Re: [PATCH v3 0/4] blk-iocost: charge flushes and zone appends

[Tejun Heo <tj@kernel.org>]

Hello, Tao.

On Mon, Sep 21, 2026 at 11:34:49AM +0800, Tao Cui wrote:
> While testing iocost's weight-based throttling under concurrent IO, we
> observed that a cgroup limited to 1% weight could issue an unbounded
> number of flushes without being throttled: an fsync loop produced ~510k
> flushes in 12s with cost.usage staying at zero the entire time.  The
> device was monopolized while iocost reported no activity.  Zone appends
> were in the same position: the builtin linear cost model defines
> coefficients only for READ and WRITE, so REQ_OP_ZONE_APPEND is priced
> at zero and excluded from the latency statistics as well.

Looks good to me. A few nits:

- 1/4: REQ_PREFLUSH and REQ_FUA are in REQ_NOMERGE_FLAGS, so !is_merge is
  always true there. Can you drop the wrapper and the separate flush_cost
  accumulator? Adding to cost before the empty bio check gives the same
  result, and the FUA condition fits on one line.

- 1/4: The comment mostly restates the two ifs and the coefficient
  formula. The FUA condition is the only non-obvious part. Can it be
  trimmed to that?

- 1/4: Pending pre-flushes get coalesced into one device flush, so the
  charge doesn't quite mirror what the block layer issues. It's a
  per-request policy charge, which the description already says. The
  other iops coefficients also have the page cost subtracted, so they
  aren't translated the same way. There's also a short broken line in
  that paragraph.

- 1/4: cgroup-v2.rst doesn't use REQ_* names anywhere else. Maybe
  describe them as a write with a preceding cache flush and a FUA write?
  The table row could follow its siblings too ("The maximum ... per
  second").

For the series,

Acked-by: Tejun Heo <tj@kernel.org>

Thanks.

-- 
tejun