OpenVZ live-migration

[user100]

I played with the live-migration of OpenVZ that should work now. But I´m not sure if live-migration is really useable now. If I´m just logged in via ssh (Putty) without doing anything there and do a live-migration it´s working fine. Sould mean: After migration I can still work in my Putty-Window without a new logon.

But if I let following run (again in Putty-window) in the bash it often hangs and does not respond when I do a live-migration:

while :;do date; done

I tried a while loop with date just to quick find out how long there is a noticeable interrupt while migrating. Sometimes it worked. Often it does not worked and it seems like the Putty-session freezed (should mean, CTRL-C does not work, Enter did not make a new empty line, date-lines stop). And I ask me, when this happens to this simple loop in the bash what happens with database-write, incoming mails (with or without SSL) or other important stuff?

I have not tried this in KVM yet - but currently I don´t really trust the live-migration with OpenVZ-VMs. Or have I missed something to think about here?


Greetings,
user100

 

[dietmar]

Maybe you can report that bug directly to the openvz people. If more people report bugs it is more likely they will fix them:

http://bugzilla.openvz.org/

- Dietmar

 

[user100]

Can somebody tell me where to find the current source of the kernel that is used in Proxmox 1.0? The date of the files that I found myself at ftp://pve.proxmox.com/sources (2008-10-29) differ from the date of the kernel that came with Proxmox 1.0 (2008-10-24), so I don´t really know...

...or maybe better - somebody of the Proxmox-maintainers get in contact with the OpenVZ guys directly?


Greetings,
user100

 

[dietmar]

the source is available at: ftp://pve.proxmox.com/sources

 

[user100]

It seems the openvz-guys don´t like the source from there (and don´t trust the "proxmox-kernel" so far). - It´s the same with proxmox 1.1. I´m in doubt that the bug is on the proxmox side. However I tried to test with another "clean" kernel on the proxmox-machines in the meanwhile (2.6.27 vanilla + OpenVZ-Patch + OpenVZ-Config) to clear this question. But with that kernel I don´t get so far. It´s booting well and openvz-machines are starting but live migration stops (verbosed output) at:

Can not dump container: Invalid argument
Error: iptables-save exited with 255
Checkpointing failed
Error:  Failed to dump container
Resuming...

I searched and found a possible workaround for that error (linking /sbin/iptables-save and /sbin/iptables-restore to /bin/true in the VE) - but it don´t really work at my test-system. Any ideas?


Greetings,
user100

 

[dietmar]

It seems the openvz-guys don´t like the source from there (and don´t trust the "proxmox-kernel" so far). - It´s the same with proxmox 1.1. I´m in doubt that the bug is on the proxmox side.

The problem is that we cant use stable 2.6.18, because that does not work well with kvm. Instead we use the ubuntu 2.6.24 kernel which is 'officially' supported by openvz.

Do you have a better idea?

 

[user100]

Do you have a better idea?

Not so far. Don´t understand me wrong - I think 2.6.24 is ok and it´s not the fault of proxmox or a to new kernel (btw. today I would prefer lenny than etch anyway ). When 2.6.27 would make the same failure than it´s clear some OpenVZ bug. But I can´t tell them it´s the same because it´s stopping migration before I can test. I´m new with OpenVZ. So my hope was somebody knows what that IP-tables can else mean at migration. The workaround with the link does not really help here.

Do you think 2.6.18 should work with live-migration and OpenVZ? I thought there was some troubles regarding live-migration in the past?


Greetings,
user100

 

[dietmar]

When 2.6.27 would make the same failure than it´s clear some OpenVZ bug.

AFAIK openvz 2.6.27 is very unstable.

Do you think 2.6.18 should work with live-migration and OpenVZ?

yes

 

[user100]

Ok... I would try with some other (vanilla+openvz-patch) kernel too.

 

[user100]

Following results:

2.6.18 (vanilla+openvz-patch/config) - No hang at live-migration but 2.6.18 is old and live-migration does not work with all templates

2.6.24 (vanilla+openvz-patch/config) - Hangs at live-migration.

2.6.26 (vanilla+openvz-patch/config) - No hang at live-migration but does not work with all templates (seems the same CentOS templates as in 2.6.18 make troubles at live-migration).

2.6.27 (vanilla+openvz-patch/config) - Live-migration does not work with my templates. Found some workaround regarding the ip-tables failure (link to /bin/true) but did not help here. So I gave up with that version.

Failure with the 2.6.18 and 2.6.26 version and CentOS templates (4 or 5):

Can not dump container: Invalid argument
Error: d_path is invisible /dev/ptyp0
Checkpointing failed
Error:  Failed to dump container
Resuming...

Greetings,
user100