[ClusterLabs] corosync/dlm fencing?

[Jan Pokorný]

Just minor clarifications (without changing the validity) below:

On 17/07/18 21:28 +0200, Jan Pokorný wrote:
> 
> On 16/07/18 11:44 +0200, Philipp Achmüller wrote:
>> Unfortunatly it is not obvious for me - the "grep fence" is attached
>> in my original message.
> 
> Sifting your logs a bit:
> 
>> -------------------
>> Node: siteb-2 (DC):
>> 2018-06-28T09:02:23.282153+02:00 siteb-2 pengine[189259]:   notice: Move stonith-sbd#011(Started sitea-1 -> siteb-1)
>> [...]
>> 2018-06-28T09:02:23.284575+02:00 siteb-2 crmd[189260]:   notice: Initiating stop operation stonith-sbd_stop_0 on sitea-1
>> [...]
>> 2018-06-28T09:02:23.288254+02:00 siteb-2 crmd[189260]:   notice: Initiating start operation stonith-sbd_start_0 on siteb-1
>> [...]
>> 2018-06-28T09:02:38.414440+02:00 siteb-2 corosync[189245]:   [TOTEM ] A processor failed, forming new configuration.
>> 2018-06-28T09:02:52.080141+02:00 siteb-2 corosync[189245]:   [TOTEM ] A new membership (192.168.121.55:2012) was formed. Members left: 2
>> 2018-06-28T09:02:52.080537+02:00 siteb-2 corosync[189245]:   [TOTEM ] Failed to receive the leave message. failed: 2
>> 2018-06-28T09:02:52.083415+02:00 siteb-2 attrd[189258]:   notice: Node siteb-1 state is now lost
>> [...]
>> 2018-06-28T09:02:52.084054+02:00 siteb-2 crmd[189260]:  warning: No reason to expect node 2 to be down
>> [...]
>> 2018-06-28T09:02:52.084409+02:00 siteb-2 corosync[189245]:   [QUORUM] Members[3]: 1 3 4
>> 2018-06-28T09:02:52.084492+02:00 siteb-2 corosync[189245]:   [MAIN  ] Completed service synchronization, ready to provide service.
>> [...]
>> 2018-06-28T09:02:52.085210+02:00 siteb-2 kernel: [80872.012486] dlm: closing connection to node 2
>> [...]
>> 2018-06-28T09:02:53.098683+02:00 siteb-2 pengine[189259]:  warning: Scheduling Node siteb-1 for STONITH
> 
>> -------------------
>> Node sitea-1:
>> 2018-06-28T09:02:38.413748+02:00 sitea-1 corosync[6661]:   [TOTEM ] A processor failed, forming new configuration.
>> 2018-06-28T09:02:52.079905+02:00 sitea-1 corosync[6661]:   [TOTEM ] A new membership (192.168.121.55:2012) was formed. Members left: 2
>> 2018-06-28T09:02:52.080306+02:00 sitea-1 corosync[6661]:   [TOTEM ] Failed to receive the leave message. failed: 2
>> 2018-06-28T09:02:52.082619+02:00 sitea-1 cib[9021]:   notice: Node siteb-1 state is now lost
>> [...]
>> 2018-06-28T09:02:52.083429+02:00 sitea-1 corosync[6661]:   [QUORUM] Members[3]: 1 3 4
>> 2018-06-28T09:02:52.083521+02:00 sitea-1 corosync[6661]:   [MAIN  ] Completed service synchronization, ready to provide service.
>> 2018-06-28T09:02:52.083606+02:00 sitea-1 crmd[9031]:   notice: Node siteb-1 state is now lost
>> 2018-06-28T09:02:52.084290+02:00 sitea-1 dlm_controld[73416]: 59514 fence request 2 pid 171087 nodedown time 1530169372 fence_all dlm_stonith
>> 2018-06-28T09:02:52.085446+02:00 sitea-1 kernel: [59508.568940] dlm: closing connection to node 2
>> 2018-06-28T09:02:52.109393+02:00 sitea-1 dlm_stonith: stonith_api_time: Found 0 entries for 2/(null): 0 in progress, 0 completed
>> 2018-06-28T09:02:52.110167+02:00 sitea-1 stonith-ng[9022]:   notice: Client stonith-api.171087.d3c59fc2 wants to fence (reboot) '2' with device '(any)'
>> 2018-06-28T09:02:52.113257+02:00 sitea-1 stonith-ng[9022]:   notice: Requesting peer fencing (reboot) of siteb-1
>> 2018-06-28T09:03:29.096714+02:00 sitea-1 stonith-ng[9022]:   notice: Operation reboot of siteb-1 by sitea-2 for stonith-api.171087 at sitea-1.9fe08723: OK
>> 2018-06-28T09:03:29.097152+02:00 sitea-1 stonith-api[171087]: stonith_api_kick: Node 2/(null) kicked: reboot
>> 2018-06-28T09:03:29.097426+02:00 sitea-1 crmd[9031]:   notice: Peer lnx0361b was terminated (reboot) by sitea-2 on behalf of stonith-api.171087: OK
>> 2018-06-28T09:03:30.098657+02:00 sitea-1 dlm_controld[73416]: 59552 fence result 2 pid 171087 result 0 exit status
>> 2018-06-28T09:03:30.099730+02:00 sitea-1 dlm_controld[73416]: 59552 fence status 2 receive 0 from 1 walltime 1530169410 local 59552
> 
>> -------------------
>> Node sitea-2:
>> 2018-06-28T09:02:38.412808+02:00 sitea-2 corosync[6570]:   [TOTEM ] A processor failed, forming new configuration.
>> 2018-06-28T09:02:52.078249+02:00 sitea-2 corosync[6570]:   [TOTEM ] A new membership (192.168.121.55:2012) was formed. Members left: 2
>> 2018-06-28T09:02:52.078359+02:00 sitea-2 corosync[6570]:   [TOTEM ] Failed to receive the leave message. failed: 2
>> 2018-06-28T09:02:52.081949+02:00 sitea-2 cib[9655]:   notice: Node siteb-1 state is now lost
>> [...]
>> 2018-06-28T09:02:52.082653+02:00 sitea-2 corosync[6570]:   [QUORUM] Members[3]: 1 3 4
>> 2018-06-28T09:02:52.082739+02:00 sitea-2 corosync[6570]:   [MAIN  ] Completed service synchronization, ready to provide service.
>> [...]
>> 2018-06-28T09:02:52.495697+02:00 sitea-2 stonith-ng[9656]:   notice: stonith-sbd can fence (reboot) siteb-1: dynamic-list
>> 2018-06-28T09:02:52.495902+02:00 sitea-2 stonith-ng[9656]:   notice: Delaying reboot on stonith-sbd for 25358ms (timeout=300s)
>> 2018-06-28T09:03:29.093957+02:00 sitea-2 stonith-ng[9656]:   notice: Operation 'reboot' [231293] (call 2 from stonith-api.171087) for host 'siteb-1' with device 'stonith-sbd' returned: 0 (OK)
>> 2018-06-28T09:03:29.096254+02:00 sitea-2 stonith-ng[9656]:   notice: Operation reboot of siteb-1 by sitea-2 for stonith-api.171087 at sitea-1.9fe08723: OK
>> 2018-06-28T09:03:29.096769+02:00 sitea-2 crmd[9660]:   notice: Peer siteb-1 was terminated (reboot) by sitea-2 on behalf of stonith-api.171087: OK
> 
>> -------------------
>> Node sideb-1 has no entries during this timeframe
>> 
>> during standby corosync should be up/running - so may the "Failed to 
>> receive the leave message" will be a Problem?
> 
> True, it should normally stay running, and that's actually the problem
> here and you just seem to be confusing consequence with consequence:

"consequence with the cause", apparently

> 
> - first: the corosync membership got broken on siteb-1 node
>   (hard to tell why, the only thing we can observe is that
>    sbd monitoring was moved there from sitea-1)
> 
> - then: dlm_controld on sitea-1, which already observed that membership
>   issue of siteb-1, proceed to ask pacemaker to fence siteb-1, which
>   was cheerfully executed by sitea-2 node
> 
> I cannot really tell what the root cause was here, but observing PID
> as high as 171087, I wonder if you have recent enough libqb (0.17.2):

if it's unclear, 0.17.2 as the lowest version that's fixed

> https://bugzilla.redhat.com/show_bug.cgi?id=1114852
> 
> (and would perhaps fit: new sbd process/es with high enough PIDs,

actually, crm_attribute processes arising from stonith:external/sbd

(fun fact: fence_sbd agent from fence-agents project, using a bit
different API for fencing, doesn't need to resort to relying on
stonith-timeout cluster property, so it doesn't involved such
"calling home")

> exercising libqb-backed IPC -> something bad happening like
> corosync getting stuck or crashing -> node kicked out...)

-- 
Nazdar,
Jan (Poki)
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 819 bytes
Desc: not available
URL: <https://lists.clusterlabs.org/pipermail/users/attachments/20180718/f1f17d06/attachment-0002.sig>