Zombie builds stuck in FileDownload

[catlee]

We have slaves running buildbot 0.7.9 attached to a master running 0.7.10. On occasion, a slave will get disconnected from the master while doing a FileDownload? step.

When it reconnects to the master, the master notices a duplicate connection, and attempts to disconnect the old slave. It doesn't manage to stop the old build, however, so you can end up with one slave running old builds for weeks at a time until the master is finally restarted.

It is impossible to Stop Build these old builds.

[dustin]

serendipitously, I saw a similar problem here today. Here's what I see in the web interface:

(view as text) Traceback (most recent call last): Failure: twisted.spread.pb.PBConnectionLost: [Failure instance: Traceback (failure with no frames): <class 'twisted.internet.error.ConnectionDone?'>: Connection was closed cleanly. ]

For unrelated reasons, our buildslave logs for this no longer exist, so that's all I've got.

[dustin]

We'll need more evidence to track this down, I think.

[dustin]

Hopefully, if this is a slave-side problem, my slave-side tests will tease it out.

[Dustin J. Mitchell]

Allow transfer steps to be interrupted

This also collects .finished and .interrupted into a parent class on the slave side. Fixes #825.

Changeset: c8d1ee63f6789d63a97ef39e62e7dd9d9a912562