On 4 January 2017 at 14:35, Vance Morris <vmor...@us.ibm.com> wrote:
> It's worse than a waste: it managed to confused the hell out of me. I had no
> idea that there even was a thing called "juju-deployer" that was separate
> from "juju deploy".
>
> Seems to me that if a model already has machines 0 and 1 deployed, and a
> bundle is deployed that refers to machines 0, 1 and 2, Juju ought to
> automatically add a 3rd machine (2) and go to work deploying the
> applications. Why should the bundle care if a machine exists prior to
> deployment or not?
>
> - Vance
The idea behind the bundle format is that the deploy command should
provide some way
to map from the machines in the bundle to existing machines in the
model. That feature isn't
implemented yet AFAIK.
Given that machine numbers are essentially ephemeral (if you need to
restart a machine,
you'll never get the same machine number again), I'm not sure it makes sense to
bake machine numbers into a bundle. The place that this was done historically
was to refer to machine 0 (the initial bootstrap machine), but that's not
available any more unless you're on the admin model and not even then
if HA has restarted some controller nodes.
Another possible solution might be to provide a dep option to
automatically map machine numbers
in the bundle to existing machines in some way. For example, it could
choose machines with no existing
units deployed on them.
cheers,
rog.
--
Juju mailing list
Juju@lists.ubuntu.com
Modify settings or unsubscribe at:
https://lists.ubuntu.com/mailman/listinfo/juju
