[patch] wrong count of pending builds

[silk <silk@…>]

Just updated to latest trunk. The pending builds counter seems nice, but the count is wrong.

Example: 15 pending builds ( Linux mono 2.2: 1 )

What may be wrong:

• I have this build configuration set up with "oldest revision" set up, to not build some few hundreds of old revisions
• I have a 600s time wait for stabilization.

I can try to find out more, if this is not enough info.

[wbell]

Interesting; I haven't tried it with not building old revisions. Right now, it's grabbing all the pending builds in the database and matching them up to the platforms of each config. Is it possible that you have old builds that are outside of your current build range (revision wise) that haven't been cleaned up? I know back when there were definitely some times where the master wouldn't delete builds that were outside of the revision range of a configuration.

In any case, more info would be useful, especially what the output of SELECT * FROM bitten_build WHERE state='P when the pending number disagrees with the platform number. A dump of the platform table would be helpful as well.

[silk <silk@…>]

Ok, I looked at the tables, my first shot was wrong.

The problem is, that I had deleted platform when it had pending builds.

I had 3 platforms for one configuration, but slave for one was down for some time, so pending build accumulated. Then I decided this platform is being dropped from support, so I deleted it. But previous pending builds remained.

Patch follows.

[silk <silk@…>]

delete pending builds, when deleting platform

[dfraser]

Looks good. Since bitten's serious about unit testing, it would be nice if there was a test case that checked that this happens too :-)

[silk <silk@…>]

True :)

Here it goes.

[silk <silk@…>]

with unittest

[dfraser]

Great, applied with some minor changes as r627