On 7 Oct 2009, at 17:25, Philip Guenther wrote:
On Mon, Oct 5, 2009 at 10:16 AM, Chris <cjd...@brokensolstice.com>
wrote:
I just reinstalled 4.5. I touched nothing on the system. I
installed
mutt through pkg_add, then created a 1M empty file from /dev/null.
I sent this email to myself thusly: mutt -a 1megfile m...@myaddress.com
</dev/null
Same exact behavior. It spools, then I have about .5 - 1 second
before then entire system locks up.
Random guess: sendmail is one of the few standard programs that makes
heavy use of fsync(). Perhaps there's something on the "deal with
stuff being forced to 'disk'" paths of the xen emulation that confuses
the kernel. If so, the turning off sendmail's SuperSafe option in the
sendmail.cf should make things more stable. That would be *JUST* a
debugging measure, as doing that on a production server, particularly
one which is unstable, is a recipe for losing email.
OpenBSD doesn't have PV-on-HVM drivers yet (this is the hybrid mode
which lets guests run in HVM mode but use high-performance I/O). This
means that you would going via the much slower qemu-dm emulated
devices. What backing store are you using for the disks? Can you dd
a large file from /dev/random into a file and not have it lockup? What
version of Xen is it?
Overall, I wouldn't really recommend running a mail server under a
pure emulated virtualised environment (Xen, KVM, VMWare, they all are
at the moment for OpenBSD) since the emulation path is usually slow
but also mainly only used to boot the guest before switching to high-
performance PV drivers.
The other question is whether you've tried breaking into ddb from the
console when this happens. Make sure you have ddb.console=1 in your
/etc/sysctl.conf and read the ddb(4) manpage. (If xen doesn't offer
console access then throw it out.)
Of course you can access the guest console.
-anil
