I''ve got (so far) 5 different OS releases running as HVMs in near identical VMs, all created by virt-manager with a file-backed disk image, all x86_64, all running under x86_64 Fedora Core 6 Dom0 (Opteron system). The rhel5 VM crashes (goes into 100% cpu usage, and becomes non responsive) every few hours. I get the impression it might be related to disk I/O, since it has also crashed doing things like downloading the giant openoffice-core update rpm. All the other VMs have been up since they were installed, many happily doing big I/O like beagle-build-index, makewhatis, etc. Anyone know of a specific problem with rhel5 and qemu disks? Or any other problem that makes rhel5 hate running as a VM? The last time it crashed, I tried to do a shutdown from virt-manager and it wouldn''t go away. Now it is showing up as a Zombie VM (great, VMs can get zombified, just like processes :-). Is rebooting the Dom0 the only way to clear a zombie VM?
Tom Horsley wrote:> > The rhel5 VM crashes (goes into 100% cpu usage, and becomes non > responsive) every few hours. I get the impression it might be > related to disk I/O, since it has also crashed doing things like > downloading the giant openoffice-core update rpm. >https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=234160 Unfortunately RHEL-5 is having problems fully virtualized. I think it has something to do with memory pressure and mmap(), but I''m not 100% sure. I''ve found that if you increase the memory available to the virtual machine, it happens far less often. Chris Lalancette
On Tue, 24 Apr 2007 18:26:47 -0400 Chris Lalancette <clalance@redhat.com> wrote:> > The rhel5 VM crashes (goes into 100% cpu usage, and becomes non > > responsive) every few hours. I get the impression it might be > > related to disk I/O, since it has also crashed doing things like > > downloading the giant openoffice-core update rpm. > > https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=234160 > > Unfortunately RHEL-5 is having problems fully virtualized.Yea, I found the bugz right after I sent the mail. At least Intel wants it fixed, that will probably bump up the priority :-). I''ll try giving the VM more memory and see if it stays up longer.