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

["Daniel Kral" <d.kral@proxmox.com>]

On Fri Mar 27, 2026 at 12:15 AM CET, Thomas Lamprecht wrote:
> Am 26.03.26 um 17:02 schrieb Daniel Kral:
>> Two other possible cases of 'implicitly transient' states might be:
>> 
>> - adding a HA rule, which makes a HA resource in 'started' state be put
>>   in 'migrate' to another node when processing the select_service_node()
>>   in next_state_started().
>> 
>> - the node of a HA resource is offline delayed in the same round as the
>>   disarm request. If none of the HA resources are in a transient state
>>   yet, the disarm request goes through, otherwise the affected HA
>>   resources might be put in 'fence'.
>> 
>> 
>> I haven't thought this through fully, but an option might be that we
>> only allow the FSM processing of the HA resources, which are in one of
>> these 4 transient states and don't process the others.
>> 
>> E.g. breaking out the FSM transition loop into its own function and in
>> normal operation we iterate through all services in $ss, but for
>> deferred disarming we only iterate through the HA resources in transient
>> states, which should be resolved.
>
> I pushed a follow-up [0] that should deal with this, another look at
> that would be appreciated!
> FWIW, I mostly pushed directly as I still wanted to do a bump+test upload
> today, because if everything is good we get a small regression fix faster
> out to users, if not we can follow-up here and no harm done.
>
> [0]: https://git.proxmox.com/?p=pve-ha-manager.git;a=commitdiff;h=b6b025a268032ff5302bede1f5eb56247af13f21
>
> [...]
>

Thanks for the quick patch! The changes look good to me too and the test
case captures the behavior well!