vsphere creates machines with multiple clashing nics

Bug #1844125 reported by Christian Muirhead
14
This bug affects 3 people
Affects Status Importance Assigned to Milestone
Canonical Juju
Fix Released
High
Christian Muirhead

Bug Description

Reported on Juju discourse https://discourse.jujucharms.com/t/vsphere-bootstrap-creates-2-nics-with-same-mac/2044

The controller machine (and subsequent deploys) creates machines with 2 nic:s with the same MAC addresses and hence the same IP-addresses. This obviously causes networking issues and I’m not sure what part juju has in this atm.

Subsequent “juju deploy foobar” equally creates machines with this very same problem (even reusing the MAC-address from the juju controller host).

Help
I’m trying to figure out if the problem is in vsphere or in juju. Since this worked fine on a previous controller (unfortunately torn down so I don’t know the version).

I bootstrap with the following:

juju bootstrap vmware01 iuba-vmware --config config.yaml --bootstrap-constraints=“mem=32G cores=8” --model-default model-default.yaml

The content of config.yaml:

primary-network: VLAN_802
datastore: H280-VIC847-0001
allow-model-access: true

The content of the model-default.yaml:

datastore: H280-VIC847-0001
primary-network: VLAN_802
apt-mirror: https://af.scania.com/artifactory/ubuntu

The resulting controller machine ends up with 2 nic:s as the picture belowScreenshot%20from%202019-09-12%2016-08-15

The MAC-addresses are the same on those NIC:s and hence also given the same IP-addresses.

This never happened before. The version of juju is: 2.6.8-bionic-amd64

snap list
juju 2.6.8 8873 stable canonical✓ classic

Anyone that has a clue on whats going wrong here?

Things I’ve tried with no avail:

Messing around with external-network and primary-network with no avail.
Messing around with templates.

Revision history for this message
Daniel Bidwell (bidwell) wrote :

I have down-graded to 2.6.6-bionic-amd64 and still have 2 nics generated instead of the requested one, but they have different mac addresses so they do function. These were deploying to non-vsan volumes.

Revision history for this message
Daniel Bidwell (bidwell) wrote :

Someone told me where the templates are for booting new vms on vsphere, but I can't find the email now. Could someone point me at them? Thanks.

Revision history for this message
Christian Muirhead (2-xtian) wrote :

Hi Daniel - the template can be found in the juju-vmdks/<controller-uuid>/<series> folder.

Revision history for this message
Christian Muirhead (2-xtian) wrote :

Could you please run the bootstrap with --debug and post the output on a pastebin? There's some debug logging for the generated network devices that would be really useful to see. We're not having any luck reproducing the problem on our vsphere so far but still trying to work out what's going on.

Revision history for this message
Christian Muirhead (2-xtian) wrote :

Also, can you tell us what OS+version you're bootstrapping on?

Revision history for this message
Erik Lönroth (erik-lonroth) wrote :

A detail I mentioned in the discourse post is that, when providing "external-network" + "primary-network" the controller gets 4 (four) nics. Maybe that would help you track down where this happens.

Revision history for this message
Christian Muirhead (2-xtian) wrote :

Ok, we can see the same behaviour when bootstrapping from windows. We're chasing down the cause of the problem now.

Revision history for this message
Daniel Bidwell (bidwell) wrote :

nohup juju bootstrap myvtestcloud myvtestvsancontroller --config primary-network=Netjuju --config datastore=vsanDatastore --debug

the contents of nohup.out can be found at https://paste.ubuntu.com/p/WrTVHjhxQP/

I am running Ubuntu 18.04.3 LTS with Linux juju 4.15.0-62-generic #69-Ubuntu SMP Wed Sep 4 20:55:53 UTC 2019 x86_64 x86_64 x86_64 GNU/Linux

The vSphere is running vmware 6.7

I should be bootstraping on bionic beaver.

Anything else I can help with?

Revision history for this message
Christian Muirhead (2-xtian) wrote :

It seems like this is fixed in the latest version of the 2.6 branch (which will be released shortly) - would you be able to test this using the edge snap?

Changed in juju:
status: In Progress → Incomplete
Revision history for this message
Daniel Bidwell (bidwell) wrote :

Yes, it works!

Any chance we could get the ability to migrate a default model to a named model?

Revision history for this message
Erik Lönroth (erik-lonroth) wrote :

I can test the edge version Christian, still edge version 2.6.8?

Revision history for this message
Richard Harding (rharding) wrote : Re: [Bug 1844125] Re: vsphere creates machines with multiple clashing nics

It should be 2.6.9.

On Sun, Sep 22, 2019 at 4:00 AM Erik Lönroth <email address hidden> wrote:

> I can test the edge version Christian, still edge version 2.6.8?
>
> --
> You received this bug notification because you are subscribed to juju.
> https://bugs.launchpad.net/bugs/1844125
>
> Title:
> vsphere creates machines with multiple clashing nics
>
> To manage notifications about this bug go to:
> https://bugs.launchpad.net/juju/+bug/1844125/+subscriptions
>

Changed in juju:
milestone: 2.6.9 → 2.6.10
Changed in juju:
milestone: 2.6.10 → 2.6.11
Changed in juju:
status: Incomplete → Fix Released
Changed in juju:
milestone: 2.6.11 → 2.6.9
To post a comment you must log in.
This report contains Public information  
Everyone can see this information.

Other bug subscribers

Remote bug watches

Bug watches keep track of this bug in other bug trackers.