Manually provisioned machines with multiple networks cannot connect to API server after reboot

[Kyle Fazzari]

I have two machines with two NICs each. Each NIC is hooked up to a different network:

Network A (eth0): Internet access only. Machines cannot communicate with each other.
Network B (eth1): Local network (no internet access). Machines _can_ communicate with each other.

Both machines were manually provisioned, so I setup my Juju environment:

    $ juju generate-config
    $ juju switch manual

I then updated the config to point to one of my machines via the

    manual:
            type: manual
            bootstrap-host: 192.168.56.100
            bootstrap-user: my-user
            storage-listen-ip: 192.168.56.100

Then I bootstrapped:

    $ juju bootstrap

It ran fine, and became Machine 0. I then added the second computer to Juju:

    $ juju add-machine ssh:my-user@192.168.56.101

That also ran fine, becoming Machine 1. Juju status showed that both machines were up, and Machine 1's log had several lines like this:

    2015-01-26 14:32:38 INFO juju.state.api apiclient.go:242 dialing "wss://192.168.56.100:17070/"
    2015-01-26 14:32:38 INFO juju.state.api apiclient.go:176 connection established to "wss://192.168.56.100:17070/"

All good!

Then I rebooted both machines. Then juju status showed this:

    $ juju status
    environment: manual
    machines:
      "0":
        agent-state: started
        agent-version: 1.20.14
        dns-name: 192.168.56.100
        instance-id: 'manual:'
        series: trusty
        hardware: arch=amd64 cpu-cores=1 mem=994M
        state-server-member-status: has-vote
      "1":
        agent-state: down
        agent-state-info: (started)
        agent-version: 1.20.14
        dns-name: 192.168.56.101
        instance-id: manual:192.168.56.101
        series: trusty
       hardware: arch=amd64 cpu-cores=1 mem=994M
services: {}

Machine 1 was down. So I checked its log, and it was filled with lines like this:

    2015-01-26 14:51:12 INFO juju.state.api apiclient.go:242 dialing "wss://10.0.2.15:17070/"
    2015-01-26 14:51:12 ERROR juju.worker runner.go:218 exited "api": unable to connect to "wss://10.0.2.15:17070/"

Note that 10.0.2.15 is the bootstrap server's eth0 address into Network A, where the machines cannot communicate.

Also note:

    $ juju api-endpoints --refresh
    192.168.56.100:17070

Why does the API server address change after a reboot? How can I set which network to use for Juju?

[Curtis Hovey (sinzui)]

This bug is probably a duplicate of bug 1303204, but this issue clearly describes the problem, where as the other bug prescribes a solution. I believe fixing this bug implicitly addresses the request for the feature in bug 1303204