> 2010/9/20 Roberto De Ioris <[email protected]>: >> >> It still could be an overload problem, we have no way (for now) to know >> the time between >> nginx connect() and uwsgi accept(). I do not know if the TCP_INFO >> feature added to 0.9.7 could be useful. >> >> I will try to make a patch to let you test the status of connections in >> real time (remember that only TCP mode is supported) > > I'm looking at your patch "the master process will signal the overload > of the listen queue on linux tcp sockets": > > http://projects.unbit.it/hg/uwsgi/rev/b8fab39d08f9 > > Is this what I should apply? Also, I'm thinking it would be great to > start using syslog, so that various priorities can be used, and the > admin can sort and filter out important messages. Direct output is > not really usable. I'm a bit scared by C code, but I should give a > hand at this, I'm already familiar with the nginx syslog patch :-) > -- > Jean-Baptiste Quenot
Hi Jean, yesterday i submitted a patch to the nginx list to export the time a request starts to the upstream environments. It would be something like this: uwsgi_param UWSGI_TIME $unixtime; Then in uWSGI we can easily measure the time between requests and uWSGI accept() and (eventually) signal overload. If it will be accepted by the nginx team your problem should be more easy to understand. Regard the TCP_INFO patch i will release an updated 0.9.6.3 version that includes it. syslog support is on work, but without using the libc api as python generated exceptions will be not logged (they are printed to stderr). -- Roberto De Ioris http://unbit.it _______________________________________________ uWSGI mailing list [email protected] http://lists.unbit.it/cgi-bin/mailman/listinfo/uwsgi
