On Mon, Sep 22, 2003 at 03:13:15PM -0400, Jesse Guardiani wrote: > Again, this all started when I upgraded my Apache version from 1.3.27 to > 1.3.28 and upgraded the supporting modules and such.
Mine is: Server: Apache/1.3.28 (Unix) mod_fastcgi/2.4.0 mod_throttle/3.1.2 mod_ssl/2.8.15 OpenSSL/0.9.6g Ah, I've just checked... one server had ~160 zombie processes, and the other had ~210. Not that it affects the server's performance... but I suppose eventually it could run out of process slots. However I find that sending a 'HUP' to Apache is sufficient to clear them, i.e. kill -1 `cat /var/run/httpd.pid` > sqwebmail doesn't link to anything that would be effected by apache code, > does it? (I don't think it does) A CGI is really a standalone application. It gets some information from environment variables, and communicates on stdin/stdout. It doesn't link to the webserver code at all. Some changes were made in 1.3.28 to the way Apache handles CGIs, to fix a race condition (what was happening, particularly under multi-CPU Solaris boxes, was that Apache would notice that stdout had been closed while the process was still finishing cleaning up after itself. However Apache decided that the process was stuck, so it then sent it a 'kill -TERM', waited 3 seconds, and sent it a 'kill -KILL'. The overall effect was that each click on Sqwebmail seemed to take 3 seconds longer than it should.) It's not impossible that those changes introduced the problem. I think it's something you should bring up on an Apache mailing list though, especially if reverting to 1.3.27 makes the problem go away. Cheers, Brian.
