Red Hat Bugzilla – Bug 962088
fence_sanlock (2.6-2.el6) agent script incompatible with pacemaker 1.1.8-7.el6
Last modified: 2016-05-31 10:26:06 EDT
Description of problem:
Currently unable to use fence_sanlock as an agent with pacemaker (1.1.8), RH fencing agent api not fully implemented (monitor, list etc)
Version-Release number of selected component (if applicable):
Can be reproduced by attempting to use fence_sanlock as a stonith agent in pacemaker.
Steps to Reproduce:
1. CMAN running and using the pacemaker redirect fencing agent.
2. Pacemaker running (pacemaker-1.1.8-7.el6.x86_64), attempt to add a stonith device using the fence_sanlock agent (note that the host list, map and custom argument were included since fence_sanlock does not use the host name)
pcs stonith create st-sanlock fence_sanlock path="/dev/mapper/vg_shared-lv_sanlock" pcmk_host_list="hostname1 hostname2" pcmk_host_map="hostname1:1;hostname2:2" pcmk_host_argument="host_id
3. Check status of stonith device using pcs status and monitor logs
Device will be added to the running cib, but will fail as the monitor operation is not supported, from /var/log/messages;
May 10 01:33:42 hostname1 stonith-ng: warning: log_operation: st-sanlock:10725 [ host_id argument required ]
Expectation had been that the provided fence_sanlock agent would be compatible with pacemaker.
I'd raised this on the pacemaker mailing list to see if anyone had previously used fence_sanlock with pacemaker, and whether it was intended to be used as such. I was requested to raise a ticket here.
Wasn't sure whether to raise this against the sanlock or fence-agents component, chose sanlock since its listed as the srpm for the fence-agent package
Adding Andrew to cc list as requested
There are at least two deviations from:
- agents need to support the monitor action
- targets are to be specified with either port or nodename
I've moved this to RHEL7; RHEL6 is tech preview and if it is moved to supported it would be with some very limited validated usages, which would not include pacemaker.
Broader support would have to be considered for RHEL7. fence_sanlock is a bit peculiar, it isn't quite like other simplistic fence agents; I'm not sure how all the details would work in a pacemaker environment. We may have to add some new
As far as "monitor" goes, it's one of those things that may not make sense here.
As far as the agent args, I believe the agent currently specifies that in its
Either its a fencing agent or its not.
If it is, then it needs to conform to the published API.
Not planning any more work on fence_sanlock.
David, in that case, please consider dropping fence-sanlock subpackage
from distribution as no package is better than not well-integrated one.
(or at least, making it not discovered by pacemaker)
I suspect dropping a package is not so simple, but I'll try and see what happens.