Using MAAS and --constraints "maas-name=$SPECIFIC_HOST" juju will deploy when that host is in commissioning or some other non-ready state

[David Ames]

Using MAAS and --constraints "maas-name=$SPECIFIC_HOST" juju will deploy when that host is in commissioning or some other non-ready state without erroring or warning.

In juju status the machine state shows as pending indefinitely even when the machine specified becomes available in a ready state.

[Kapil Thangavelu (hazmat)]

This seem like a maas issues, juju doesn't interpret constraints, it merely provides them to the provider to do the 'right' thing. On the other end (status, post deploy), juju is directly querying the maas server for status of the host using the value given by the maas server for deploy/add-unit.

[Kapil Thangavelu (hazmat)]

constraint validation on the juju side for maas is limited at the moment to vocabulary validation against tags. the maas-name imo is a deprecated constraint in that its effectively broken from a capabilities perspective... ie you can never add-unit to a service deployed with a name-constraint. none the less maas should never return a provisioning error if no machines matching constraints are in a ready state/good state imo.

[John A Meinel (jameinel)]

I don't think this is specific to --maas-name. The issue is that you go to deploy and no nodes can match your requested constraints. So juju gets 0 nodes available for that request (right now). When the node becomes ready later, it doesn't seem that juju is retrying the deploy request?

[Martin Packman (gz)]

Short version: don't use the maas-name constraint.

So, only nodes in the READY state are offered for acquisition. If tell juju to only deploy to a machine with a particular node, and that machine does not exist, or is not yet ready, it will get NodesNotAvailable back from maas. This is fine, apart from that error does not propagate from the provisioning agent back to the client, so without looking at the log on the master node you have no idea anything went wrong. See the existing bug for more discussion.