OpenVZ -...

Portal Home > Knowledgebase > Industry Announcements > Web Hosting Main Forums > OpenVZ -...

Posted by PCS-Chris, 12-02-2013, 05:01 PM
Yesterday CentOS 6.5 was released and if you've not already seen this, you will likely see floods of tickets come in because when you upgrade an OpenVZ container to CentOS 6.5 and reboot, it will lose all network connectivity.

No active network interfaces:
Code:
[root@test ~]# ifconfig
[root@test ~]#
We cant bring interfaces up either:
Code:
[root@test log]# service network restart
Shutting down loopback interface:                          [  OK  ]
FATAL: Module ipv6 not found.
Bringing up loopback interface:  RTNETLINK answers: Operation not supported
Failed to bring up lo.
                                                           [FAILED]
Bringing up interface venet0:  RTNETLINK answers: Operation not supported
Failed to bring up venet0.
                                                           [FAILED]
FATAL: Module ipv6 not found.
How to solve (CentOS 6 64Bit):

Until the OpenVZ developers create an official fix, you can downgrade the iproute package as follows:

1. From your OpenVZ Hostmachine, download the old RPM:

# wget http://repo.smartservermanagement.co...el6.x86_64.rpm

2. Copy this inside the container:
# cp /root/iproute-2.6.32-23.el6.x86_64.rpm /vz/private/<CTID>/root/

3. Console into the VPS, forcefully remove the current iproute package, and replace with old one:

# vzctl enter <CTID>
# rpm -e iproute-2.6.32-31.el6.x86_64 --nodeps
# rpm -ivh /root/iproute-2.6.32-23.el6.x86_64.rpm

4. Restart the network service

# service network restart

..and voila!

Don't forget if you have automated yum updates configured (e.g. cPanel does this), add iproute* to the exclude= line in /etc/yum.conf for the time being.

If this fix doesn't work for you please let me know as we would be interested to hear about it!

Cheers
Posted by Jon-RackSRV, 12-02-2013, 05:04 PM
Great work Chris, thanks for sharing!
Posted by PCS-Chris, 12-02-2013, 05:20 PM
No worries, same problem and fix applies to RHEL 6.5 32bit variants but you will want this RPM:

http://repo.smartservermanagement.co...el6.x86_64.rpm

Note that I have made no changes to this RPM whatsoever, it's actually from Scientific Linux 6.4 as it was the easiest source to find.
Posted by Tom,, 12-02-2013, 05:40 PM
Thanks guys - really appreciate your efforts. Was a fix for my problem - http://www.webhostingtalk.com/showthread.php?p=8934604
Posted by ZonedHost, 12-02-2013, 05:53 PM
Great work! Intresting that a lot of newer Linux releases are having really big issues with networking on OpenVZ Kernels. This has been going on for months (Is a problem we have experienced) and still there is no public/kernel fix for the issues;

bugzilla.openvz.org/show_bug.cgi?id=2712
Posted by PCS-Chris, 12-02-2013, 06:03 PM

Quote:

Originally Posted by ZonedHost

bugzilla.openvz.org/show_bug.cgi?id=2712

Is that your bugreport? Those issues look specific to OVH and their configuration on that switch you are plugged into. Quick and dirty fix for this is to bring up all of your IP's on eth0:X on your server which will update the arp table. Then you can assign to containers without error. I've had to do that in the past after migrating VPS with certain server providers. Certainly saved some headaches!
Posted by ZonedHost, 12-02-2013, 06:26 PM

Quote:

Originally Posted by PCS-Chris

Is that your bugreport? Those issues look specific to OVH and their configuration on that switch you are plugged into. Quick and dirty fix for this is to bring up all of your IP's on eth0:X on your server which will update the arp table. Then you can assign to containers without error. I've had to do that in the past after migrating VPS with certain server providers. Certainly saved some headaches!

Cheers for the reply. Nope its not my bugreport (We are not at OVH), but I and a few others appear to have the same issue. Ill give your suggestion a whirl, but its the same issue on two seperate nodes running 32Bit Kernels that don't seem to be compatible with newer Ubuntu/Debian OS's IP configs (Even if you trick the Kernel to get it to boot!)
Posted by RRWH, 12-02-2013, 09:58 PM
Interestingly, we just jumped onto one of our DNS-only installs and a yum list iproute shows that both 2.6.32-23.el6 AND 2.6.32-31.el6 Are installed!

I think the bug might have something to do with having both versions installed.

I just did a yum remove iproute-2.6.32-31.el6.x86_64 and I get an "unfinished transactions" error from Yum.

Me thinks there is more going on here!

Edit: well, it was only on this 1 host, so no problem...
Posted by RRWH, 12-02-2013, 11:14 PM
Anyone (yet) done any testing where the node and VM are both updated?

I have not got access to my dev/test box for a few hours to run this one up and see how it fares if both the HV and VM are updated.
Posted by rds100, 12-03-2013, 01:38 AM
Here is another possible workaround i wrote a few days ago - https://www.fitvps.com/members/annou...orkaround.html

What it does is modifying the ifup-eth script to use "ifconfig xxxx up" instead of "ip link set dev xxxx up", which is the failing command leaving your container without networking.
Posted by epadicom, 12-03-2013, 08:35 AM
its work on my server.
Thank you for your advise.
Posted by serve-you, 12-03-2013, 09:58 AM
I just updated a host node and all the containers on it, and didn't run into this issue. I wonder if there's some magic combination that's causing this. I run CentOS 6.5 x86_64 on the host node, and a mix of 6.5 x86 & x86_64 containers.
Posted by rds100, 12-03-2013, 10:03 AM
The problem only appears if the host node is CentOS 5 based and the container is CentOS 6.
Posted by serve-you, 12-03-2013, 10:07 AM

Quote:

Originally Posted by rds100

The problem only appears if the host node is CentOS 5 based and the container is CentOS 6.

That would definitely make sense.
Posted by mperkel, 12-03-2013, 11:23 AM
OK - I upgraded with no problems. but I have Centos 6 everywhere.
Posted by foobic, 12-05-2013, 09:38 PM

Quote:

Originally Posted by rds100

The problem only appears if the host node is CentOS 5 based and the container is CentOS 6.

If that's the case it seems like something that providers running CentOS 5 nodes should be warning their clients about. As a VPS user I'd certainly appreciate some advance warning that a simple yum update will leave me without networking, particularly where it's due to incompatibility with the host-node's OS.
Posted by bawhbb, Today, 09:36 PM
Upstream knows about the issue: kb.parallels.com/en/118992

Quote:

The issue is recognized as PCLIN-32219 and the fix is planned to be included in an upcoming Parallels Virtuozzo Containers 4.6 update.

CentOS 5 uses OpenVZ 4.6, which is based on Virtuozzo Containers 4.6