lxd provider: machines stuck at "pending"

[Casey Marshall]

Sometimes when I try deploying on the lxd provider, the machines never provision.

See https://paste.ubuntu.com/14852745/, contains juju status, lxc list and the controller log file.

Rebooting the controller usually recovers from this condition (juju ssh 0 "sudo reboot").

[Eric Snow (ericsnowcurrently)]

This log entry seems to be a good clue:

cannot rename temporary config file: open /.config/lxc/config.yml.new: no such file or directory

[John A Meinel (jameinel)]

If the bug really is: /.config/lxc/config.yml.new: no such file or directory
there was a buggy race condition where we were creating a config file in root /.config/... and further we were doing something buggy internally that was causing us to create lots of them as:
/.config/juju-lxc/juju-lxc/juju-lxc/...
(I saw as deep as 6).
Potentially this would be multiple threads trying to initialize juju-lxc at the same time.

however, the latest version of the provider gets rid of the disk config entirely, and we feed it the information it needs directly from state, instead of caching it.
So I'd guess this is Fixed.