[Buildbot-devel] stalling when downloading log files ?

[Brian Warner]

> For example, try getting
> http://build.fluendo.com:8081/fc2-core-full/builds/70/distcheck/0

I see that page stalling too. I'd point you at the twisted buildbot, because
it's at 0.6.6, but its host was upgraded to twisted-2.0.0 recently.

Do you see a big spike in CPU usage when somebody hits that page and their
download stalls? I'm wondering if there's some sort of O(n^2) problem biting
us.

The logfile delivery code was changed a lot in 0.6.5, making it a bit more
dependent upon Twisted's Producer/Consumer mechanisms. It's possible that
there's a bug in twisted-1.3 that got fixed in -2.0 .

My quick advice would be to upgrade to Twisted-2.0.1, since my personal
experience with it has been good. I'd like to understand the bug better and
find a workaround for -1.3, but that may not get you up and running the
fastest. (Unfortunately I'm heading out of town for a week, so I won't be
able to help out until I get back). To track this down, you'll need to put a
bunch of debug log.msg's in buildbot.status.html.ChunkConsumer.writeChunk and
buildbot.status.builder.LogFileProducer.resumeProducing . The idea is to find
out whether resumeProducing() is still being called.

cheers,
 -Brian