On 08/17/2016 04:06 AM, Josh Boyer wrote:

On Tue, Aug 16, 2016 at 9:31 PM, Zbigniew Jędrzejewski-Szmek zbyszek@in.waw.pl wrote:

...

Unfortunately, due to the disagreements in the kernel development community, CPU controller cgroup v2 support has not been merged and enabling it requires applying two small out-of-tree kernel patches. The situation is explained in the following documentation:

https://git.kernel.org/cgit/linux/kernel/git/tj/cgroup.git/tree/Documentatio...

While it isn't clear what will happen with CPU controller cgroup v2 support, there are critical features which are possible only on cgroup v2 such as buffered write control making cgroup v2 essential for a lot of workloads. Also we finally have reliable empty cgroup reporting which avoids various systemd issues with zombie scope units and such.

Systemd recently gained experimental support for the new cgroup v2 cpu controller [1], which of course only works for people who a) are running with systemd.unified_cgroup_hierarchy=1 and b) have a patched kernel. We would like to move closer towards switching to v2 hierarchy, with an eye towards using unified hierarchy by default, but we have a chicken and egg problem where we cannot encourage people to switch and test systemd with v2 hierarchy, since the kernel support is incomplete. And the kernel support does not get enough testing because it's incomplete and requires a substantive effort.

Would it be possible to compile the rawhide kernels with the cgroup-v2-cpu patches [2]? This should cause no issue for v1 users, and would let the kernel and systemd functionality get wider testing, and hopefully nudge the kernel upstream towards finalizing the unified hierarchy support.

So the past 3 times we've done this, it didn't work out that way at all. Despite the kernel developers saying they want in-distro testing, it hasn't actually helped the upstream merge conversation. The utrace, Secure Boot, and kdbus code all was in Fedora for a very long time and only portions of utrace were ever merged. With kdbus is wasn't a huge issue because it was self contained within a module, but we're still carrying the Secure Boot patches today. The cgroup-v2 patches aren't self contained and if the upstream community continues to disagree we could be setting ourselves up for another round of patches we're carrying that aren't upstream.

Whether or not Fedora carries these isn't my call, but I'd approach this with caution. If the systemd developers are looking to simply have a kernel to test against, perhaps reviving the kernel-playground COPR with these would be enough.

I echo everything Josh wrote. I'm also wary because this is adding a userspace ABI. It's one thing to carry an in kernel work around or set of APIs but exposing something to userspace is a bigger issue. If the objections were "requires more testing" I'd be more willing but it sounds like the disagreement is over design. I'd also like to see a better long term plan than "hope the kernel community reaches a consensus that matches with this code". Convince me this will eventually get merged.

Thanks, Laura