On Tue, Jan 12, 2021 at 10:59:58AM -0800, Shakeel Butt wrote:

On Tue, Jan 12, 2021 at 9:12 AM Johannes Weiner [off-list ref] wrote:

quoted

When a value is written to a cgroup's memory.high control file, the
write() context first tries to reclaim the cgroup to size before
putting the limit in place for the workload. Concurrent charges from
the workload can keep such a write() looping in reclaim indefinitely.

Is this observed on real workload?

Yes.

On several production hosts running a particularly aggressive
workload, we've observed writers to memory.high getting stuck for
minutes while consuming significant amount of CPU.

Any particular reason to remove !reclaimed?

It's purpose so far was to allow successful reclaim to continue
indefinitely, while restricting no-progress loops to 'nr_retries'.

Without the first part, it doesn't really matter whether reclaim is
making progress or not: we do a maximum of 'nr_retries' loops until
the cgroup size meets the new limit, then exit one way or another.