maria-discuss team mailing list archive

[William Edwards <wedwards@xxxxxxxxxxxxxx>]

Hi,

A customer ran an 'ALTER TABLE' query, which failed. This caused Galera 
slaves to shut themselves down:

    2021-11-23  9:35:25 1 [ERROR] Slave SQL: Column 2 of table 
'inserve_fpfvxiqm.invoices' cannot be converted from type 'date' to type 
'int(10) unsigned', Internal MariaDB error code: 1677
    2021-11-23  9:35:25 1 [Warning] WSREP: RBR event 17 Update_rows_v1 
apply warning: 3, 28891485
    2021-11-23  9:35:25 1 [Warning] WSREP: Failed to apply app buffer: 
seqno: 28891485, status: 1
             at 
/home/buildbot/buildbot/build/galera/src/trx_handle.cpp:apply():353
    [ 3 more times ]
    2021-11-23  9:35:25 1 [Warning] WSREP: RBR event 17 Update_rows_v1 
apply warning: 3, 28891485
    2021-11-23  9:35:26 1 [ERROR] WSREP: Failed to apply trx: source: 
41c479ab-0527-11ec-9fcc-9f02b1141194 version: 4 local: 0 state: APPLYING 
flags: 1 conn_id: 13475392 trx_id: 63620823 seqnos (l:     312653, g: 
28891485, s: 28891454, d: 28891449, ts: 10603037948851154)
    2021-11-23  9:35:26 1 [ERROR] WSREP: Failed to apply trx 28891485 4 
times
    2021-11-23  9:35:26 1 [ERROR] WSREP: Node consistency compromised, 
aborting...

The node on which the query was executed, crashed ("mysqld got signal 
11"). At first sight, this seems to be the result of "WSREP: Send action 
{(nil), 528, TORDERED} returned -107 (Transport endpoint is not 
connected)" - as all the slaves were down at this point.

As 'ALTER' is a DDL statement that does not support rollback, I 
understand that the only option for Galera may be to abort. But is there 
a way to handle something like this more gracefully?

--
With kind regards,

William Edwards