Metadata warning is logged when Pacemaker starts about fence_delay HOT 10 CLOSED

It seems I was able to fix this by deleting the log file at /var/log/fence_delay.log.

I'm unsure what the ownership and permissions were before, but now owner/group is root with perms -rw-r----- and starting/stopping the fence resource no longer causes the warnings.

from fence_delay.

I think I fixed this by removing op monitor interval=60m from the resource config. Nope.

from fence_delay.

Thanks for this. If/when you find the fix, I'll merge.

from fence_delay.

Reopening because after rebooting the node, the error is back! It's a really weird issue...

from fence_delay.

For the record, this is using Pacemaker 2.0.5 on Debian 11.

from fence_delay.

I've never tested this on Debian. If you find the problem though, I'll be happy to pull merge a PR.

from fence_delay.

I wish I knew enough Perl to test a modified version with logging to file switched off. Somehow I think its related to that.

from fence_delay.

Alright so I was able to reproduce the error reliably: Pacemaker needed to restart in order for it to call the metadata op (restarting the resource wasn't enough). After a bit of testing I found that the error went away for good when ownership of the log file was set to hacluster:haclient instead of root:root.

from fence_delay.

Excellent, thank you. I can probably update fence_delay to set the ownership directly.

from fence_delay.

This should hopefully be resolved now. Please note that current / future dev on fence_delay will by on the anvil repo - https://github.com/ClusterLabs/anvil/blob/master/tools/fence_delay

If the problem persists, please open an issue over there.

from fence_delay.