[Pacemaker] Announce: Making Resource Utilization Dynamic

[Andrew Beekhof]

On 12/06/2013, at 3:13 PM, Michael Schwartzkopff <misch at clusterbau.com> wrote:

> Am Mittwoch, 12. Juni 2013, 08:06:43 schrieb Andrew Beekhof:
> > On 08/06/2013, at 6:49 PM, Michael Schwartzkopff <misch at clusterbau.com> wrote:
> > > Am Donnerstag, 6. Juni 2013, 14:08:26 schrieb Andrew Beekhof:
> > > > On 06/06/2013, at 4:44 AM, Michael Schwartzkopff <misch at clusterbau.com> wrote:
> > > > > Hi,
> > > > > 
> > > > > I was not satisfied with the situation that the utilization of
> > > > > resources
> > > > > is static. This is not how real world resources behave. Especially
> > > > > virtual guests in a clustered environment show daily load patterns. So
> > > > > I
> > > > > thought it would be a good idea to make the utilization of resources
> > > > > dynamic. I worte a small patch for the VirtualDomain resource agent
> > > > > that
> > > > > updates the CPU utilization every time it monitors the resource.
> > > > > 
> > > > > Please read about the background in the following article:
> > > > > 
> > > > > http://www.sys4.de/de/blog/54/
> > > > > 
> > > > > 
> > > > > Any comments?
> > > > > Where can I send the patch to?
> > > > 
> > > > What patch :)
> > > > I see a code fragment but no context.
> > > 
> > > Patch:
> > > 
> > > --- VirtualDomain.orig  2013-06-08 10:24:54.403863313 +0200
> > > +++ VirtualDomain.new   2013-06-08 10:43:16.184253755 +0200
> > > @@ -187,6 +187,15 @@
> > > 
> > >         dom_mem=$(LANG=C virsh $VIRSH_OPTIONS dominfo ${DOMAIN_NAME} | awk
> > >         '/Max memory/{printf("%d", $3/1024)}') test -n "$dom_mem" &&
> > >         set_util_attr hv_memory "$dom_mem"
> > >      
> > >      fi
> > > 
> > > +
> > > +    if [ -x /usr/bin/virt-top ]; then
> > > +      CPUS=$(/usr/bin/virsh nodeinfo | awk '/CPU\(s/ {print $2}')
> > > +      CPU_TICKS=$(/usr/bin/virt-top -n 2 -d 5 --stream | grep
> > > $DOMAIN_NAME | tail -1 | awk '{printf "%.0f",$7}') +     
> > > OLD_TICKS=$(crm_resource -r $OCF_RESOURCE_INSTANCE -z -g ticks) +     
> > > NEW_TICKS=$(((OLD_TICKS + CPU_TICKS * CPUS) / 2))
> > > +      /usr/sbin/crm_resource -r $OCF_RESOURCE_INSTANCE -z -p ticks -v
> > > ${NEW_TICKS} +    fi
> > > +
> > > 
> > >  }
> > >  
> > >  # Set options to be passed to virsh:
> > > > Also, what does the resource/constraint config look like?
> > > 
> > > primitive resMyDomain ocf:heartbeat:VirtualDomain params config="..."
> > > 
> > > no constraints.
> > 
> > Ok, let me rephrase... is there any part of the config that makes use of the
> > values the RA defines?
>  
> Perhaps I did not get the question right. But the rest is done by pacemaker automagically. If you set the utilizations of a node to 400 ticks and a VirtualDomain resources together consume more than 400 ticks pacemaker migrates at least one of the resources away from that node.

You know that, and I know that, but not everyone reading this thread might know that :-)
They might have appreciated seeing the cib parts.

Also, I missed the "/ 2" part the first time I saw the patch, so I wasn't sure what you had in mind.

>  
> On the other hand if I set the utiliztion-strategy="minimal" pacemaker would migrate all resources to the minimal nodes possible without overloading one node. But whom do I tell ...
>  
> My idea is to switch off nodes if no resources run on that nodes and to switch on them again as utilization consumtion rises again. For that purpose I wrote a new Resource Agent that I want to publish in a next blog entry.
>  
> Given that setup you could save energy by running only as many nodes as really nescessary to deal with the utilizations of the resources.
>  
> -- 
> Dr. Michael Schwartzkopff
> Guardinistr. 63
> 81375 München
>  
> Tel: (0163) 172 50 98
> _______________________________________________
> Pacemaker mailing list: Pacemaker at oss.clusterlabs.org
> http://oss.clusterlabs.org/mailman/listinfo/pacemaker
> 
> Project Home: http://www.clusterlabs.org
> Getting started: http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf
> Bugs: http://bugs.clusterlabs.org