNING *; key | other_val --------------------------------------------------------------------- 15 | 2 (1 row) -- confirm the shard id SELECT * FROM run_command_on_placements('serial_move_test', 'SELECT DISTINCT key FROM %s WHERE key = 15') WHERE result = '15' AND shardid = 13000034; nodename | nodeport | shardid | success | result --------------------------------------------------------------------- localhost | 57637 | 13000034 | t | 15 (1 row) SELECT master_move_shard_placement(13000034, 'localhost', :worker_1_port, 'localhost', :worker_2_port); master_move_shard_placement --------------------------------------------------------------------- (1 row) -- confirm the successfull move SELECT * FROM run_command_on_placements('serial_move_test', 'SELECT DISTINCT key FROM %s WHERE key = 15') WHERE result = '15' AND shardid = 13000034; nodename | nodeport | shardid | success | result --------------------------------------------------------------------- localhost | 57638 | 13000034 | t | 15 (1 row) -- finally show that serials work fine afterwards INSERT INTO serial_move_test (key) VALUES (15) RETURNING *; key | other_val --------------------------------------------------------------------- 15 | 3 (1 row) INSERT INTO serial_move_test (key) VALUES (15) RETURNING *; key | other_val --------------------------------------------------------------------- 15 | 4 (1 row) -- we should be able to move shard placements of partitioend tables CREATE SCHEMA move_partitions; CREATE TABLE move_partitions.events ( id serial, t timestamptz default now(), payload text ) PARTITION BY RANGE(t); SET citus.shard_count TO 6; SELECT create_distributed_table('move_partitions.events', 'id', colocate_with := 'none'); create_distributed_table --------------------------------------------------------------------- (1 row) CREATE TABLE move_partitions.events_1 PARTITION OF move_partitions.events FOR VALUES FROM ('2015-01-01') TO ('2016-01-01'); INSERT INTO move_partitions.events (t, payload) SELECT '2015-01-01'::date + (interval '1 day' * s), s FROM generate_series(1, 100) s; SELECT count(*) FROM move_partitions.events; count --------------------------------------------------------------------- 100 (1 row) -- try to move automatically SELECT master_move_shard_placement(shardid, 'localhost', :worker_2_port, 'localhost', :worker_1_port) FROM pg_dist_shard JOIN pg_dist_shard_placement USING (shardid) WHERE logicalrelid = 'move_partitions.events'::regclass AND nodeport = :worker_2_port AND shardstate != 4 ORDER BY shardid LIMIT 1; master_move_shard_placement --------------------------------------------------------------------- (1 row) SELECT count(*) FROM move_partitions.events; count --------------------------------------------------------------------- 100 (1 row) -- add a primary key to the partition ALTER TABLE move_partitions.events_1 ADD CONSTRAINT e_1_pk PRIMARY KEY (id); -- should be able to move automatically now SELECT master_move_shard_placement(shardid, 'localhost', :worker_2_port, 'localhost', :worker_1_port) FROM pg_dist_shard JOIN pg_dist_shard_placement USING (shardid) WHERE logicalrelid = 'move_partitions.events'::regclass AND nodeport = :worker_2_port AND shardstate != 4 ORDER BY shardid LIMIT 1; master_move_shard_placement --------------------------------------------------------------------- (1 row) SELECT count(*) FROM move_partitions.events; count --------------------------------------------------------------------- 100 (1 row) -- should also be able to move with block writes SELECT master_move_shard_placement(shardid, 'localhost', :worker_2_port, 'localhost', :worker_1_port, 'block_writes') FROM pg_dist_shard JOIN pg_dist_shard_placement USING (shardid) WHERE logicalrelid = 'move_partitions.events'::regclass AND nodeport = :worker_2_port AND shardstate != 4 ORDER BY shardid LIMIT 1; master_move_shard_placement --------------------------------------------------------------------- (1 row) SELECT count(*) FROM move_partitions.events; count --------------------------------------------------------------------- 100 (1 row) -- should have moved all shards to node 1 (2*6 = 12) SELECT count(*) FROM pg_dist_shard JOIN pg_dist_shard_placement USING (shardid) WHERE logicalrelid::text LIKE 'move_partitions.events%' AND nodeport = :worker_1_port; count --------------------------------------------------------------------- 12 (1 row) DROP TABLE move_partitions.events;