Status: Closed (View Workflow)
Resolution: Cannot Reproduce
5 node MariaDB galera cluster on CentOS 7.x
A simple alter crashed complete cluster after changing ENUM with mariadb internal code: 1677
ALTER TABLE `table_name`
CHANGE COLUMN `column_name_9` `column_name_9` ENUM('1','2','3','4') NULL DEFAULT NULL COLLATE 'utf8_general_ci' AFTER `column_name_8`;
All 5 nodes exited with:
2018-09-26 8:40:42 87095127 [ERROR] Slave SQL: Column 7 of table 'xxx.table_name' cannot be converted from type 'enum' to type 'varchar(300)', Internal MariaDB error code: 1677
column_name_1 int(1) unsigned
column_name_2 int(1) unsigned
column_name_4 int(1) unsigned
column_name_5 smallint(1) unsigned
column_name_6 int(1) unsigned
column_name_7 int(1) unsigned
So server can't change column 7 from type of column 8 to type of column 9 ?
Why would server issue a change on wrong column?
This happened on both TOI and RSU, while on RSU just the local node crashed, on TOI this happened to all nodes.
We were checking if slave_type_conversations could fix the issue, but this can't fix the problem of changing the type of a wrong column?
Hopefully we can resolve this quickly, since this never happened on 10.1.x versions.
- relates to
MDEV-17301 Change of COLLATE unnecessarily requires ALGORITHM=COPY