Test A Bug Theory For Me?
- From:
- Zed A. Shaw
- Date:
- 2010-08-24 @ 02:05
Hi folks, I'm trying to track down the 100% CPU problem that comes from
0MQ handlers getting double messages. I've so far tracked it down to
something related to the IO threads setting, so I'm suspecting that
there's a race condition in 0MQ that's making it send two messages, but
then barf when it gets the double responses.
If you could test setting zeromq.threads to 1 and then thrash your 0MQ
handlers to see if you still get 100% CPU that'd help me test that.
What you should see when you produce the bug is that the handler dies,
and mongrel2 reports 100% CPU. Your handler should also get the
messages once, even though mongrel2 sent it once. If you then restart
the handler, mongrel2 goes back to normal and everything is good.
With zeromq.threads=1 the theory is we shouldn't see the double messages
and the 100% CPU should not happen.
Thanks!
--
Zed A. Shaw
http://zedshaw.com/