Re: [PATCH ha-manager v3 08/21] queue resource motion: support migration options

[Fiona Ebner <f.ebner@proxmox.com>]

Am 13.10.26 um 2:22 PM schrieb Daniel Kral:
> On Fri Sep 18, 2026 at 6:08 PM CEST, Fiona Ebner wrote:
>> @@ -651,7 +651,7 @@ sub queue_resource_motion {
>>              "crm command '$cmd' - $task service '$csid' to node '$target'"
>>                  . " (service '$csid' in positive affinity with service '$sid')",
>>          );
>> -        $ss->{$csid}->{cmd} = [$task, $target];
>> +        $ss->{$csid}->{cmd} = [$task, $target, $migrate_options];
> 
> The HA resource affinity rule could contain mixed HA resource types, so
> we cannot pass the same $migrate_options to all of them...
> 
> e.g. if ct:100 and vm:101 are in the positive resource affinity rule,
> running
> 
>     ha-manager migrate vm:101 node2 --with-conntrack-state
> 
> would make the manual migration for ct:100 fail.
> 
> (Though as 'vm:101' is successfully migrated to node2, ct:100 would
> automatically be moved to node2 by the HA Manager to make sure the
> positive resource affinity rule is fullfilled.)
> 
> 
> 
> I'm not sure how to handle this best, as it could be useful if there are
> only HA resource of the same type, but at the same time this is quite
> implicit right now.

Here, I'll just filter out unknown options for the type before passing
it along. Will need to make the schema accessible by making
migrate_json_properties() a proper plugin method.

> The migrateall API endpoint doesn't have the option to set the
> --with-conntrack-state option right now for VMs or other shared
> parameters such as --bwlimit, etc.

There, we could have a schema with all options and do the same, i.e.
just pass along the applicable ones to each type? But not going to
change that in this series.