[MDEV-5110] init script for mariadb galera cluster Created: 2013-10-08  Updated: 2015-10-30  Resolved: 2015-10-30

Status: Closed
Project: MariaDB Server
Component/s: Scripts & Clients
Fix Version/s: N/A

Type: Task Priority: Minor
Reporter: erkan yanar Assignee: Nirbhay Choubey (Inactive)
Resolution: Done Votes: 0
Labels: galera

Issue Links:
Relates
relates to MDEV-703 LP:870310 - killall -9 in init-script Closed

 Description   

Having a look into the init script from mariadb-galera-server-5.5 I see some things that are not good. Some of them also fit to mariadb in general.

  • kill -9 (See Bug #870310)
  • killall mysqld (Do I really want to kill all running mysqld?)

Things that are not good for Galera or are missing for a better Galera support.

  • Relying on a credentials file (/etc/mysql/debian.cnf). This is always bad, because it is inconsistent. Restarting/Status-Check a server will break, because the credentials in the file does not match the ones in the database. I wonder why not use any socket-auth.
  • Missing Bootstrap for Galera. PXC ships with a bootstrap option. May be worth to think about it.
  • The --wsrep-recover Option is ignored. This will most likely prevent a SST after a crash, as it can be used to get the last committed seqno

 Comments   
Comment by Elena Stepanova [ 2013-10-08 ]

Bug #870310 was (somewhat) fixed as MDEV-703. Now the timeout is 10 min, and I'd say if it didn't shutdown properly in this time, something is most likely wrong, and it will have to be killed anyway. The reason for doing it in the script is the upgrade i think, which would hung if the server hangs on shutdown.

killall is still present though.

For the Galera issues, Jan, could you please take a look and decide whether it should be done on our side or we should reassign it to Seppo? And if it's to be done on our side, please decide between you and Holyfoot who would be doing it.
As far as I know, the script gets merges from Codership, but at some point it was touched by Holyfoot as well, as a part of MDEV-507.

Thanks.

Comment by erkan yanar [ 2013-10-31 ]

Forget the wsrep-recover thing. It is in mysqld_safe and not in the init script.

Comment by Daniel Black [ 2014-03-04 ]

ref kill/killall - perhaps check its not in a DONOR state.

bootstrap - see MDEV-4253

credentials - see MDEV-5500 however socket-auth sounds like a much better option

Comment by Nirbhay Choubey (Inactive) [ 2015-10-30 ]

kill/killall : MDEV-703
Credentials : MDEV-5500 (unix socket)
Missing bootstrap : MDEV-5759, MDEV-8804

Comment by Nirbhay Choubey (Inactive) [ 2015-10-30 ]

On shutdown, if the SST is in progress, it may fail, and thus the joiner node needs to find another donor node.

Generated at Thu Feb 08 07:01:45 UTC 2024 using Jira 8.20.16#820016-sha1:9d11dbea5f4be3d4cc21f03a88dd11d8c8687422.