On Fri, Dec 19, 2014 at 12:06 AM, Yann Ylavic <ylavic.dev@xxxxxxxxx> wrote: > If you can manage to recompile the APR library used by your Apache 2.4 > (you probably did that already since 2.4 does not seem to be the > version packaged with RHEL 6.4), you can use "./configure > --enable-posix-shm ..." to use another shared memory mechanism than > the default one (IPC SysV) which is limited to 32K segments (system > wide). > The number of segments would then be limited by the usual number of > file descriptors per process (ulimit -n). > > Unfortunately, there no equivalent to the Mutex directive for shared > memories, this can't be changed by httpd's configuration (as far as I > know). Oups, this leading part was not meant to be sent, and is rephrased anyway below according to message, more accurately. Please ignore it... > > On Thu, Dec 18, 2014 at 2:30 PM, Sylvain Goulmy <sygoulmy@xxxxxxxxx> wrote: >> >> I also noticed that a single proxy_balancer creates almost 350 shm on the >> system. >> >> Is it normal that a single proxy_balancer creates so many shm ? > > How many VirtualHost(s) do you use in your configuration? > Also, how do you declare the balancer(s), in each VirtualHost or in > the main section using BalancerInherit on? > >> >> Do i really have to increase the memory segment on my system in huge >> proportion to handle all my proxy_balancers ? > > The number of IPC SysV shared-memory segments is limited to 32768 on > linux (system wide), and you are already above with 100 * 350, so you > would need to use another SHM mechanism. > Unfortunately, there no equivalent to the Mutex directive for shared > memories, AFAIK this can't be changed by httpd's configuration. > So you would have to recompile the APR library used by your Apache 2.4 > (you probably did that already since RHEL 6.4 does not seem to package > 2.4 by default), and do for example "./configure --enable-posix-shm > ..." to use another shared-memory mechanism. > The number of segments would then be limited by the usual number of > file descriptors per process (ulimit -n). > > However 350 shm per balancer looks weird to me, your configuration may > do something not optimal... > > Regards, > Yann. --------------------------------------------------------------------- To unsubscribe, e-mail: users-unsubscribe@xxxxxxxxxxxxxxxx For additional commands, e-mail: users-help@xxxxxxxxxxxxxxxx