Il giorno 10/set/2012, alle ore 21:08, Łukasz Mierzwa <[email protected]> ha scritto:
> Hi, > > some time ago I've noticed that my carbon stats have random holes, > once in a while some random node has one missing cycle. > I've looked at logs and I see no errors, it looks like some cycles are > skipped. > It's not nptd adjusting clock (my first guess), so maybe master was > busy with something and did not executed carbon cycle? I have a lot of > crons and I also see that cheaper spawned new worker at the time when > I would expect missing carbon cycle to be executed. > Does anyone else have such issue? > > carbon update log: > > 10/09/2012 20:00:21 :: wrote 1 datapoints for > uwsgi.a178.myapp.requests in 0.00048 seconds > 10/09/2012 20:01:22 :: wrote 1 datapoints for > uwsgi.a178.myapp.requests in 0.00023 seconds > 10/09/2012 20:02:22 :: wrote 1 datapoints for > uwsgi.a178.myapp.requests in 0.00024 seconds > 10/09/2012 20:03:22 :: wrote 1 datapoints for > uwsgi.a178.myapp.requests in 0.00019 seconds > 10/09/2012 20:04:21 :: wrote 1 datapoints for > uwsgi.a178.myapp.requests in 0.00091 seconds > 10/09/2012 20:06:24 :: wrote 1 datapoints for > uwsgi.a178.myapp.requests in 0.00036 seconds > 10/09/2012 20:07:25 :: wrote 1 datapoints for > uwsgi.a178.myapp.requests in 0.00022 seconds > 10/09/2012 20:08:25 :: wrote 1 datapoints for > uwsgi.a178.myapp.requests in 0.00025 seconds > 10/09/2012 20:09:24 :: wrote 1 datapoints for > uwsgi.a178.myapp.requests in 0.00022 seconds > > No update at 20:05:2x. > I'm not sure if crons and cheaper handling could cause carbon cycle to > be skipped (I would expect it to be delayed at worst), so I'm not sure > what is happening and if this is uWSGI, carbon/graphite or my nodes > issue. > It's not a big issue but it's annoying. > Any tips? > > Your "update resolution", is too near the "master frequency". As soon as the master slowdown a bit (like when calling fork()), the master clock will lose of a bunch of milliseconds (or even a whole second on a loaded system). Probably setting the carbon update frequency to 2 seconds will fix the problem, but a better solution would be allowing milliseconds cycles in the master. Something to look at for 1.4 (expecially now that we have high-resolution/highly-reliable clock plugins) -- Roberto De Ioris http://unbit.it JID: [email protected] _______________________________________________ uWSGI mailing list [email protected] http://lists.unbit.it/cgi-bin/mailman/listinfo/uwsgi
