Hey everyone! I'm *not* using PHP anywhere... My application is a TurboGears application, which is running it's on server and Cherokee is standing infront of it as a Reverse Proxy. For what it worth - it's a Python application.
Alvaro, I'll try to test r3488 and report here. Meanwhile I did this test: I reverted only handler_proxy.c to r3165, which seems to me like the first change since 0.99.20, and it didn't help. I checked the same TurboGears server - when connecting directly it all works, and through Cherokee 0.99.22 I get '502 Bad Gateway'. I noticed that Cherokee says "NOTICE: Taking source='localhost:8008' off-line" when I try to get the page, and after a while it takes the source 'back on-line'. It happens for all my sources (not just the one on port 8008), of course, on the same pages. Thank you. Yo'av 2009/8/13 LinuxInsight <[email protected]> > Yo'av Moshe wrote: > >> Hey, >> Anyone seen this behavior? Any ideas on what to check? >> As I said, I'm pretty sure something has changed with Cherokee between >> 0.99.20 to 0.99.21 that causes these "502 Bad Gateway" errors. >> >> My server is a production server and I'm right now it's not quite working >> as it should... :-p >> >> > Yo'av, don't know enough about your setup, sorry, joined list just today. > > But, from some extensive testing I did few days ago, I know one thing. > Using UNIX sockets for communication between cherokee and php-cgi farm is > very fragile. At some point in time, under high load, php-cgi's become > unresponsive and cherokee starts retunring 504 Gateway timeout. To make > matters worse, the problem won't remedy itself, full restart (cherokee + > php-cgi) is required to fix it. Now, I don't know the cause of the problem, > it might be even in kernel (some race condition in socket code under high > load), php-cgi mechanism or the way cherokee communicates with it... > > What I do know is that using TCP/IP sockets for the communication (e.g. > 127.0.0.1:9000, instead of /tmp/cherokee-php.socket) makes it much more > robust. And before anyone says that using TCP/IP is slower, I don't think > so, couldn't find any difference in speed, although I didn't measure it > extensively. It seems that localhost TCP/IP transport on Linux is heavily > optimized. But, my primary concern at this time is robustness. > > Anybody having gateway problems, try switching to TCP/IP communication, you > might be pleasantly surprised. > -- > http://www.linuxinsight.com/ > -- Yo'av Moshe
_______________________________________________ Cherokee mailing list [email protected] http://lists.octality.com/listinfo/cherokee
