function registration is incorrect after gearman server restart

[James E. Blair]

Steps to reproduce:

1) Start gearmand
2) Enable gearman plugin in jenkins
3) gearadmin --workers shows correct function registration
4) Disable gearman plugin
5) gearadmin --workers shows correct function registration (no functions registered)
6) Enable gearman plugin
7) gearadmin --workers shows correct function registration
8) Stop gearmand
9) Start gearmand
10) gearadmin --workers shows correct function registration
11) Disable gearman plugin
12) gearadmin --workers shows INCORRECT function registration (functions still registered)
13) Enable gearman plugin; config page hangs on saving

[Khai Do (zaro0508)]

I'm seeing something a little different between steps 9 & 10. The functions are not exactly correct.

functions from step #7:
31 50.46.196.80 master-manager : stop:15.185.117.66
32 50.46.196.80 master-exec-0 : build:pear build:pear:!ubuntu build:guava build:grapefruit build:grapefruit:air
33 50.46.196.80 oneiric-668621-exec-0 : build:peach build:peach:linux build:lemon build:lemon:linux&&gcc build:kiwi:ubuntu build:kiwi build:kiwi:centos build:guava build:apple build:apple:oneiric-668621
34 50.46.196.80 precise-668665-exec-0 : build:tangerine:torrent build:tangerine build:tangerine:centos build:pear build:pear:!ubuntu build:peach build:peach:linux build:kiwi build:mango build:kiwi:ubuntu build:mango:torrent build:guava build:kiwi:centos

functions from step #10:
32 50.46.196.80 - : stop:15.185.117.66
31 50.46.196.80 - : build:grapefruit:air build:pear:!ubuntu build:pear build:guava build:grapefruit
33 50.46.196.80 - : build:apple:oneiric-668621 build:kiwi build:lemon build:apple build:kiwi:ubuntu build:guava build:lemon:linux&&gcc build:peach:linux build:peach build:kiwi:centos
34 50.46.196.80 - : build:kiwi build:mango build:kiwi:ubuntu build:mango:torrent build:guava build:tangerine:centos build:kiwi:centos build:tangerine:torrent build:pear:!ubuntu build:pear build:tangerine build:peach:linux build:peach

The functions seem to be correct, however the worker IDs seemed to have dropped off. My initial hunch is that this is a problem with the underlying gearman-java library that does the work to re-establish a lost connection with the server. I will investigate to see wether this is indeed the problem. Will also investigate options for making the connection more robust.

[Khai Do (zaro0508)]

I was calling the unregisterAll() method before shutting down the worker. But I can't do that due to an issue in the gearman-java library: https://bugs.launchpad.net/gearman-java/+bug/708153

I've removed which seems to have fixed the NPE.

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.openstack.org/25846

[James E. Blair (corvus)]

In the latest build of gearman-plugin, in step 12, only the "stop" method is incorrectly listed, all the other methods are gone (which is correct). I also observe the worker ids issue you noted. And finally, the config screen still does hang for me at step 13.

[James E. Blair (corvus)]

This also happens when I try to trigger a build via gearman a second time.

[James E. Blair (corvus)]

Oops, strike that last comment, wrong bug.

[Khai Do (zaro0508)]

I can no longer repro after applying commit: https://review.openstack.org/#/c/26911/
Seems like this issue is fixed.