[ClusterLabs] Antw: Re: Antw: Re: Antw: [EXT] Two node cluster and extended distance/site failure

[Ulrich Windl]

>>> Andrei Borzenkov <arvidjaar at gmail.com> schrieb am 24.06.2020 um 12:42 in
Nachricht <1f4a89ce-1945-b9ab-18ee-06aaf71b15b9 at gmail.com>:
> 24.06.2020 12:20, Ulrich Windl пишет:
>>>
>>> How Service Guard handles loss of shared storage?
>> 
>> When a node is up it would log the event; if a node is down it wouldn't 
> care;
>> if a node detects a communication problem with the other node, it would 
> fence
>> itself.
>> 
> 
> So in case of split brain without witness both nodes fence itself and
> become unavailable. Which is exactly what I'd like to avoid if possible.
> 
>> But hoestly: What sense does it make to run a node if the shared storage
is
>> unavailable?
>> 
> 
> Cluster nodes may use NFS which is not suitable for SBD (although I
> wonder if shared file on NFS may work) and shared SAN storage used for
> witness only. In this case it makes all sort of sense to continue when
> witness is not available.

I think multiple hosts writing the same file via NFS does not give any
ordering guarantees for interleaved reads and writes. Still this might work.

> _______________________________________________
> Manage your subscription:
> https://lists.clusterlabs.org/mailman/listinfo/users 
> 
> ClusterLabs home: https://www.clusterlabs.org/