Juju upgrade times out, never completes

[Curtis Hovey]

13 out of 14 upgrade tests across all series, architectures, and providers fail consistently. Only aws-upgrade-precise-amd64 tends to pass. All the tests are fine under 1.20. The tests timeout after
    http://juju-ci.vapour.ws:8080/job/joyent-upgrade-precise-amd64/879/console
    http://juju-ci.vapour.ws:8080/job/hp-upgrade-trusty-amd64/1746/console
    http://juju-ci.vapour.ws:8080/job/aws-upgrade-trusty-amd64/1135/console

The last reliable revision was commit bb271ce3. Since commit ad1f35e3, the tests fail
    Merge pull request #700 from axw/state-tools-take2

[Curtis Hovey (sinzui)]

Note that machines 1 and 2 do not have logs, probably because juju forgot the machines. We can see they were fine before the upgrade.

[Menno Finlay-Smits (menno.smits)]

all-machines.log or machine[1-2].log (available for local provider upgrade tests) shows that machine-1 and/or machine-2 are unable to download tools for the upgrade. Which machine fails seems to vary.

This is likely to be related to recent changes around tools storage.