Re: [PATCH ha-manager v2 3/4] fix #2751: implement disarm-ha and arm-ha for safe cluster maintenance

[Thomas Lamprecht <t.lamprecht@proxmox.com>]

Am 25.03.26 um 16:49 schrieb Fiona Ebner:
> Am 22.03.26 um 12:57 AM schrieb Thomas Lamprecht:
>> +    if ($mode eq 'freeze') {
>> +        for my $sid (sort keys %$ss) {
>> +            my $sd = $ss->{$sid};
>> +            my $state = $sd->{state};
>> +            next if $state eq 'freeze'; # already frozen
>> +            if (
>> +                $state eq 'started'
>> +                || $state eq 'stopped'
>> +                || $state eq 'request_stop'
>> +                || $state eq 'request_start'
>> +                || $state eq 'request_start_balance'
>> +                || $state eq 'error'
> Should it really happen for the 'error' state too? Because when
> re-arming, the state will become 'started':
> 
> Mar 25 16:20:06 pve9a1 pve-ha-crm[242553]: disarm: freezing service
> 'vm:400' (was 'error')
> ...
> Mar 25 16:20:36 pve9a1 pve-ha-crm[242553]: service 'vm:400': state
> changed from 'freeze' to 'started'
> 
> Which feels rather surprising to me. For comparison, after a cold
> cluster start, services in 'error' state are not (attempted to be)
> started either.

You're right, thanks for noticing this, fixed in:
https://git.proxmox.com/?p=pve-ha-manager.git;a=commitdiff;h=890673383c27a2949315a306218ef035042b46b0;hp=b6b025a268032ff5302bede1f5eb56247af13f21

Albeit, I'd like to revisit the error state and its being a final
status in the FSM, as somethings might be often better to just be
retried with a rate-limit forever in the context of HA. Anyhow,
definitively orthogonal to this series.