Migration issue with local iso storage

[udo]

Hi,
if I wan't to migrate an kvm-vm which booted from an live-cd (iso on local storage) I get the error message

migration aborted (duration 00:00:00): Failed to sync data - can't migrate local cdrom 'local:iso/dl-1.4.3-2011-09-29-k8.iso'
TASK ERROR: migration aborted

Of course, with an shared iso-storage should it possible, but we have some importand services running with live-cd (dhcp, dns, ftp, mail, ...) and in case of HA we need an iso-storage which is also absolutly fail-save!
And normaly this isn't so.
Or i need an extra 3-node-pve-cluster with drbd for this... much oversized! If I use extra nodes in the existing cluster, it's nessesary to have the right order for HA-recovery (first iso-nfs-server must run, then the others...).
local iso-storage (synced for all, or only importand isos?) fit much better - all running nodes have access to the local storage.

Udo

 

[dietmar]

local iso-storage (synced for all, or only importand isos?) fit much better - all running nodes have access to the local storage.

feel free to sync your iso images - or what exactly is the question?

 

[dietmar]

Oh, I can see the bug now ;-)

not sure how we avoid that?

 

[dietmar]

maybe you can setup a directory storage local on each node, but mark it as shared and sync with rsync?

 

[udo]

maybe you can setup a directory storage local on each node, but mark it as shared and sync with rsync?

Hi,
tested with nfs-share and rsynced local directory.

Live migration fails:

Feb 15 11:18:38 starting migration of VM 411 to node 'pve2-test2'
Feb 15 11:18:38 copying disk images
Feb 15 11:18:38 starting VM 411 on remote node 'pve2-test2'
Feb 15 11:18:39 starting migration tunnel
Feb 15 11:18:39 starting online/live migration on port 60000
Feb 15 11:18:41 ERROR: online migrate failure - VM 411 not running
Feb 15 11:18:42 ERROR: migration finished with problems (duration 00:00:04)
TASK ERROR: migration problems

The VM can started without trouble on both nodes.

Udo

 

[dietmar]

The VM can started without trouble on both nodes.

And you can online migrate that VM when the CD is on real shared storage (just wonder)?

 

[udo]

And you can online migrate that VM when the CD is on real shared storage (just wonder)?

Hi,
no - with the iso on a nfs-share (real shared storage) the migration fails with the same error too.

Udo

 

[dietmar]

Hi,
no - with the iso on a nfs-share (real shared storage) the migration fails with the same error too.

We will release a new kernel soon.

 

[udo]

We will release a new kernel soon.

Fine!

BTW. if I connect an cd-iso to an running linux-VM I can live migrate from node1 to node2. The way back the VM is off after migration (log say all migrated fine).
Without connected cdrom i can livemigrate between the two nodes without trouble (just do it seven times).

Perhaps this will also work better with the new kernel?!

Udo

 

[udo]

We will release a new kernel soon.

Hi,
with the new kernel (2.6.32-60) and 2.0rc1 the live migration of an live-distro (cd boot) work like a charm.

Very good job!!

Udo

 

[e100]

Hi,
with the new kernel (2.6.32-60) and 2.0rc1 the live migration of an live-distro (cd boot) work like a charm.

Very good job!!

Udo

Hi Udo,

Did you test with with only NFS?
I am curious to know if it also worked with a directory that is rsynced and marked as shared storage, if you tested that?

Still fails using local storage with iso copied to other node.

 

[udo]

Hi Udo,

Did you test with with only NFS?
I am curious to know if it also worked with a directory that is rsynced and marked as shared storage, if you tested that?

Still fails using local storage with iso copied to other node.

Hi e100,
no i've test with defined directory which contains on both nodes the same content. Work very good but since the last update the VM freeze after migration (and use 100% CPU). But i have changed the setup also (before the drbd-connection goes through ininiband, after the normal cluster-communication line was used (1GB Ethernet)).

But i think, that the drbd-changes are not related to the issue, because i build this weekend at home an tree-node cluster with the same issue (squeeze-VMs, but also with an 1GB-drbd-link on vmbr0).

Udo