[adding sanlock-devel ML into the loop as deemed appropriate]
On 18/05/16 12:49 +0000, Da Shi Cao wrote:
> After some try and error, fence_sanlock can be used as a stonith
> resource in pacemaker+corosync.
> 1. Add a "monitor" action, which is exactly the same action as
> "status".
Per documented API (already pacemaker-aware) for fence agents:
https://fedorahosted.org/cluster/wiki/FenceAgentAPI#agent_ops
Also it looks like the resource relying on this fence agent should
declare "provides='unfencing'" as its meta-attribute(?),
but that falls into the configuration domain.
> 2. Make "status" action return "false" if a resource belongs to a
> host is acquired and owned by another host. It returned "true"
> erroneously since it didn't make a test on the owner id of a
> resource in version 3.3.0.
Might be a bug?
(I only have a very vague idea how sanlock works)
> 3. Make fence_sanlockd try for several times before it failed if the
> resource for a host is owned by another host. This gives a time
> window for the resource to be released manually at the other host.
>
> Sometimes a resource of a host get locked permanently by another
> host if the "off" action failed, often in time out.
Can you remove ambiguities in the above sentence (on which of the
machines the "off" action failed, on "another host"?), please?
(again, not that I will be very helpful afterwards, but someone could)
--
Jan (Poki)
