Subordinate charm dies when deployed with --storage option

[Jacek Nykis]

I wrote a subordinate charm with the following storage in metadata.yaml:
storage:
  srv:
    type: filesystem
    description: Srv storage
    location: /srv
    multiple:
      range: 0-1

If I deploy like this everything works:
juju deploy --repository=. local:trusty/my-charm
juju add-relation apache2 my-charm

But of course with the above storage is not used.

When I try to use storage:
juju deploy --repository=. --storage srv=1G local:trusty/my-charm

The service deploys then I add relation:
juju add-relation apache2 my-charm

And agents end ups up in the lost state:

apache2/0 unknown idle 1.25.5.1 1 10.0.3.144
  my-charm/0 unknown lost 1.25.5.1 10.0.3.144 agent is lost, sorry! See 'juju status-history my-charm/0'

And logs on the unit show this:
2016-06-23 15:22:17 ERROR juju.worker.uniter.filter filter.go:137 tomb: dying
2016-06-23 15:22:21 ERROR juju.worker.uniter.filter filter.go:137 tomb: dying
2016-06-23 15:22:24 ERROR juju.worker.uniter.filter filter.go:137 tomb: dying

juju 1.25.5-trusty-amd64

[Cheryl Jennings (cherylj)]

Can you please attach the following?

1 - /var/log/juju/machine-0.log from the state server

2 - The unit log for the primary and subordinate charm?

[Jacek Nykis (jacekn)]

I uploaded logs here:
https://private-fileshare.canonical.com/~jacek/lp1595617.tgz

Please don't share outside of the company.

[Ryan Beisner (1chb1n)]

I am experiencing death of subordinates with 1.25.6-trusty-amd64, but without use of the --storage option:

2016-07-20 16:23:47 ERROR juju.worker.uniter.filter filter.go:137 tomb: dying

cinder/4 blocked idle 1.25.6 32 8776/tcp 10.5.2.171 Services not running that should be: cinder-backup
  cinder-backup/4 unknown failed 1.25.6 10.5.2.171 Waiting for agent initialization to finish

To date, I've only observed this on Xenial deployments.

[Andrew Wilkins (axwalk)]

In 2.0-rc1 I see what I believe is the same issue, but the error message is different:

    2016-09-27 00:31:24 ERROR juju.worker.dependency engine.go:539 "uniter" manifold worker returned unexpected error: watching storage attachment: getting storage filesystem: filesystem for storage instance "filesystem/1" not found

[Andrew Wilkins (axwalk)]

https://github.com/juju/juju/pull/6336