All,
I am currently running a xen-2.0-testing snapshot from April 20. I''m
having sporadic problems with migration.
I have two xen machines, 10.130.2.35 and 10.130.2.36, booting from a
read-only, iso image loopback iscsi target from a third machine. I''m
using the Cisco iscsi-initiator and iscsi-init module for the boot. The
iscsi has been solid so far.
The scsi target ends up mounted to /dev/sda in Dom 0 on both the
machines. I then use that same read-only mount and, as the following
xenU config file shows, gets exported up to /dev/hda when a xenU gets
created:
kernel = "/boot/kernel-2.6.11-xen-2.0.5-domU"
ramdisk = "/boot/initrd"
memory = 64
name = "test"
vif = [ ''mac=00:55:4F:44:00:01'' ]
disk = [ ''phy:sda,hda,r'' ]
dhcp="dhcp"
root = "/dev/ram0 ro init=/linuxrc cdroot"
Everything boots just fine. The "test" xenU runs flawlessly; I can
ssh
into it, run whatever. No problems there. And it''s surprisingly fast
over iscsi, even though I''ve only got 100 Mbit Ethernet adapters.
BUT...
I''ve been migrating between the machines, both live and non-live, with
mixed success. Sometimes, I''d say every 1 in 10 migrations, I get the
errors posted in the attached xfrd.log files. The .1 file is the source
of the migration and the .2 is the destination. The other 9 of 10
times, it migrates just fine.
I don''t seem to get these problems when I do not export /dev/sda to a
domU. For example, if I use just a simple domU (using the same kernel)
with no mounts and an initrd file system, I don''t have these problems.
I saw mailing list messages a while back dealing with migration and the
possibility of a crash under heavy network load. Further, I saw a patch
that had been applied:
<QUOTE>
[PATCH] stream fixes for migration
I''ve attached a patch for libxutil/libxc. This fixes one of the hangs
I''ve seen during migrations. It applies against 2.0 and 2.0-testing.
Changes:
* Encountering EOF or error when xfrd reads from stream could cause an infinite
loop.
* Cleaned up the closing of streams.
* Fixed several memory leaks.
Signed-off-by: Charles Coffing <ccoffing@novell.com>
</QUOTE>
The version of 2.0-testing I''m using has this patch applied. But the
comments in this patch imply that there are still more "hangs" during
migration. Have a stumbled on another one of these?
I believe this patch fixed a previous problem, I would get a looping
hang under 2.0.5 stable; I haven''t seen that after going to
2.0-testing.
Am I making incorrect assumptions that I can read-only mount an iscsi
target twice?
Or could hardware be a factor? For testing, I''m just running cheap-o
VIA
Rhine 100-TX controllers. I thought I would post this before shelling
out for some Intel gig nics and gig switches though.
Thank you very much for your help.
-James Henderson
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel