> Looking for some ideas as to why my Django 1.4.2, nginx + uwsgi is slowing > down on some requests for the same exact page. > > I am hitting the page over and over with apache bench with concurrency 1. > What I am seeing is every couple of requests I see a spike in how long it > takes. I have enabled memory-profiler in uwsgi and I am greping it for the > time and how long it took: > > tail -f uwsgi.log | grep -oh "bytes in .* msecs " > > bytes in 500 msecs > bytes in 5542 msecs > bytes in 487 msecs > bytes in 416 msecs > bytes in 474 msecs > bytes in 368 msecs > bytes in 437 msecs > bytes in 416 msecs > bytes in 409 msecs > bytes in 445 msecs > bytes in 417 msecs > bytes in 465 msecs > bytes in 386 msecs > bytes in 460 msecs > bytes in 400 msecs > bytes in 390 msecs > bytes in 478 msecs > bytes in 437 msecs > bytes in 5489 msecs > bytes in 5608 msecs > bytes in 394 msecs > bytes in 5436 msecs > bytes in 458 msecs > bytes in 424 msecs > bytes in 368 msecs > bytes in 400 msecs > bytes in 5395 msecs > bytes in 397 msecs > bytes in 5 msecs > bytes in 383 msecs > bytes in 69 msecs > bytes in 397 msecs > bytes in 338 msecs > bytes in 372 msecs > bytes in 368 msecs > bytes in 379 msecs > bytes in 368 msecs > bytes in 443 msecs > bytes in 388 msecs > bytes in 435 msecs > bytes in 5449 msecs > bytes in 500 msecs > bytes in 383 msecs > bytes in 410 msecs > bytes in 397 msecs > bytes in 408 msecs > bytes in 425 msecs > bytes in 407 msecs > bytes in 10762 msecs > > > Every few requests I get a spike of 5-10s which I find really odd. Django > profiler is indicating that the time to process the request is about 2 > secs > and when I run it on slow requests it doesn't show anything out of the > ordinary. It seems like uwsgi is actually taking the time and not the wsgi > application. > > Running against another page shows consistent times. > > > Setup: > > app server on linode 2GB RAM > uwsgi --version: 1.9.21.1 > _______________________________________________ > uWSGI mailing list > [email protected] > http://lists.unbit.it/cgi-bin/mailman/listinfo/uwsgi >
This is hardly a uWSGI/nginx problem (unless some info is missing). Start your app with a single worker and attach a strace to it (strace -p <pid>) as there is a big gap in response times you should clearly see where the slowdown is -- Roberto De Ioris http://unbit.it _______________________________________________ uWSGI mailing list [email protected] http://lists.unbit.it/cgi-bin/mailman/listinfo/uwsgi
