> On May 27, 2014, 4:59 p.m., Chiradeep Vittal wrote: > > I like the idea of taking action on OOM in the VR/systemvm. However there > > is a missing (very important) piece: the VR stays unconfigured. So, it > > should actually stop, not reboot in order for the management server to take > > notice and perform HA.
As far as I know Debian does not have a way to halt after a panic. I agree that a reboot of the SVM is not the best solution but it is still better then having the OOM killer randomly kill processes ending up partially functional. What's you though on this: have cloud-early-config detect a non-ACS initiated reboot, f.i. a panic, and stop the SVM? - Joris ----------------------------------------------------------- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/21773/#review43985 ----------------------------------------------------------- On May 26, 2014, 2:34 p.m., Joris van Lieshout wrote: > > ----------------------------------------------------------- > This is an automatically generated e-mail. To reply, visit: > https://reviews.apache.org/r/21773/ > ----------------------------------------------------------- > > (Updated May 26, 2014, 2:34 p.m.) > > > Review request for cloudstack, Chiradeep Vittal, daan Hoogland, edison su, > Hugo Trippaers, and Sheng Yang. > > > Repository: cloudstack-git > > > Description > ------- > > net.ipv4.ip_nonlocal_bind = 1 > For smooth transition on redundant routers in case of a failover > > kernel.panic = 10 > kernel.panic_on_oops = 1 > vm.panic_on_oom = 1 > This way, if the instance oops/ooms, it dies in a way that ACS sees it and > can act on it > > net.nf_conntrack_max = 1000000 > net.netfilter.nf_conntrack_max = 1000000 > also up the limit of nf_conntrack for more performance > > > Diffs > ----- > > systemvm/patches/debian/config/etc/sysctl.conf 586d5bd > > Diff: https://reviews.apache.org/r/21773/diff/ > > > Testing > ------- > > we are running these settings in our beta and prod env for months allready, > with good results. > > > Thanks, > > Joris van Lieshout > >
