Juju 2.0 Trunk launches disconnected nodes

Bug #1551842 reported by Tom Barber
16
This bug affects 2 people
Affects Status Importance Assigned to Milestone
Canonical Juju
Fix Released
Critical
John A Meinel

Bug Description

I have no idea how to reproduce this. Its Juju Trunk LXD compat branch and LXD 2.0 alpha 4.

Happened twice today. This time doing:

juju add-unit pdi

at which point a new container is started and juju tells me everything is installed and running, but the container itself hasn't even installed the charm in /var/lib/juju/agents.....

http://ibin.co/2YlapnApEXm7

As you can see on the screenshot, pdi/2 is considered to be running, but its ip address is 127.0.0.1

If I look at where it would have been installed on the host, its not their either. So somehow, somewhere Juju is running through the installation hooks and passing them, but seemingly nowhere I can see.

The only logs available are: http://paste.ubuntu.com/15260117/

Revision history for this message
Tom Barber (spicule) wrote :
Revision history for this message
Tom Barber (spicule) wrote :

So I decided to destroy my controller and sync up and start-afresh. This time my controller is on 127.0.0.1 which breaks the init of the first charm... something fishy going on inside LXD I think.

Revision history for this message
Tom Barber (spicule) wrote :
Revision history for this message
Tom Barber (spicule) wrote :

After a reboot, my controller is back to an IP, my first node is also assigned an ip, add-unit once again sets the IP address to 127.0.0.1 even though LXC has given it a 10.0.3. ip address

Changed in juju-core:
status: New → Triaged
importance: Undecided → Critical
milestone: none → 2.0-beta2
Changed in juju-core:
assignee: nobody → John A Meinel (jameinel)
Revision history for this message
John A Meinel (jameinel) wrote :

From the duplicate bug #1552423
I also noticed Juju is trying to connect to localhost during bootstrap, which seems to me like it'd never be the right thing to do:

Bootstrapping model "lxd" [156/953]
Starting new instance for initial controller
Launching instance
 - juju-bbeb3f29-cca8-4d8a-8ceb-c13ad3ead94e-machine-0
Building tools to upload (2.0-beta2.1-trusty-amd64)
Installing Juju agent on bootstrap instance
Waiting for address
Attempting to connect to 10.0.3.68:22
Attempting to connect to fe80::216:3eff:fe5f:f34e:22
Attempting to connect to 127.0.0.1:22
Attempting to connect to ::1:22

I wonder if this is carry over from when we had "local" provider. Or if it is just that we added filtering for all the other providers to remove "local only" addresses, and when LXD was implemented it didn't do that step.

Revision history for this message
Tom Barber (spicule) wrote :

Another quirk I noticed John, was, if I do juju deploy realtime-syslog-analytics all those nodes in the bundle get valid IP's. So, weirdly bundle deployment works fine, but single charm deployment is broken(but sometimes not)

Revision history for this message
John A Meinel (jameinel) wrote :
Revision history for this message
Tom Barber (spicule) wrote :

Patch works well!

John A Meinel (jameinel)
Changed in juju-core:
status: Triaged → Fix Committed
Curtis Hovey (sinzui)
Changed in juju-core:
status: Fix Committed → Fix Released
tags: added: 2.0-count
tags: added: juju-release-support
affects: juju-core → juju
Changed in juju:
milestone: 2.0-beta2 → none
milestone: none → 2.0-beta2
To post a comment you must log in.
This report contains Public information  
Everyone can see this information.

Duplicates of this bug

Other bug subscribers

Related blueprints

Remote bug watches

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