juju-tools 1.25.0 with juju-agent 1.25.3 has networking issues

[Ali]

I've been using an install of juju-core 1.25.0 and upon bootstrapping a new environment juju automatically upgraded the agent to 1.25.3. I've been running into issues with the script that creates juju-br0 and restarts the interface. When the new version of that script is executed for 1.25.3 the client the instance doesn't have working networking. executing ifconfig eth0 down; ifup eth0 resolves the issue, but the actions that cloud-init was trying to do have already failed. This issue doesn't appear to occur when the agent and tools version is matched (the bridge works and there is connectivity). With juju automatically picking up new patch releases of the tools, however this is a problem.

[Cheryl Jennings (cherylj)]

You can pin the agent version, and prevent the upgrade by using 'agent-version: 1.25.0' in your environments.yaml.

There are some known networking issues with 1.25.3 on MAAS that we are currently investigating. I think I'm running into this same issue trying to deploy with wily (was able to deploy fine with trusty).

[Cheryl Jennings (cherylj)]

Spoke with frobware who mentioned there may also be problems with the bridge script in upgrade scenarios.

[Andrew McDermott (frobware)]

If you still have the environment around please could you attach all of:

 /etc/network/interfaces*

to the bug. Thanks.

[Andrew McDermott (frobware)]

If you still have the environment booted please could you attach all of:

 /etc/network/interfaces*

to the bug. Thanks.

[Ali (asaidi)]

Hi Andrew,

Sorry I don't have it anymore, however it all looked 'correct' juju-br0 was there, set to auto, with the correct bridge port, address and gateway. enp1s0f1 was there and set to manual. Restarting the system resulted in networking working correctly, however at that point juju didn't send any status command to the orchestration node, so it was stuck waiting for agent initialization.

Ali

[Andrew McDermott (frobware)]

Please could you summarise your juju commands.

If I understood this you had en environment already bootstrapped but an upgrade to .3 made the network script fail. Is that correct? Or you bootstrapped a new environment but the tools had been upgraded to 1.25.3 (presumably an apt-get update) and the new environment failed to bootstrap. I'm trying to understand whether this was an upgrade failure or a new deployment failure. Thanks.

[Andrew McDermott (frobware)]

Hi Ali,

Which version of MAAS were you using? (1.9, 1.8.2?)

[Ali (asaidi)]

Hi Andrew,

You're correct, I had an environment already bootstrapped with 1.25.0 on the orchestration node. and juju-tools 1.25.0. I added another node to the environment and it installed juju 1.25.3 and the network script failed to bring up the network as described above. Rebooting the instance or manually if up/down the interface solved the issue with connectivity although the node never became part of the environment.

Bootstrapping a new environment with --no-auto-upgrade and adding a machine works fine, however adding a node to the existing one fails.

Also, bootstrapping a new environment with 1.25.3 and then adding new machine (which is 1.25.3) works as well.

In both cases MaaS 1.9 is the provider.

Ali

[Andrew McDermott (frobware)]

I am not looking into this bug at the moment.

[Anastasia (anastasia-macmood)]

Our current 1.25 version is 1.25.6. Also 1.25 series is "critical" bugs only mode.

Our next release will be 1.25.7.

If you are experiencing this failure with 2.0.x, please provide newer logs.