Обсуждение: OOM Killer / PG9 / RHEL 6.1
Howdy, just a quick check, is vm.overcommit_memory = 2 vm.swappiness = 0 Still the way to go with PG9.0 / RHEL 6.1 (64bit) ? I know we gained some control over the OOM Killer in newer kernels and remember reading that maybe postgres could handle it in a different way now. Thanks Dave
On Wed, Nov 2, 2011 at 10:42, David Kerr <dmk@mr-paradox.net> wrote: > Howdy, > > just a quick check, is > vm.overcommit_memory = 2 > vm.swappiness = 0 > > Still the way to go with PG9.0 / RHEL 6.1 (64bit) ? IMHO yes (although I never touch swappiness...) > I know we gained some control over the OOM Killer in newer kernels > and remember reading that maybe postgres could handle it in a different way now. If you compile with -DOOM_ADJ (which I think the red hat rpms do?) yes. As usual the docs are fairly good: http://www.postgresql.org/docs/9.1/interactive/kernel-resources.html#LINUX-MEMORY-OVERCOMMIT
On Wed, Nov 2, 2011 at 10:42 AM, David Kerr <dmk@mr-paradox.net> wrote: > Howdy, > > just a quick check, is > vm.overcommit_memory = 2 > vm.swappiness = 0 > > Still the way to go with PG9.0 / RHEL 6.1 (64bit) ? If you've got lots of ram, it's better off to throw a "swapoff -a" at the end of rc.local, as I found that on large mem machines with late model kernels, setting swappiness to 0 I still had the occasionaly kswapd storm being my production servers to their knees.