Hi, On 2010-08-03 at 13:10:10 [+0200], Matt Madia <mattmadia@xxxxxxxxx> wrote: > On Tue, Aug 3, 2010 at 10:46, <support@xxxxxxxxxx> wrote: > > Notification for 188.40.89.174: > > > > ==> HTTP <== > > State is now: ERROR (12:46h 2010.08.03) > > State before: OK (was about 1 month) > > Service params: port=80 path=/ hostname=dev.haiku-os.org > > Service uptime: 99.91% > > Check output: > > > timeout after 25 seconds - execution expired > > ssh was non-responsive. > according to top on baron, vmdev's cpu & ram usage were pegged. > 7184 root 20 0 1683m 1.5g 772 S 200 18.8 4281:30 qemu-kvm > > here's the server stats: > http://server-stats.haiku-os.org/stats?host=vmdev.haiku-os.org&plugin=apache&plugin=cpu&plugin=df&plugin=disk&plugin=interface&plugin=load&plugin=memory&plugin=postgresql&plugin=processes&plugin=vmem×pan=hour&button=Ok > > > ...Not sure what caused the memory & processes to spike. > baron:/etc/init.d # service VM-dev restart > > > After restarting ( baron:/etc/init.d # service VM-dev restart ), it's > memory usage went past 1536, causing it to be non-responsive. > I increased the ram to 2048 in /etc/init.d/VM-dev and after another > restart, it seems to be happy. > 1390 root 20 0 2190m 405m 1380 S 1 5.1 0:48.23 qemu-kvm Thanks for the info. I have checked the logs, but did not find anything that would explain the increased memory use, especially not directly after boot. :-\ Just now, I have reconfigured (and restarted) all VMs to let their I/O bypass the hosts cache, which should considerably lower the memory pressure on baron. Before, the image-file as well as the VMs disk blocks had to travel through the host cache. Now it should be only the image-file. Additionally, I noticed that the NTP configuration got hosed during one of the recent package upgrades, so I invested some time to get a proper NTP setup on baron and on all VMs, which hopefully shouldn't get busted again (for the curious: we are now using the "locale" ntp servers provided by Hetzner). cheers, Oliver