[Pacemaker] dependent resource reach max fail count

[Andrew Beekhof]

On Sat, Jul 3, 2010 at 5:10 AM, Michael Fung <mike at 3open.org> wrote:
> On 2010/7/2 下午 09:16, Michael Fung wrote:
>>
>> It seems to indicate that pacemaker do not shutdown dependent resources
>> in an orderly manner when the multi-state depended resource change
>> state. The dependent resources just "crashed". Am I right?
>>
>
> I was wrong.
>
> From the extracted log file, lrmd stopped the dependent resources orderly:
>
> lrmd: [806]: info: rsc:ve1011:119: stop
> lrmd: [806]: info: rsc:vz_svc:121: stop
> lrmd: [806]: info: rsc:vz_fs:122: stop
> lrmd: [806]: info: rsc:drbd_r0:0:123: demote
> lrmd: [806]: info: RA output: (drbd_r0:0:demote:stdout)
> lrmd: [806]: info: rsc:drbd_r0:0:124: notify
> lrmd: [806]: info: RA output: (drbd_r0:0:notify:stdout)
> lrmd: [806]: info: rsc:drbd_r0:0:125: notify
> lrmd: [806]: info: rsc:drbd_r0:0:126: notify
> lrmd: [806]: info: RA output: (drbd_r0:0:notify:stdout)
> lrmd: [806]: info: rsc:vz_fs:127: start
> lrmd: [806]: info: RA output: (vz_fs:start:stderr) /dev/drbd0: Wrong
> medium type
>
> But the problem is, at the second last line, we see lrmd was trying to
> start vz_fs while it already demoted drbd_r0. This results in error and
> probably lead to vz_fs fail-counts set as INFINITY. Isn't this behavior
> violates:
>  order ms_drbd_r0-b4-vz_fs inf: ms_drbd_r0:promote group_vz:start
>  group group_vz vz_fs vz_svc ve1011
>
> Any ideas?
>

We fixed a couple of things that might be related in 1.0.9.
Perhaps give that a try?