blk-iolatency: deal with small samples
authorJosef Bacik <josef@toxicpanda.com>
Fri, 28 Sep 2018 17:45:41 +0000 (13:45 -0400)
committerJens Axboe <axboe@kernel.dk>
Fri, 28 Sep 2018 17:47:29 +0000 (11:47 -0600)
There is logic to keep cgroups that haven't done a lot of IO in the most
recent scale window from being punished for over-active higher priority
groups.  However for things like ssd's where the windows are pretty
short we'll end up with small numbers of samples, so 5% of samples will
come out to 0 if there aren't enough.  Make the floor 1 sample to keep
us from improperly bailing out of scaling down.

Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
block/blk-iolatency.c

index 8daea7a4fe492ebd732100ddb43fbb981d31eec8..e7be77b0ce8beb1c684383edfce21a7134814fb1 100644 (file)
@@ -366,7 +366,7 @@ static void check_scale_change(struct iolatency_grp *iolat)
                 * scale down event.
                 */
                samples_thresh = lat_info->nr_samples * 5;
-               samples_thresh = div64_u64(samples_thresh, 100);
+               samples_thresh = max(1ULL, div64_u64(samples_thresh, 100));
                if (iolat->nr_samples <= samples_thresh)
                        return;
        }