Slave connection timeout incorrectly measured

[dabrahams]

The description of the slave connection timeout is "The timeout in seconds after which a build started by a slave is considered aborted, in case there has been no activity from that slave in that time," but in fact Bitten measures the time from the beginning of the build. If you have enough short steps, it's easy for a slave to exceed any given timeout without actually being unresponsive.

[hodgestar]

Patch from Walter being added shortly. Walter's notes:

Add new last_activity field to build. I considered reusing stopped, but this seemed cleaner and more obvious, which seems like the right way to go.

On the completion of every step, last activity is updated to the current time. All orphaning/invalidation based on slave_timeout happens based on that time, not started. Previously, builds were orphaned if it had been more than slave_timeout (config setting) seconds since they had started, now it's since the last interaction (which is actually what the documentation on the field already says.)

This requires a database upgrade.

[hodgestar]

Walter's patch.

[hodgestar]

See also #411.

[wbell]

One more try.

[hodgestar]

Fixed up database upgrade script, few other minor tweaks.

[wbell]

Fixed as of [840]