I think that will change a lot per-project basis, if we can get that info per
job, with grafana then we can aggregate and create secondary stats (like bilds
per hour as you say).
So I'd say just to collect the 'bare' data, like job built event, job ended,
duration and such.
agree. will need to improve that, right now it 'pulls' each X seconds via the CLI,
instead of Jenkins sending the events, so it is limited to what the CLI can
provide and not that efficient. I plan to install [1] and do the opposite
(Jenkins will send a POST request with the data on each build
event and then it would be sent to graphite)

Have you checked the current ds fabric checks?
There are already a bunch of fabric tasks that monitor jenkins, if we install
the nagiosgraph (see ds for details) to send the nagios performance data into
graphite, we can use them as is to also start alarms and such
Icinga2 has integrated graphite support, so after the upgrade we will
get all of our alarms data sent to graphite 'out-of-the-box'.

    dcaro@akhos$ fab -l | grep nagi
    do.jenkins.nagios.check_build_load                      Checks if the bui...
    do.jenkins.nagios.check_executors                       Checks if the exe...
    do.jenkins.nagios.check_queue                           Check if the buil...
    do.provision.nagios_check                               Show a summary of...

Though those will not give you the bare data (were designed with nagios in
mind, not graphite so they are just checks, the stats were added later)

There's also a bunch of helpers functions to create nagios checks too.

cool, wasn't aware of those fabric checks.
I think for simple metrics(loads and such) we could use that(i.e. query Jenkins from fabric)
but for more complicated queries we'd need to query graphite itself,
with this[2] I could create scripts that query graphite and trigger Icinga alerts.
such as: calculate the 'expected' slaves load for the next hour(in graphite)
and then:
Icinga queries graphite -> triggers another Icinga alert -> triggers custom script(such as
fab task to create slaves)

for now, added two more metrics: top 10 jobs in past X time, and
avg number of builds running / builds waiting in queue in the past X time.
some metrics might 'glitch' from time to time as there is not a lot of data yet
and it mainly counts integer values while graphite is oriented towards
floats, so the data has to be smoothed(usually with movingAverage())



[1] https://wiki.jenkins-ci.org/display/JENKINS/Statistics+Notification+Plugin
[2] https://github.com/klen/graphite-beacon

On Fri, Apr 15, 2016 at 9:39 AM, David Caro <dcaro@redhat.com> wrote:
On 04/15 01:24, Nadav Goldin wrote:
> Hi,
> I've created an experimental dashboard for Jenkins at our Grafana instance:
> http://graphite.phx.ovirt.org/dashboard/db/jenkins-monitoring
> (if you don't have an account, you can enrol with github/google)

Nice! \o/

>
> currently it collects the following metrics:
> 1) How many jobs in the Build Queue are waiting per slaves' label:
>
> for instance: if there are 4 builds of a job that is restricted to 'el7'
> and 2 builds of another job
> which is restricted to 'el7' in the build queue we will see 6 for 'el7' in
> the first graph.
> 'No label' sums jobs which are waiting but are unrestricted.
>
> 2) How many slaves are idle per label.
> note that the slave's labels are contained in the job's labels, but not
> vice versa, as
> we allow regex expressions such as (fc21 || fc22 ). right now it treats
> them as simple
> strings.
>
> 3) Total number of online/offline/idle slaves
>
> besides the normal monitoring, it can help us:
> 1) minimize the difference between 'idle' slaves per label and jobs waiting
> in the build queue per label.
> this might be caused by unnecessary restrictions on the label, or maybe by
> the
> 'Throttle Concurrent Builds' plugin.
> 2) decide how many VMs and which OS to install on the new hosts.
> 3) in the future, once we have the 'slave pools' implemented, we could
> implement
> auto-scaling based on thresholds or some other function.
>
>
> 'experimental' - as it still needs to be tested for stability(it is based
> on python-jenkins
> and graphite-send) and also more metrics can be added(maybe avg running time
> per job? builds per hour? ) - will be happy to hear.

I think that will change a lot per-project basis, if we can get that info per
job, with grafana then we can aggregate and create secondary stats (like bilds
per hour as you say).
So I'd say just to collect the 'bare' data, like job built event, job ended,
duration and such.

>
> I plan later to pack it all into independent fabric tasks(i.e. fab
> do.jenkins.slaves.show)

Have you checked the current ds fabric checks?
There are already a bunch of fabric tasks that monitor jenkins, if we install
the nagiosgraph (see ds for details) to send the nagios performance data into
graphite, we can use them as is to also start alarms and such.

    dcaro@akhos$ fab -l | grep nagi
    do.jenkins.nagios.check_build_load                      Checks if the bui...
    do.jenkins.nagios.check_executors                       Checks if the exe...
    do.jenkins.nagios.check_queue                           Check if the buil...
    do.provision.nagios_check                               Show a summary of...

Though those will not give you the bare data (were designed with nagios in
mind, not graphite so they are just checks, the stats were added later)

There's also a bunch of helpers functions to create nagios checks too.


>
>
> Nadav

> _______________________________________________
> 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