[ClusterLabs] How to correctly stop cluster with active stonith watchdog?

[Andrei Borzenkov]

30.04.2019 9:53, Digimer пишет:
> On 2019-04-30 12:07 a.m., Andrei Borzenkov wrote:
>> As soon as majority of nodes are stopped, the remaining nodes are out of
>> quorum and watchdog reboot kicks in.
>>
>> What is the correct procedure to ensure nodes are stopped in clean way?
>> Short of disabling stonith-watchdog-timeout before stopping cluster ...
> 
> Do you want the cluster to continue to operate, just with less nodes? If
> so, you probably want "last_man_standing: 1" in corosync/votequorum.
> 

It does not work for two node cluster (I do not mean necessary
two_node=1). Only one of two nodes may be quorate in this case, which
returns us to possible race condition in case of (unattended) shutdown.

It is the same as with qdevice - everything works as long as we can
ensure qdevice server remains up and reachable until all cluster nodes
are stopped.