27.11.2018 3:50, Ken Gaillot пишет:
I don't know if a cleanup would help.
Currently, it helps.
My first thought is to set the target-role to Slave, then back to
Master. This would of course leave the cluster with no master for a
period of time, but I think it would reschedule the monitors.
I think I found more effective solution. At this moment I have a script
that runs via cron every 5 minutes on all machines and it does the
following actions:
- runs `lrmadmin -L` and filters all Master/Slave clones running on the host
- for every Master/Slave clone the script runs `lrmadmin -S <rsc_id>`
and detects if the clone has no operations and its state is LRM_RSC_IDLE
- for every idle rsc_id with no operations it runs `crm_resource -C -r
<rsc_id>`
This way the cleanup operation runs only if it's needed and Pacemaker
only refreshes its internals (lrm, cib) without stopping/demoting
resources. In fact, it tries to start a resource, so Pacemaker executes
monitor op, figures out the resource is already in running-master state
and does nothing else
--
С уважением,
Субботин Никита Андреевич,
Сибирский Федеральный Университет
_______________________________________________
Users mailing list: Users@clusterlabs.org
https://lists.clusterlabs.org/mailman/listinfo/users
Project Home: http://www.clusterlabs.org
Getting started: http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf
Bugs: http://bugs.clusterlabs.org