Parallels to merge OpenVZ and Cloud Server [LWN.net]

Parallels has announced that it will be merging its open-source OpenVZ and proprietary Parallels Cloud Server projects. "Now it's time to admit -- over the course of years OpenVZ became just a little bit too separate, essentially becoming a fork (perhaps even a stepchild) of Parallels Cloud Server. While the kernel is the same between two of them, userspace tools (notably vzctl) differ. This results in slight incompatiblities between the configuration files, command line options etc. More to say, userspace development efforts need to be doubled." The result of the merger will be open source; the name will be "Virtuozzo Core."

to post comments

Parallels to merge OpenVZ and Cloud Server

Posted Dec 29, 2014 1:07 UTC (Mon) by einstein (subscriber, #2052) [Link] (6 responses)

This is great news. Virtuozzo has long been the undisputed leader in Linux containers, with features that completely outclass any other implementation e.g. solaris zones. The openvz folks are responsible for virtually all of the container technology that exists today in the mainline kernel. In my former job as a unix engineer at a fortune 100 company, we used vz containers to run full blown multi-role production servers, and I can tell you they are full featured, rock solid and bare metal fast. My only complaint was that the vz kernel was not mainline, and the version lagged mainline by an appreciable amount of time. For instance, the latest vz server still ships with a kernel based on 2.6.32. However this latest move ought to bring everything into sync and reduce that gap to the bare minimum.

Parallels to merge OpenVZ and Cloud Server

Posted Dec 29, 2014 2:49 UTC (Mon) by dowdle (subscriber, #659) [Link] (1 responses)

The bare minimum you speak of will almost certainly 3.10.x for some value of 3.10.x as it is based on the RHEL7 kernel sources... and quite a bit of stuff (drivers especially) get backported over time. And again, the current stable OpenVZ kernel isn't stock 2.6.32... but the RHEL6-based 2.6.32 that has gotten more and more updates over time.

Parallels to merge OpenVZ and Cloud Server

Posted Dec 29, 2014 4:45 UTC (Mon) by einstein (subscriber, #2052) [Link]

Yep, assuming all the right pieces eventually get backported. From what I've heard, you can actually bring up a vz container on a stock 3.14 kernel, even though it lacks a few critical pieces of functionality related to the memory controller. The vz folks hope to get the most important missing pieces merged by 3.20 - and if all that good stuff gets backported to RHEL 3.10, well then we're good to go.

Parallels to merge OpenVZ and Cloud Server

Posted Dec 29, 2014 21:24 UTC (Mon) by ebiederm (subscriber, #35028) [Link] (2 responses)

Small correction. The openvz folks have done a lot. Some years and in some places more than others. It is not at all true that the openvz folks are responsible for virtually all of the container technology that exists today in the mainline kernel.

The code that exists in the mainline kernel has been a collaboration by a lot of companies with very significant contributions from folks other than Parallels employees. IBM, Ubuntu, Vserver, Arista Networks, Linux Networks, Suse, NEC, Redhat, Google, Twitter, grad students, and a bunch of companies I can't think of off the top of my head.

Go ahead and root for your favorite group or favorite userspace implementation but please remember up in kernel land it is a lot of folks collaborating to put together code that is better than any of us could build individually.

Parallels to merge OpenVZ and Cloud Server

Posted Dec 31, 2014 6:39 UTC (Wed) by dowdle (subscriber, #659) [Link] (1 responses)

The impression I always got was that because none of the mainline container feature developers cared much about a full-featured container... they all created the separate little piece they were interested in... and as a result the various pieces don't fit together well... so the sum of the pieces falls way short when compared to OpenVZ. It is even more clear when comparing the userspace tools.

So I think this is an instance where a single company trying to solve the complete problem does a much better job then a bunch of companies solving smaller, individual problems that are sort of related.

Parallels to merge OpenVZ and Cloud Server

Posted Dec 31, 2014 8:51 UTC (Wed) by Lennie (subscriber, #49641) [Link]

The reason is: the Linux kernel has no container support.

It has a whole bunch of features which together can be used to create container or container-like things or something totally different.

Parallels to merge OpenVZ and Cloud Server

Posted Dec 31, 2014 0:14 UTC (Wed) by LightDot (guest, #73140) [Link]

Hm, I don't think these news mean that remaining OpenVZ kernel code will get mainlined any sooner.

A lot has already been done and, yes, you can play with OpenVZ userland tools and a mainline kernel today. The vzctl-core package has been in Fedora for a while now, give it a spin.

That being said, OpenVZ production kernel has always been based on a RHEL kernel, and that's IMHO going to continue. Even if eventually all the crucial OpenVZ pieces get into mainline (or moved to userland), I seriously doubt Parallels will ever do much testing on vanilla kernels. So, the "gap" mentioned here is actually a gap between a vanilla kernel and a RHEL kernel. And that won't change.

But any OpenVZ or RHEL user should understand the (substantial!) difference between a vanilla 2.6.32.zz kernel and a RHEL's 2.3.32-xxx.n.m kernel.

As I write this, current stable OpenVZ RHEL 6 kernel is based on RHEL's 2.6.32-504.1.3.el6. OpenVZ RHEL 5 kernel (soon to be EOL) is based on RHEL's 2.6.18-400.el5. There is also an OpenVZ RHEL 7 kernel in development, based on RHEL 7's variant of 3.10.0-xxx.n.m.el7.

Anyone interested can check RHEL kernel's release dates and compare how quickly OpenVZ follows. There is also enough information about how a RHEL kernel relates to a vanilla kernel (backporting, etc.) freely available.

Parallels to merge OpenVZ and Cloud Server

Posted Dec 29, 2014 13:05 UTC (Mon) by dskoll (subscriber, #1630) [Link] (4 responses)

I wonder what Debian will do? They stopped shipping OpenVZ as of Wheezy and we had to make some rather annoying and painful migrations to LXC.

The OpenVZ userspace tools are much better than the LXC ones; I wonder if Debian will eventually reintegrate them? I'm willing to do one more painful container system migration... but only one!

Parallels to merge OpenVZ and Cloud Server

Posted Dec 29, 2014 14:39 UTC (Mon) by Lennie (subscriber, #49641) [Link] (3 responses)

I don't know why Debian stopped supporting OpenVZ, my guess would be because OpenVZ was a very large patch on some distribution Linux kernels and not a set of small patches.

But these days, all the container and container-like tools: LXC, Ubuntu LXD, OpenVZ, systemd-nspawn, Docker, CRIU and p.haul are all userspace tools which all use the same kernel features: cgroups, (network, user, etc.)namespaces and maybe a MAC like SELinux, AppArmor.

So packaging it should be a lot easier.

Parallels to merge OpenVZ and Cloud Server

Posted Dec 29, 2014 19:43 UTC (Mon) by tpo (subscriber, #25713) [Link] (1 responses)

> I don't know why Debian stopped supporting OpenVZ

I can tell you that, but can't give you the respective link to the mailing list posting (currently my internet is too slow for searching, yay!):

there was a call if anybody would be willing to support the migration to the next Debian release and on, and nobody stepped forward.

Parallels to merge OpenVZ and Cloud Server

Posted Dec 29, 2014 20:07 UTC (Mon) by Lennie (subscriber, #49641) [Link]

> there was a call if anybody would be willing to support the migration to the next Debian release and on, and nobody stepped forward.

A lot of the time this is because of an underlying issue like: it's a lot of work to package it because upstream doesn't offer patches in a usable way.

I was hoping someone knew. :-)

Parallels to merge OpenVZ and Cloud Server

Posted Dec 29, 2014 23:08 UTC (Mon) by dowdle (subscriber, #659) [Link]

Why did Debian stop providing their own OpenVZ packages? That is a real easy question.

While the OpenVZ Project has gone through quite a few kernel branches over the years, a few years back they dropped all branches except those based on RHEL kernels (RHEL5 and RHEL6 and RHEL7 coming). Trying to take patches from RHEL6 (2.6.32.x + lots of additions) and applying them to Debian's build of 2.6.32 in Debian 6... maybe doable... but when they moved to a much newer kernel with Debian 7... not so much.

So for those who insist on using Debian as their OpenVZ host node, they are recommended to the OpenVZ Project provided repo that basically repackages the .rpm builds of the RHEL-based branch to .deb files.

Even with the 2.6.32.x-based packages that Debian was providing, they couldn't really maintain them, there were several significant known bugs, and the userland tools were ancient.