Re: fixing jenkins builds on pull requests

[Loic Dachary <loic@dachary.org>]

[-- Attachment #1: Type: text/plain, Size: 2431 bytes --]

Hi,

I triaged the jenkins related failures (from #24 to #49):

CentOS 6 not supported:

  https://jenkins.ceph.com/job/ceph-pull-requests/26/console
  https://jenkins.ceph.com/job/ceph-pull-requests/28/console
  https://jenkins.ceph.com/job/ceph-pull-requests/29/console
  https://jenkins.ceph.com/job/ceph-pull-requests/34/console
  https://jenkins.ceph.com/job/ceph-pull-requests/38/console
  https://jenkins.ceph.com/job/ceph-pull-requests/44/console
  https://jenkins.ceph.com/job/ceph-pull-requests/46/console
  https://jenkins.ceph.com/job/ceph-pull-requests/48/console
  https://jenkins.ceph.com/job/ceph-pull-requests/49/console

Ubuntu 12.04 not supported:

  https://jenkins.ceph.com/job/ceph-pull-requests/27/console
  https://jenkins.ceph.com/job/ceph-pull-requests/36/console

Failure to fetch from github

  https://jenkins.ceph.com/job/ceph-pull-requests/35/console

I've not been able to analyze more failures because it looks like only 30 jobs are kept. Here is an updated summary:

 * running on unsupported operating systems (CentOS 6, precise and maybe others)
 * leftovers from a previous test (which should be removed when a new slave is provisionned for each test)
 * keep the last 300 jobs for forensic analysis (about one week worth)
 * disable reporting to github pull requests until the above are resolved (all failures were false negative).

Cheers

On 23/12/2015 10:11, Loic Dachary wrote:
> Hi Alfredo,
> 
> I forgot to mention that the ./run-make-check.sh run currently has no known false negative on CentOS 7. By that I mean that if run on master 100 times, it will succeed 100 times. This is good to debug the jenkins builds on pull requests as we know all problems either come from the infrastructure or the pull request. We do not have to worry about random errors due to race conditions in the tests or things like that.
> 
> I'll keep an eye on the test results and analyse each failure. For now it would be best to disable reporting failures as they are almost entirely false negative and will confuse the contributor. The failures come from:
> 
>  * running on unsupported operating systems (CentOS 6 and maybe others)
>  * leftovers from a previous test (which should be removed when a new slave is provisionned for each test)
> 
> I'll add to this thread when / if I find more.
> 
> Cheers
> 

-- 
Loïc Dachary, Artisan Logiciel Libre


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 198 bytes --]