On Thu, Mar 10, 2011 at 03:13:52PM +0000, Joe Gooch wrote: > I have several servers that have run Debian Lenny w/ kernel 2.6.22.19 > for quite some time without any issues. I've been doing rolling > upgrades to Squeeze... and I've encountered some strange limiting > behavior. > Some of my vservers have limits on resource usage (i.e. rlimits/as, > rlimits/nproc, rlimits/rss), and have been experiencing out of memory > issues to the extent that I can't even vserver exec processes into the > namespace anymore. > Looking at all my /proc/virtual/xxx/limit I see the following: > VM: 73456 0/ -1 -1/ -1 0 > VM: 7264 0/ -1 -1/ -1 0 > *VM: -27253 0/ -1 -1/ -1 0 > VM: 6751 0/ -1 -1/ -1 0 > VM: 183464 0/ -1 -1/ -1 0 > VM: 1261030 0/ 2224643 -1/ -1 0 > VM: 41791 0/ 199142 -1/ -1 0 > VM: 191196 0/ 258038 -1/ -1 0 > VM: 199354 0/ 1603723 -1/ -1 0 > VM: 670480 0/ -1 -1/ -1 0 > -VM: 45315 0/ 1200000 1200000/ 1200000 0 > VM: 31508 0/ 300000 300000/ 300000 0 > *VM: -88860 0/ -1 -1/ -1 0 > VM: 431386 0/ 502421 -1/ -1 0 > *VM: -14804 0/ -1 -1/ -1 0 > VM: 25737 0/ -1 -1/ -1 0 > Notice have 3 vservers on different hosts that have negative "current" > values for virtual memory. (at the moment) The ones that have no > hard/soft limit are ok... but if I have a hard limit configured, I > can't spawn any more processes when this happens. I have a vserver > where this happens at least once a day. (the dashed one above) > I would assume that a negative current value is a bug... definitely a bug, you might check with the mainline 2.6.32.28 kernel and appropriate patch, if you can recreate the issue there, let me know and we'll investigate ... > Is there a delta to fix this? no idea, the VM/RSS accounting and limits have been dropped in favor to the mainline cgroup accounting which is used in all newer kernels > Kernel is 2.6.32-5-vserver from squeeze. (mostly 686 kernels)... > In looking at their deb it looks like they're using vs2.3.0.36.29.6. best, Herbert > Thanks. > Joe >