[ClusterLabs] Coming in Pacemaker 3.0.1: improvements for systemd resources

[Eugen Block]

This message felt like you were reading my mind. :-D

We've been struggling for a while with an older pacemaker cluster,  
unfortunately it's a highly used production cluster on a customer  
site, so there's only limited testing possible. But this week I  
stumbled across probably the same root cause on a test cluster, where  
stopping and starting some systemd resources fails. I figured out that  
after an upgrade those failing resources had new "Wants" entries in  
their unit files:


# old unit
grep Wants /usr/lib/systemd/system/glance-api.service
Wants=network-online.target

# new unit
grep Wants /usr/lib/systemd/system/glance-api.service
Wants=postgresql.service mysql.service keystone.service ntp.service  
rabbitmq-server.service network-online.target


These are OpenStack control nodes, and their mysql is managed by  
galera for high availability. So the standalone mysql service will  
never be active here. I verified for several services (like nova-api,  
nova-conductor etc.) that simply removing mysql from the "Wants"  
statement fixed the issue for me. During research I stumbled upon this  
report (https://bugs.clusterlabs.org/show_bug.cgi?id=5404) which also  
mentioned the approach to talk to systemd via dbus.
Is it reasonable to assume that the new pacemaker version will fix  
what I described above? For now I'm thinking about using a drop-in  
file to remove mysql from the "Wants" statement for all affected units.

Thanks!
Eugen

Zitat von Chris Lumens <clumens at redhat.com>:

> Pacemaker 3.0.1, which will be released in the summer of 2025, will
> include an improvement for systemd resources.
>
> Previously, pacemaker would poll systemd resources to detect when start
> and stop actions were complete.  However, some resources may take a long
> time to stop or start, resulting in pacemaker's polling timing out and
> assuming the action failed.
>
> Now, pacemaker will talk to systemd via dbus to be signaled when an
> action is complete.  This will allow properly supporting those
> resources.  It will also allow pacemaker to more quickly react to
> changes in state of all systemd resources.
>
> There is nothing you need to do to take advantage of this improvement.
>
> - Chris
>
> _______________________________________________
> Manage your subscription:
> https://lists.clusterlabs.org/mailman/listinfo/users
>
> ClusterLabs home: https://www.clusterlabs.org/