Details
-
Bug
-
Status: Closed (View Workflow)
-
Critical
-
Resolution: Fixed
-
N/A
-
None
Description
--source include/have_binlog_format_row.inc
|
--source include/master-slave.inc
|
|
|
set drop_before_create_or_replace= on; |
|
|
create table t (a int); |
--error ER_DUP_ENTRY
|
create or replace table t (a int primary key) as select 1 as a union all select 1; |
show binlog events;
|
show tables;
|
|
|
--sync_slave_with_master
|
show tables;
|
|
|
--connection master
|
--source include/rpl_end.inc |
|
bb-main-monty 6cb6e8285c50a0bbcb2b599535b8d02b7bba5256 |
show binlog events;
|
Log_name Pos Event_type Server_id End_log_pos Info
|
master-bin.000001 4 Format_desc 1 256 Server ver: 12.0.0-MariaDB-debug-log, Binlog ver: 4
|
master-bin.000001 256 Gtid_list 1 285 []
|
master-bin.000001 285 Binlog_checkpoint 1 329 master-bin.000001
|
master-bin.000001 329 Gtid 1 371 GTID 0-1-1
|
master-bin.000001 371 Query 1 469 use `test`; create table t (a int)
|
show tables;
|
Tables_in_test
|
connection slave;
|
show tables;
|
Tables_in_test
|
t
|
Before MDEV-25292 (Atomic CREATE OR REPLACE), and with MDEV-28933 but in the statement binlog format, the non-atomic CREATE OR REPLACE drops the table, and a DROP event is written into the binary log. But after MDEV-25292, while it still drops the table because of drop_before_create_or_replace, the event is no longer generated, thus the table on the replica remains, and the servers diverge.
Attachments
Issue Links
- is caused by
-
MDEV-25292 Atomic CREATE OR REPLACE TABLE
-
- Stalled
-
Activity
Thanks Elkin.
I am not sure how to check whether it's quite the same problem. My test case from the description (minus the new option of course) doesn't fail even on 10.11.11 or on the current 10.6 which don't have the fix for MDEV-35499 – that is, both of them produce DROP event, and both primary and replica don't have the table after the failed CREATE OR REPLACE; and the patch for MDEV-35499 doesn't apply to Monty's branch, too many conflicts.
I suppose we should better keep it open until the fix is merged up to main, and Monty's branch is rebased on it; or, alternatively, maybe monty can take a look and say right away whether MDEV-35499 will fix it or not.
> I suppose we should better keep it open until the fix is merged up
Totally makes sense.
I think my issue is different as the code that handles writing the drop from a failed create or replace is rewritten in my code.
I have fixed this in my branch. Will push together with MDEV-36555 shortly
Fix pushed to bb-main-monty.
The issue was that old code wrote to the binlog cache before binlog_reset_cache() was executed in sql_insert.cc:select_create::abort_result_set()
Moving the writing to binlog cache after the above call fixed this issue
Could be a duplicated of just fixed
MDEV-35499Errored-out CREATE-or-REPLACE-SELECT does not log DROP table into binlog?Its commit is not present in the Monty's branch.