PVE 1.4 beta2: migration KVM on storages crashed

[valshare]

Hi,

after migrate a KVM from Master to cluster node, the KVM server not reachable anymore over network and crashed.

The KVM is stored an a iscsi shared storage.

/usr/sbin/qmigrate --online 192.168.1.3 112
Oct 07 11:50:19 starting migration of VM 112 to host '192.168.1.3'
Oct 07 11:50:20 copying disk images
Oct 07 11:50:20 starting VM on remote host '192.168.1.3'
Oct 07 11:50:21 starting migration tunnel
Oct 07 11:50:21 starting online/live migration
Oct 07 11:50:23 migration status: active (transferred 30108KB, remaining 1037220KB), total 1065216KB)
Oct 07 11:50:25 migration status: active (transferred 57444KB, remaining 1011088KB), total 1065216KB)
Oct 07 11:50:27 migration status: active (transferred 95456KB, remaining 974404KB), total 1065216KB)
Oct 07 11:50:29 migration status: active (transferred 154216KB, remaining 919540KB), total 1065216KB)
Oct 07 11:50:31 migration status: active (transferred 194508KB, remaining 881788KB), total 1065216KB)
Oct 07 11:50:33 migration status: active (transferred 233164KB, remaining 844932KB), total 1065216KB)
Oct 07 11:50:35 migration status: active (transferred 294212KB, remaining 786820KB), total 1065216KB)
Oct 07 11:50:40 migration status: completed
Oct 07 11:50:40 migration speed: 53.89 MB/s
Oct 07 11:50:41 migration finished successfuly (duration 00:00:22)
VM 112 migration done

/etc/qemu-server/112.conf
name: Management
ide2: none,media=cdrom
sockets: 1
bootdisk: ide0
ostype: wxp
memory: 1024
onboot: 0
vlan0: e1000=2A:80:A0:A0:8B:45
boot: cad
freeze: 0
cpuunits: 5000
acpi: 1
kvm: 1
description:
cores: 2
ide0: KVM-DATA-01:vm-112-disk-1

 

[dietmar]

Are there different CPUs on the nodes (AMD vs. INTEL)?

 

[valshare]

Hi,

no both are Dell PowerEdge R805 with AMD Processor.

 

[dietmar]

And the error is reproducable? What runs inside the VM?

 

[valshare]

i have tested it with another kvm, that crashed too. In the machine runs windows xp. The error is reproducable.

 

[dietmar]

And what kind of iSCSI Storage do you have (Vendor/Model)?

 

[valshare]

i have a Thecus I4500r.

http://de.thecus.com/products_over.php?cid=53&pid=168&set_language=german

 

[valshare]

no idea? what i can see, is that windows crashes and make a reboot.

 

[dietmar]

So far i am unable to reproduce the error.

 

[valshare]

anything i can help?! do you mean it is the storage?
i can try it over nfs ...

 

[dietmar]

Just try different configurations - for example use NFS as storage, use other OS, use only one core, ...

 

[valshare]

ok, will post the results tomorrow.

 

[valshare]

ok, what i can say.

Installed a fresh Windows XP SP3 guest on iscsi node and lvm shared storage.

Try with 2048 MB RAM / 1 CPU / 1 CORE = failed
Try with 2048 MB RAM / 1 CPU / 4 CORE = failed

all tests makes WITH and WITHOUT network and cdrom interface. All tests with Windows XP crashed.

Tryed with Linux distribution with 2.6 Kernel

512 MB Ram / 1 CPU / 1 Core in as KVM. No problem...same NODE and same LVM. Mirgated 20 times ... no crashes ...

Haven´t tested with nfs, think the iscsi works also.

 

[stiaga]

Hi,
I have the same problem.

My configuration is: 2 node 1.4 beta2 with primary/primary drbd.

I migrate the WinXP machine and it crashes (blue screen).
The vm reboots but the S.O. vm crashes again. I need to stop the vm and then restart it.

Why?

 

[valshare]

Hi,
I have the same problem.

My configuration is: 2 node 1.4 beta2 with primary/primary drbd.

I migrate the WinXP machine and it crashes (blue screen).
The vm reboots but the S.O. vm crashes again. I need to stop the vm and then restart it.

Why?

same here. I think it is from beginning from version 1.4b2

 

[dietmar]

It works perfectly here - what XP version do you use (SP3)?

 

[stiaga]

I don't know if it is a problem of the 1.4 b2 version but the I believe that the problem is the lvm locking system.
It seems that in the destination host the logical volume associated with the virtual machine remains in read-only.

This theory can be confirmed from the fact that if I stop the vm and start it in the destination node where I have migrated the vm it works well.

Anyone can explain me how lvm locking works?

 

[stiaga]

I use WinXP SP2

 

[valshare]

here we use SP3

 

[dietmar]

I don't know if it is a problem of the 1.4 b2 version but the I believe that the problem is the lvm locking system.
It seems that in the destination host the logical volume associated with the virtual machine remains in read-only.

We do not use lvm locking, and we do not activate volumes read only (we have a simple lock manager on the master - not lvm specific).