[ClusterLabs] How does failure-timeout works, will the resource not be scheduled when setting too short?

[Ken Gaillot]

On Sun, 2018-05-20 at 10:19 +0800, lkxjtu wrote:
> 
> I have two pacemaker resources. We call them A and B. Because of
> environmental reasons, their start methods and monitor methods always
> return failure 
> 
> (OCF_ERR_GENERIC). The following are their configurations:(The
> cluster property of start-failure-is-fatal is false)
> 
> primitive A A \
>         op monitor interval=20 timeout=120 \
>         op stop interval=0 timeout=120 on-fail=restart \
>         op start interval=0 timeout=240 on-fail=restart \
>         meta failure-timeout=60s
> primitive B B \
>         op monitor interval=20 timeout=120 \
>         op stop interval=0 timeout=120 on-fail=restart \
>         op start interval=0 timeout=240 on-fail=restart \
>         meta failure-timeout=60s
> clone A_cl A
> clone B_cl B
> 
> The time consuming of their methods is different:
> A:
> start = 60s       monitor < 1s        stop = 80s
> B:
> start < 1s        monitor < 1s        stop < 1s     
> 
> Resource of A is scheduled normally, always start and stop. But for
> resource B, there is only circular monitor fails, without start and
> stop.
> . And there is no fail-count showing of B in "crm status -f". 
> 
> Two operations can solve the problem of B not being scheduled:
> 1，Set failure-timeout of B from 60s to 600s
> 2，Modify ocf of A，make the stop method return as soon as possible
> 
> I tested it several times, and the results were the same. Why does
> the resource not be scheduled when failure-timeout setting too short?
> And what does 
> 
> it have to do with the time consuming stop of another resource?  Is
> this a bug?
> 
> My pacemaker version is 1.1.16. Any suggestion is welcome. Thank you!
> 
> 
> James
> 2018-05-20

That behavior is unexpected. Can you share logs?
-- 
Ken Gaillot <kgaillot at redhat.com>