PJ You have increased "Max Open files" setting BUT i think you should increased "open files descriptor" limit at operating system level.
There is a linux kernel parameter "file-max" in /proc that does the trick. Try increasing to 10000000. Hope this will help Thanks! Anam ________________________________ From: P J <pauljfli...@gmail.com> To: users@httpd.apache.org Sent: Saturday, 5 May 2012, 0:20 Subject: Re: [users@httpd] Help troubleshooting performance issue, after "1000 total children" Apache no longer responds to HTTP requests. Not MaxClients issue? On Fri, May 4, 2012 at 8:46 AM, Tom Evans <tevans...@googlemail.com> wrote: On Fri, May 4, 2012 at 4:18 PM, P J <pauljfli...@gmail.com> wrote: >> On Fri, May 4, 2012 at 5:28 AM, Jim Jagielski <j...@apache.org> wrote: >>> >>> The closer MinSpareServers/MaxSpareServers are, the more time >>> Apache needs to spend killing and creating child processes >>> to ensure it stays within that narrow limit >>> On May 3, 2012, at 2:40 PM, P J wrote: >>> >> >> Thanks. Yes, I understand that. I set them both to that number to try and >> induce the issue we've been seeing. >> If I set: >> >> MinSpareServers 1024 >> MaxSpareServers 2048 >> >> 1000+ simultaneous connections is not uncommon this server. >> > >This is unlikely to have a major effect on your issue, as I believe >something strange is going on when you reach 1024 processes, but why >do you have MinSpareServers so high? > >If you start 1024 children initially, and then 900 of them are busy, >you are asking httpd to ensure there are at least 1924 children. It >seems massive overkill to ensure that there are so many idle >processes. > >If your high water mark is (say) 1536 clients, and you start 1024 >processes, then I would go with settings like this: > >MinSpareServers 128 >MaxSpareServers 1536 >StartServers 1024 >MaxClients 2048 >ServerLimit 2048 > >So, you start with 1024 processes, increasing up to a maximum of 2048 >processes. If there are ever more than 1536 processes idle, kill some >off. > >Cheers > >Tom > > > Thanks Tom, I agree. I don't normally have MinSpareServers at 1024, I was just trying to see if I can force it to throw the: "[debug] mpm_common.c(663): (70007)The timeout specified has expired: connect to listener on [::]:80 error". error. Which I can. The million $ questions is what is the magic 1024 limit I'm hitting. I'll hopefully try with a later version of Apache and let the list know if that resolves it. Thanks again. -- PJ
