On 2018-05-16 06:04 -0500, Dave Sherohman wrote:
> I'm running a galera cluster of three mariadb servers. It's been
> brought down twice because one node has detected an inconsistency,
> killed itself, and then systemd automatically restarted it. This is all
> good so far.
>
> The problem comes in when it tries to restart and, because it shut down
> in an inconsistent state, it wants to do a full state transfer to get
> back in sync with the rest of the cluster, which involves copying
> (potentially) 24G of data. Performing this transfer takes long enough
> that systemd times out, assumes the restart failed, and kills it, so
> it's not possible to bring the node back online without bypassing
> systemd and running mysqld_safe manually.
>
> Based on some web searches, I've tried using `systemctl edit mysql` to
> set "TimeoutStartUSec=infinity", but this does not appear to actually
> have any effect, even after reloading the systemd daemon:
>
> # systemctl edit mysql
> <add the setting using vim>
> # systemctl show mysql -p TimeoutStartUSec
> TimeoutStartUSec=1min 30s
> # systemctl daemon-reload
> # systemctl show mysql -p TimeoutStartUSec
> TimeoutStartUSec=1min 30s
> # systemctl daemon-reexec
> # systemctl show mysql -p TimeoutStartUSec
> TimeoutStartUSec=1min 30s
>
> What do I need to do to make this actually work?
Have you tried setting TimeoutStartSec rather than TimeoutStartUSec?
Though I have to admit that I did not perform a web search but cheated
by looking at the systemd.service(5) manpage, which mentions the former
but not the latter.
Cheers,
Sven
