juju run tries to reach manually added machines through private IPs instead of public

[Laurent Sesquès]

Hi,

I manually added a machine using its public IP (which is the one juju traffic should use):
juju machine add ssh:ubuntu@machine-fqdn

Everything went fine, I deployed services successfully, etc.

Then I noticed that:
juju run --unit blah/X "pwd"
would just timeout

I saw that juju tries to reach the machine with another IP, which is a private one (10.24.0.4), not reachable from the bootstrap node.

From juju logs:
2017-10-05 13:54:38 INFO juju.worker.machiner machiner.go:132 setting addresses for machine-18 to ["local-machine:127.0.0.1" "public:<public IP>" "local-cloud:10.24.0.4" "local-machine:::1"]

I would expect juju to keep using the IP provided at the machine add stage.

This is juju 1.25.13.

Thanks.

[John A Meinel (jameinel)]

We don't currently have plans to release another 1.25. If this is critical, it can be escalated through the appropriate channels, or if it can be reproduced in the 2.2/2.3 series then we will certainly take another look.