Why SST on restart

[Philip Brown]

Done a little websearching, and found info on cleaner full cluster restarts, but not on single node.

Background: I'm new to Galera,and coming in as a sysadmin for an existing galera cluster at a new company.

I added a new node  (mariadb 10.1, galera 25.3.28, on Centos).
It did its initial SST, finished up, and was all happy.   wsrep_cluster_size reported as 4.

I wanted to do an OS patch. So i did a quick shutdown and restart of the mariadb service.

probably took only around 10 min or so.

it decides it needs to do a full SST resync, instead of IST.

I can find no clue as to why. So I would appreciate it if folks could help me figure out:

* Why it chose that

* How to avoid it in future, for "quick" node restarts

FYI, cache.size=16G. across all nodes in the cluster. SO I dont think it overflowed the gcache.

[Garrett Boast]

It's possible the system did not let the node shut down gracefully. What system/service manager are you using?

If it is systemd there is a shutdown timeout ( https://www.freedesktop.org/software/systemd/man/systemd.service.html#TimeoutStopSec= )

Ensure the value is sane (e.g. Infinity), especially if you have  innodb_fast_shutdown = 0.

Logs would be helpful to diagnose.

Regards,

Garrett

[Fco. Javier Lopez]

Although long time ago, this may help someone else ...

Could be related to gcache.recover, set by default to no.

Old doc to read about this, although the galera documentation is very clear: https://www.percona.com/blog/2016/11/30/galera-cache-gcache-finally-recoverable-restart/

Regards