On Nov 30, 2022 / 05:14, Chaitanya Kulkarni wrote: > On 11/29/22 18:40, Shin'ichiro Kawasaki wrote: > > The test case block/017 often fails on slow test systems. When it runs > > on QEMU and kernel with LOCKDEP, it fails around 50% by chance with > > error message as follows: > > > > block/017 (do I/O and check the inflight counter) [failed] > > runtime 1.715s ... 1.726s > > --- tests/block/017.out 2022-11-15 15:30:51.285717678 +0900 > > +++ /home/shin/kts/kernel-test-suite/src/blktests/results/nodev/block/017.out.bad 2022-11-25 16:23:50.778747167 +0900 > > @@ -6,7 +6,7 @@ > > sysfs inflight reads 1 > > sysfs inflight writes 1 > > sysfs stat 2 > > -diskstats 2 > > +diskstats 1 > > sysfs inflight reads 0 > > sysfs inflight writes 0 > > ... > > > > The test case issues one read and one write to a null_blk device, and > > checks that inflight counters reports correct numbers of inflight IOs. > > To keep IOs inflight during test, it prepares null_blk device with > > completion_nsec parameter 0.5 second. However, when test system is slow, > > inflight counter check takes long time and the read completes before the > > check. Hence the failure. > > > > To avoid the failure, extend the inflight duration of IOs. Prepare a > > null_blk device without completion_nsec parameter and measure time to > > check the inflight counters. Prepare null_blk device again specifying > > completion_nsec parameter 0.5 seconds plus the measured time of inflight > > counter check. > > > > Signed-off-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@xxxxxxx> > > overall it looks good to me, I trust that you have tested this on the > fast machine also and make sure it doesn't regress, with that said :- > > Reviewed-by: Chaitanya Kulkarni <kch@xxxxxxxxxx> Thanks for the review. I've applied it. Yes, I tested with multiple machines including a bare metal fast machine, and saw no regression. -- Shin'ichiro Kawasaki
