juju takes a long time to deploy using the local provider

[Ricardo Kirkner]

When running juju on canonistack (flavor=1137) using the local provider it takes over 20 minutes to deploy a simple unit (like the wordpress charm).

As the initial unit fails (see bug #1231686) even though the lxc template gets properly created, the second time you try to deploy the unit it will pick up the template and go straight to deploy the unit.

Steps to reproduce:

- create an instance on canonistack lcy02 region

- deploy a unit
juju bootstrap
juju deploy wordpress
(this fails, as stated in bug #1231686)

- reset the environment
juju destroy-environment local -y

- retry
juju bootstrap
juju deploy wordpress

Adding an strace dump to aid in diagnosing the issue (it looks like most the waiting time is spent in epoll_wait related to veth).

[Curtis Hovey (sinzui)]

I don't think this is a bug with juju. Canonistack is ill. Juju CI has tested on the cloud for 6 moths.

3 months ago a simple wordpress stack could be stood up in 5 minutes. Juju CI was located in canonistack at the time and the lxc tests took about 5 minutes.

But when canonistack became over-resourced, Juju CI was moved to a different cloud. We will test in canonistack.
It now take 25 minutes to do *less* work than it used to do in 5 minutes. We do less work by not deploying wordpress since it doesn't work on private clouds.
    http://juju-ci.vapour.ws:8080/view/Providers/

[Ricardo Kirkner (ricardokirkner)]

After the UOS session on using the local provider, I've started using the juju-clean plugin, and this solved all the issues I was having with the local provider on my laptop (the problem was evidently some unclean state left over, quite possibly some very old cloud images in the cache).