[ClusterLabs] trace of resource - sometimes restart, sometimes not

[Ken Gaillot]

On Thu, 2022-10-06 at 21:05 +0200, Lentes, Bernd wrote:
> Hi,
> 
> i have some problems with our DLM, so i wanted to trace it. Yesterday
> i just set a trace for "monitor". No restart of DLM afterwards. It
> went fine as expected.
> I got logs in /var/lib/heartbeat/trace_ra. After some monitor i
> stopped tracing.
> 
> Today i set a trace for all operations.
> Now resource DLM restarted:
> * Restart    dlm:0                       (             ha-idg-1
> )   due to resource definition change
> I didn't expect that so i had some trouble.
> Is the difference in this behaviour intentional ? If yes, why ? Is
> there a rule ?

Yes, trace_ra is an agent-defined resource parameter, not a Pacemaker-
defined meta-attribute. Resources are restarted anytime a parameter
changes (unless the parameter is set up for reloads).

trace_ra is unusual in that it's supported automatically by the OCF
shell functions, rather than by the agents directly. That means it's
not advertised in metadata. Otherwise agents could mark it as
reloadable, and reload would be a quick no-op.

> 
> Furthermore i'd like to ask where i can find more information about
> DLM, because it is a mystery for me.
> Sometimes the DLM does not respond to the "monitor", so it needs to
> be restarted, and therefore all depending resources (which is a lot).
> This happens under some load (although not completely overwhelmed).
> 
> Thanks.
> 
> Bernd
> 
-- 
Ken Gaillot <kgaillot at redhat.com>