Re: [PATCH] xfs: add per-mount read/write I/O completion counters

[Eric Peterson]

On Mon, Aug 31, 2026 at 07:26:52AM +1000, Dave Chinner wrote:
> Hence this really doesn't seem like something we should be trying to
> infer from indirect filesystem stats. Why can't you use the bdev
> stats to get the actual filesystem wide queue depth information?

The block device measures the device queue, which is a different
quantity than filesystem outstanding I/O - not just a lower-layer view
of the same thing.

Below are three cases where filesystem queue depth is not what the block
layer sees:

1. Cache hits never reach the block layer. Under a heavy read workload
   with a warm cache, a large share of ops are serviced from the page
   cache and are never seen at the block level. Device queue depth can
   sit near zero while the filesystem is servicing a very high op rate.

2. Filesystem ops don't map 1:1 to block I/O. A single read or write can
   produce one block I/O, several (metadata, readahead, writeback
   coalescing), or none at all. So device queue depth isn't the
   filesystem's outstanding-operation count.

3. Work can be outstanding inside the filesystem before any block I/O is
   issued - waiting on locks, log space/reservation, delalloc, etc.
   Such I/O has entered the filesystem but is invisible at the bdev.

The block-device queue depth answers "how deep is the device queue,"
which is not the same as "how much work is outstanding in the
filesystem." When the filesystem is just one layer an I/O passes
through, the block stats fold the layers together and structurally
cannot isolate the filesystem's own contribution.

To be clear about scope: I'm not proposing a queue-depth feature in
the kernel. The change just adds read/write completion counters to pair
with the existing call (submission) counters, so userspace can compute
outstanding I/O and derive a response-time estimate itself. The kernel
side is only exposing the complementary raw signal that's currently
missing - calls are counted, completions are not.

Being upfront: what userspace derives from this is an instantaneous
approximation, not a precise time-weighted queue length. It's meant as
a cheap, always-on aggregate, not a replacement for accurate per-op
tooling.

Does exposing the completion side of the existing call counters seem
reasonable on that basis?

-Eric