Data corruption will happen as soon as the galera cluster size change
Let's take a table t made of single auto increment and the following binlog
# at 770118590
|
#160203 21:31:19 server id 10 end_log_pos 770118618 Intvar
|
SET INSERT_ID=134024921/*!*/;
|
# at 770118618
|
#160203 21:31:19 server id 10 end_log_pos 770121328 Query thread_id=961224213 exec_time=0 error_code=0
|
SET TIMESTAMP=1454531479/*!*/;
|
INSERT INTO t values (NULL),(NULL ),(NULL ),(NULL ),(NULL ),(NULL ),(NULL ),(NULL ),(NULL ),(NULL ),(NULL ),(NULL )
|
/*!*/;
|
# at 770121328
|
#160203 21:31:19 server id 10 end_log_pos 770121355 Xid = 701968214
|
COMMIT/*!*/;
|
# at 770121355
|
#160203 21:31:19 server id 10 end_log_pos 770121414 Query thread_id=961224213 exec_time=0 error_code=0
|
SET TIMESTAMP=1454531479/*!*/;
|
BEGIN
|
/*!*/;
|
# at 770121414
|
#160203 21:31:19 server id 10 end_log_pos 770121442 Intvar
|
SET INSERT_ID=134024933/*!*/;
|
# at 770121442
|
#160203 21:31:19 server id 10 end_log_pos 770124812 Query thread_id=961224213 exec_time=0 error_code=0
|
SET TIMESTAMP=1454531479/*!*/;
|
INSERT INTO t values (NULL),(NULL ),(NULL ),(NULL ),(NULL ),(NULL ),(NULL ),(NULL ),(NULL ),(NULL ),(NULL ),(NULL )
|
Replication will break because the step of auto increment will change as soon the cluster size change. In statement base the master did not change the step and auto increment overlay will happen and corrupt the data ?
Last_SQL_Error: Error 'Duplicate entry '134024933
|
|
MariaDB [db_calameo_v2]> SELECT ID FROM t WHERE id >= 134024921;
|
+-----------+
|
| LinkID |
|
+-----------+
|
| 134024921 |
|
| 134024923 |
|
| 134024925 |
|
| 134024927 |
|
| 134024929 |
|
| 134024931 |
|
| 134024933 |
|
| 134024935 |
|
| 134024937 |
|
| 134024939 |
|
| 134024941 |
|
| 134024943 |
|
+-----------+
|
So this proposal is to introduce verbose error message at replication start to either change wsrep_auto_increment_control=1 or to use ROW based .
Bug
Major
MDEV-6445 UUID column type addition for distributed systems
