Dnia sobota, 3 grudnia 2011 10:58:51 Roberto De Ioris pisze: > > Dnia sobota, 3 grudnia 2011 10:29:18 Åukasz Mierzwa pisze: > >> Dnia sobota, 3 grudnia 2011 08:22:10 Roberto De Ioris pisze: > >> > Sleeping helped me a bit :) > >> > > >> > I finally found the last leak in the master popping up during > >> > socket > >> > logging. My rant on socketpair/epoll was obviously wrong, they > >> > helped > >> > spotting the leak because their usage in uWSGI was wrong. Now the > >> > >> master > >> > >> > will make its health check at regular interval and not every time > >> > is > >> > >> is > >> > >> > waken up by a event (like remote logging). This should push back > >> > >> master > >> > >> > cpu usage to low-values even on event storming and maintain the > >> > memory > >> > consumption at a static value. The whole thing has been tested > >> > with > >> > >> ksm > >> > >> > too. > >> > > >> > Let me know. > >> > >> With uWSGI 1.0-dev-1745 if I have log-socket option enabled in vassal > >> or > >> fastrouter config, master cpu jumps from 0 to 100% on regular, small > >> intervals and I don't get any messages on the log-socket destination > >> other > >> than: "received message: too fast" > > > > Memory problem seems to be gone, after 400k requests per backend and 6 > > reloads > > per worker, memory usage does not grow any longer, master memory usage > > did not > > grow since start. I'll leave it running for whole weekend. > > This is without using log-socket option in workers, after it is fixed > > I'll redo > > the test. > > should be fixed now
Yes, it works now, thanks. I've upgraded to 1746 and restarted test, I'll report back later. Łukasz Mierzwa _______________________________________________ uWSGI mailing list [email protected] http://lists.unbit.it/cgi-bin/mailman/listinfo/uwsgi
