Re: [PATCH v8] nvmet: add cgroup_id to charge namespace I/O to a cgroup

[peng yu]

On Wed, Oct 21, 2026 at 2:57 PM Tejun Heo <tj@xxxxxxxxxx> wrote:
>
> Hello, Peng.
>
> On Tue, Oct 20, 2026 at 06:01:12PM -0700, peng yu wrote:
> > Hi Tejun, does this patch look good from the cgroup perspective?
>
> From the cgroup core side, this looks okay to me. However, the current
> implementation will block the shared submitter when throttled, and it
> would probably need more work to behave well.

Currently the io.max won't block the shared submitters but io.cost and
io.latency will. If a nvmet subsystem has two namespaces, one namespace is
blocked by the io.cost or io.latency, the IOs of another namesapce will be
blocked too. I plan to create a v9 patch, if a namespace has a cgroup, it
will submit the IOs via the nvmet_wq, then this namespaces won't impact each
other. If it is not the correct way, please let me know.


> Thanks.
>
> --
> tejun