Juju 2.9.9 fails to bootstrap on AWS

[Simon Fels]

I am trying to bootstrap a new controller with 2.9.9 on AWS but always fail with

10:41:38 ERROR juju.cmd.juju.commands bootstrap.go:883 unable to contact api server after 1 attempts: unable to connect to API: dial tcp 172.31.40.101:17070: connect: network is unreachable

The client tries to connect the created controller instance on both it's public and private address but seems to directly fail on the first attempt. I attached the entire debug log of the `juju bootstrap` command.

I get this on two different AWS accounts so it doesn't seem to be related to the VPC setup within the account. In addition I also tried to bootstrap in a totally clean environment with Juju fresh installed, same problem happened.

Any idea what is wrong here?

[Ian Booth (wallyworld)]

The issue is there's a whitelist of errors which are considered non fatal when polling the controller API address to validate that the controller agent is alive at the end of bootstrap. Any error not in this list is considered fatal. And "network is unreachable" is not currently in the list so the bootstrap process exits with an error. We normally see "i/o timeout" or "connection refused". We just need to add "network is unreachable" to the list.

[Ian Booth (wallyworld)]

I can't reproduce locally as I never get that same underlying error but it should be a one line fix
https://github.com/juju/juju/pull/13196

[Simon Fels (morphis)]

Thanks Ian for the quick fix!

Should I be able to test this with the latest edge snap?