On 02/01/2014 10:44 AM, Adrian Reyer wrote:
> Hi,
>
> On Sat, Feb 01, 2014 at 09:26:23AM +0000, Gordan Bobic wrote:
>> Can some of the LXC components in the kernel be used in place of the
>> relevant VServer components to make the patch smaller and easier to
>> maintain? Or if the VServer component is "better" (I use the term
>> loosely because what kernel maintainers deem "better" isn't
>> necessarily what other people deem "better"), trying to push a patch
>> to do that upstream?
>
> I have not checked the code, but I have the impression Herbert uses new
> features that arrive in the kernel and enhance or replace features he
> implemented himself earlier. E.g. memory limits now done with cgroups.
> I have the impression Herbert is not intrested in the hassle involved in
> trying to get some code included into mainline kernels.
> I personally regard the vserver an ever decreasing patch in terms of
> size as more features arrive within the kernel. If it evolves like I
> hope, one day we will have all features in mainline accompanied by
> util-vserver, which I like much, much better than any lxc-tools I have
> seen so far.
I rather suspect at least the CoW hard-link breaking patches are not
going to go upstream any time soon, and those is probably the biggest
reason why I use VServer.
>> I'm not sure about the general use case, but the impression I get
>> from listening on the list is that most people tend to use LTS
>> kernels because that is what the distro kernels tend to be based on.
>> Would it perhaps be more reasonable to focus on maintaining the
>> patch only for LTS kernels? They are still moving targets, but at
>> least they would be fewer moving targets.
>
> We mostly use the 3.4 and 3.10 kernels we provide ourselves and for many
> szenarios a longterm supported kernel is just the right thing. However,
> there are szenarios where you need more recent kernels and would like to
> have VServer support nonetheless. Obvious one are new hardware onlyrelatively
> supported by recent kernels and missing backports. A new feature would
> e.g. live migration of a VServer to some other host by using criu
> (http://criu.org/). I have not looked deeply into this one, minimum
> kernel is 3.11. But it seems to me criu could enable me to actually
> live-migrate VServers I run on top of DRBD-Clusters. If I get it right,
> criu actually came from OpenVZ and the most recent LXC uses criu for
> exactly the same purpose.
> To my understanding, VServer can do everything mainline can plus the
> added features. Same time not limited to a specific cpu architecture. If
> mainline supports freezing tasks, VSevrer will. If mainline starts and
> cook dinner, VServer will be able to do so as well.
I understand where you are coming from, but LTS kernels aren't that few.
Between Ubuntu LTS release kernels and RHEL release kernels, there is a
decent compromise between chasing goalposts against mainline and saving
some of Herbert's many man-hours spent on this project with a negligible
impact on the vast majority of users.
I know just how much effort it takes to maintain an open source project,
and sometimes cutting the odd unnecessary corner is the only sane way to
stay on top of things.
Gordan
Received on Sat Feb 1 12:11:15 2014