[ClusterLabs] Antw: Re: Antw: Re: SLES11 SP4:SBD fencing problem with Xen (NMI not handled)?

Ulrich Windl Ulrich.Windl at rz.uni-regensburg.de
Wed Aug 1 09:13:04 UTC 2018


>>> Klaus Wenninger <kwenning at redhat.com> schrieb am 01.08.2018 um 08:28 in
Nachricht <5149ea3c-3c14-57be-034e-4f1e0d8fb01e at redhat.com>:
> On 08/01/2018 08:06 AM, Ulrich Windl wrote:
>> Hi Klaus,
>>
>> sorry for the late response, but in the meantime I found out some more 
> facts:
> np
>> 1) Triggering of the watchdog was most likely triggered by inconsistent SBD
>> timing parameters (mainly due to incomplete SBD documentation (I'll write
>> another thread on that soon))
>> 2) When using softdog instead of hpwdt, no fencing happened (without having
>> resolved item 1)).
> that is strange ... either it should have triggered
> because of 1) and you've hit a case where softdog
> doesn't work that well or hpwdt is somehow
> malfunctioning so that kicking it doesn't always
> work - somehow contradicts your experience after
> having solved 1) though
>> 3) For reasons unknown to me, SBD of SLES11 SP4 lacks the test-watchdog
>> command.
> my guess is that as I've quite recently implemented that
> SLES11 SP4 is just too old for having it in. would guess
> Suse is patching in fixes but no new features.
>>
>> At the moment we run hpwdt with updated SBD timing parameters, so there is 
> no
>> reason for the watchdog to trigger. However if it triggers, the (Xen) kernel
>> will probably freeze again...
> are you running sbd inside xen-vms or on dom0?
> for running sbd inside xen-vms what would be
> speaking against using xen_wdt as suggested by
> the cytrix-guy?

We run the Cluster and SBD in Dom0 (which is also a kind of virtual machine).

Regards,
Ulrich





More information about the Users mailing list