> On Fri, Aug 23, 2013 at 9:50 PM, yi huang <[email protected]> wrote: >> On Fri, Aug 23, 2013 at 9:43 PM, yi huang <[email protected]> >> wrote: >>> I've found my wsgi application always get harakiri-ed, I've >>> investigated a little by adding some logs to all the places that will >>> set harakiri, and i found this: >>> >>> Fri Aug 23 21:40:49 2013 - set harakiri 2 >>> Fri Aug 23 21:40:49 2013 - set harakiri 4 >>> Fri Aug 23 21:40:49 2013 - set harakiri 3 >>> Fri Aug 23 21:40:49 2013 - clear harakiri [4] >>> Fri Aug 23 21:41:00 2013 - *** HARAKIRI ON WORKER 2 (pid: 27010, try: >>> 1) *** >>> ... >>> Fri Aug 23 21:41:01 2013 - *** HARAKIRI ON WORKER 3 (pid: 27012, try: >>> 1) *** >>> ... >>> >> >> I should have been more clear, I only issue one http request, there >> are three set harakiri on three worker process, and only one is >> cleared, the other two process is recycled by harakiri. >> > > After further investigation into the source code, i'm pretty sure this > is a bug of gevent loop in multiple worker processes, I've issued a > bug report: https://github.com/unbit/uwsgi/issues/365 . And probably > come up with a patch soon. >
i have found the problem. Basically harakiri is not reset if the accept() returns EAGAIN. The problem is present in coroae too. I am about to push the fix. Thanks -- Roberto De Ioris http://unbit.it _______________________________________________ uWSGI mailing list [email protected] http://lists.unbit.it/cgi-bin/mailman/listinfo/uwsgi
