juju-core broken with OpenStack Havana for tenants with multiple networks

This is impacting actual users, and given that we need to have a place to test the work -- it should be something we try to get done as quickly as possible.

Martin, I think IS will be very interested in helping us test this work, so feel free to contact them about this work.

Thanks!

Please can we have an update on plans to fix this issue; its impacting havana cloud's and I keep getting asked about it :-)

"havana cloud's" ?

People deploying onto an Openstack Havana cloud?

Anyway, I'm guessing we can do some of the bits that we're doing with VPC on AWS, but it depends on the details. One of the primary needs to actually fix this is to have a cloud where we can test it and actually know it is fixed. Is there one available? Is lcy02 upgraded to Havana?

To clarify, my question is whether this is people deploying a Havana cloud, or people using Juju to deploy *onto* a Havana cloud.

And what specific configuration of all the different bits of those clouds, so that we can reproduce and actually fix the problem they are experiencing.

AFAIK there aren't any *public* clouds that are running Havana.

John - the answer is people deploying onto havana based clouds.

The juju-core team have access to a havana based cloud with neutron networking - ping me on irc for details.

See also lp:1188126

We have to add a field in openstack juju config which allows you to set a default network interface to use.

As I understand from Martin, the issue is that if you have 2 networks available in Havana, Openstack doesn't define which one is "default". So we need to add an environments.yaml level configuration which allows the user to specify which network to bind to. And then this needs to chain down into some lp:goose level changes to accept a network and configure it.

FWIW we've been experiencing this bug with our new Openstack cloud, and when I upgrade to 1.17.2, the error message goes away.

Do we have any idea on a release timeframe for this bug into a stable release? Ie, either when will 1.18 be out, or will there be a backport to 1.16? This is a critical issue for our new production cloud to be able to go live.

I tested it today and this was still a problem:
{"badRequest": {"message": "Multiple possible networks found, use a Network ID to be more specific.", "code": 400}}

I went through the diff and found undocumented "network" option, after adding it to my environments.yaml it worked.
I think this bug should stay as "Fixed Commited" until work is finished off. For average user who does not want to read source code this is still broken.

I think we should:
1. Update docs
2. Provide better error message. Instead of returning json maybe print "Please specify network using 'network' option in environments.yaml"

And if we want juju to "just work" it could check which network is private and try using it first before giving up.

You may find it useful to read the message I posted to the list, describing why the 'networks' config workaround was used here.

<https://lists.ubuntu.com/archives/juju/2014-January/003392.html>

I submitted a patch for the docs, I'll chase Nick to get it landed.

We don't have great options on making the error message better, as what we get from Openstack isn't a specific exception type, so it'd be detecting that string and rewriting it... which is somewhat vulnerable to changes their end.