Comment 29 for bug 118625

Hmm, okay, so mwhudson reports that there are still hangs, though they may be less frequent. From the log that I was sent, it looks like things are generally performing better (I see at least one thread that processed its max of 100 requests and then ended, so that's good) but there's still significant slowdowns before the server was restarted.

Here's what I see:

The earliest slowdown is around 21:53, like this:

* Somebody simultaneously requests several revisions of http://bazaar.launchpad.net/%7Egary/launchpad/fix-db-devel-sourcecode/
* It takes 11 seconds to build the revision graph, but thanks to the patch only one of the threads builds it.
* All told, each thread thinks takes a total of about 15 seconds to get the information for the RevisionUI, which means that each thread spent 4 seconds processing the data after the cache was generated.

Overall, that's slow, but it's not hanging slow, and after that brief slowdown things are speedy again for a while.

In fact, the original problem that I saw here is indeed fixed, and there is now some other problem. What's really weird with this log is there appears to be a real hang, not just a slowdown. Loggerhead stops logging for eight minutes before it's restarted, so the log is basically little help at all, other than to see what it was doing before everything was flushed. Further investigation underway.