Details
-
Bug
-
Status: Open (View Workflow)
-
Minor
-
Resolution: Unresolved
-
10.11.16
-
None
-
Related to performance
Description
Example Test
(It's based on timing; don't do this on a slow CI.)
--let $rpl_skip_start_slave= 1
|
--source include/master-slave.inc
|
|
|
--save_master_pos
|
--connection slave
|
CHANGE MASTER TO MASTER_USE_GTID=NO, MASTER_DELAY= 5; |
SET @@GLOBAL.replicate_events_marked_for_skip= FILTER_ON_SLAVE; |
--source include/start_slave.inc |
# MDEV-38825
|
--sync_with_master
|
|
|
|
|
--connection master
|
SET @@SESSION.skip_replication= ON; |
CREATE TABLE t (a INT); |
DROP TABLE t; |
SET @@SESSION.skip_replication= OFF; |
--save_master_pos
|
--connection slave
|
--sleep 2 |
--let $status_items= Slave_SQL_Running_State, SQL_Remaining_Delay |
--source include/show_slave_status.inc |
# Serial: 'Slave has read all relay log; waiting for more updates' |
# Parallel: 'Waiting until MASTER_DELAY seconds after master executed event' |
--sync_with_master
|
|
|
|
|
--source include/stop_slave.inc
|
CHANGE MASTER TO MASTER_DELAY= 0; |
SET @@GLOBAL.replicate_events_marked_for_skip= REPLICATE; |
|
|
--let $rpl_only_running_threads= 1
|
--source include/rpl_end.inc |
Cause
Parallel Replication's design is flawed.
It skips events on worker threads, but delays events on the driver thread before queueing to workers.
Attachments
Issue Links
- causes
-
MDEV-30458 Consolidate Serial Replica to Parallel Replica with 1 Worker Thread
-
- Open
-
- is part of
-
MDEV-38672 Consolidate replica-side event group filtering
-
- Open
-
- relates to
-
MDEV-38825 Slave delays when there is "nothing" to replicate
-
- Open
-