I enhanced this[1] graph to compare slaves utilization vs build queue, note that
the slaves utilization is measured in percentages and the number of builds in the queue
is absolute. basically when the red lines are high(large queue size) and the green ones(slaves
utilization) are low we could have possibly had more builds running. we can see in the past few days
we've reached nice utilization of ~ 90% and following that the queue size decreased pretty
quickly, on the other hand there were times of only 16% utilization and a large queue ~ 70.
last I checked the least significant problem is the OS as most standard-ci jobs
are agnostic to EL/FC, usually it was the jobs limit, or sudden peeks in patches sent
but I didn't get to add 'reason each job is waiting' metric yet, so its just a feeling.

maybe the Priority Sorter Plugin[2] which comes bundled with Jenkins
could address the problem of jobs waiting 'unfairly' a long time in the queue,
though it will require to define the priorities in the yamls.



[1] http://graphite.phx.ovirt.org/dashboard/db/jenkins-monitoring?panelId=16&fullscreen&from=1462654800000&to=1462966158602&var-average_interval=12h&var-filtered_labels=All&var-filtered_jobs_labels=All
[2] https://wiki.jenkins-ci.org/display/JENKINS/Priority+Sorter+Plugin

On Wed, May 11, 2016 at 1:43 PM, Sandro Bonazzola <sbonazzo@redhat.com> wrote:


On Wed, May 11, 2016 at 12:34 PM, Eyal Edri <eedri@redhat.com> wrote:
From what I saw, it was mostly ovirt-engine and vdsm jobs pending on the queue while other slaves are idle.
we have over 40 slaves and we're about to add more, so I don't think that will be an issue and IMO 3 per job is not enough, especially if you get idle slaves.


+1 on raising then.

 
We are thinking on a more dynamic approach of dynamic vm allocation on demand, so in the long run we'll have more control over it,
for now i'm monitoring the queue size and slaves on a regular basis [1], so if anything will get blocked too much time we'll act and adjust accordingly.



On Wed, May 11, 2016 at 1:10 PM, Sandro Bonazzola <sbonazzo@redhat.com> wrote:


On Tue, May 10, 2016 at 1:01 PM, Eyal Edri <eedri@redhat.com> wrote:
Shlomi,
Can you submit a patch to increase the limit to 6 for (i think all jobs are using the same yaml template) and we'll continue to monitor to queue and see if there is an improvement in the utilization of slaves?

Issue was that long lasting jobs caused queue to increase too much.
Example: a patch set rebased on master and merged will cause triggering of check-merged jobs, upgrade jobs, ...; running 6 instance of each of them will cause all other projects to be queued for a lot of time.

 

E.

On Tue, May 10, 2016 at 1:58 PM, David Caro <dcaro@redhat.com> wrote:
On 05/10 13:54, Eyal Edri wrote:
> Is there any reason we're limiting the amount of check patch & check merged
> jobs to run only 3 in parallel?
>

We had some mess in the past where enabling parallel runs did not really force
not using the same slave at the same time, I guess we never reenabled them.

> Each jobs runs in mock and on its own VM, anything presenting us from
> removing this limitation so we won't have idle slaves while other jobs are
> in the queue?
>
> We can increase it at least to a higher level if we won't one specific job
> to take over all slaves and starve other jobs, but i think ovirt-engine
> jobs are probably the biggest consumer of ci, so the threshold should be
> updated.

+1

>
> --
> Eyal Edri
> Associate Manager
> RHEV DevOps
> EMEA ENG Virtualization R&D
> Red Hat Israel
>
> phone: +972-9-7692018
> irc: eedri (on #tlv #rhev-dev #rhev-integ)

> _______________________________________________
> Infra mailing list
> Infra@ovirt.org
> http://lists.ovirt.org/mailman/listinfo/infra


--
David Caro

Red Hat S.L.
Continuous Integration Engineer - EMEA ENG Virtualization R&D

Tel.: +420 532 294 605
Email: dcaro@redhat.com
IRC: dcaro|dcaroest@{freenode|oftc|redhat}
Web: www.redhat.com
RHT Global #: 82-62605



--
Eyal Edri
Associate Manager
RHEV DevOps
EMEA ENG Virtualization R&D
Red Hat Israel

phone: +972-9-7692018
irc: eedri (on #tlv #rhev-dev #rhev-integ)

_______________________________________________
Infra mailing list
Infra@ovirt.org
http://lists.ovirt.org/mailman/listinfo/infra




--
Sandro Bonazzola
Better technology. Faster innovation. Powered by community collaboration.
See how it works at redhat.com



--
Eyal Edri
Associate Manager
RHEV DevOps
EMEA ENG Virtualization R&D
Red Hat Israel

phone: +972-9-7692018
irc: eedri (on #tlv #rhev-dev #rhev-integ)



--
Sandro Bonazzola
Better technology. Faster innovation. Powered by community collaboration.
See how it works at redhat.com

_______________________________________________
Infra mailing list
Infra@ovirt.org
http://lists.ovirt.org/mailman/listinfo/infra