Comment 15 for bug 717345

I can now reproduce this locally, without needing tons of free memory, etc, thanks to the glories of 'ulimit'.

ulimit -n 100
make run_codehosting

hammer_ssh.py
python hammer_ssh.py --load=20 --request-count=200 --server bazaar.launchpad.dev --run-time=20

Starts crashing pretty quickly. I'll attach the lsof output. But basically, it still has a bunch of the handles open. Not only that, but my machine still has a bunch of 'ssh' processes that are still connected, and a fair number of launchpad-forking-service children.

So looking pretty good from the point of being able to reproduce it.