[ClusterLabs] cofigured trace for Virtual Domains - automatic restart ?

[Ken Gaillot]

On Sat, 2021-09-18 at 08:29 +0300, Andrei Borzenkov wrote:
> On 17.09.2021 22:13, Ken Gaillot wrote:
> > On Fri, 2021-09-17 at 15:54 +0200, Lentes, Bernd wrote:
> > > Hi,
> > > 
> > > today i configured tracing for some VirtualDomains:
> > > 
> > > ha-idg-2:~ # crm resource trace vm_documents-oo migrate_from
> > > INFO: Trace for vm_documents-oo:migrate_from is written to
> > > /var/lib/heartbeat/trace_ra/
> > > INFO: Trace set, restart vm_documents-oo to trace the
> > > migrate_from
> > > operation
> > > 
> > > ha-idg-2:~ # crm resource trace vm_genetrap migrate_from
> > > INFO: Trace for vm_genetrap:migrate_from is written to
> > > /var/lib/heartbeat/trace_ra/
> > > INFO: Trace set, restart vm_genetrap to trace the migrate_from
> > > operation
> > > 
> > > I thought "Trace set, restart vm_genetrap to trace the
> > > migrate_from
> > > operation" is a hint to not forget the restart of the resource.
> > > But all resources i configured tracing for did an automatic
> > > restart. 
> > > Is that behaviour intended ?
> > > 
> > > Bernd
> > 
> > Tracing works by setting a special parameter, which to pacemaker
> > looks
> > like a configuration change that requires a restart. 
> 
> Has it always been this way? Because crmsh obviously assumes that
> resource needs manual restart. If pacemaker does it automatically,
> crmsh
> probably should warn *before* committing change that this operation
> is
> disruptive.

Yes on all counts

> 
> > With the new OCF
> > 1.1 standard, the trace parameter could be marked reloadable, but
> > the
> > agents need to be updated to do that.
-- 
Ken Gaillot <kgaillot at redhat.com>