[pve-devel] [PATCH ha-manager v2 2/2] cleanup backup & mounted locks after recovery (fixes #1100)

[Thomas Lamprecht]

On 09/13/2016 07:26 AM, Dietmar Maurer wrote:
>>   # after a node was fenced this recovers the service to a new node
>>   my $recover_fenced_service = sub {
>>       my ($self, $sid, $cd) = @_;
>> @@ -266,6 +289,8 @@ my $recover_fenced_service = sub {
>>   
>>   	$haenv->steal_service($sid, $sd->{node}, $recovery_node);
>>   
>> +	&$after_recovery_cleanup($self, $sid);
>> +
> I would prefer an earlier check in steal_service(). We can then avoid
> moving config files if a resource has a lock, or use local devices?

The thought behind doing this only after we got control over the service 
was to avoid situations
where a lock was removed, but then - e.g. as a result of quorum loss - 
the service is still on the
failed node but now without lock, which seems not ideal to me.

Doing a check upfront would be reasonable to me, but there cleaning 
already up there seems as it could introduce also problems.