We have discovered a bug in our use of the Python engine. It has to do with a debug feature that we turned on in ALPHA to hunt reference leaks. Basically we added
which adds the sys.getobjects function. There was an unfortunate side effect that the counter used was only 32 bit. Which turned out to be a problem when ~6000 users had been playing for some time. We never understood why we had node deaths, after the peak hour. This was the reason, a build up in the alloc counters. A patch will be issued tomorrow, so expect a lot less node off-lining as of tomorrows downtime.

Comments are closed.