zbd/012: Test higher queue depths

[bvanassche]

Instead of only testing queue depth 1, also test higher queue depths.
Use libaio for higher queue depths since io_uring does not preserve the
write order if qd >= 2 and a request queue is frozen.

[bvanassche]

Cc-ing @damien-lemoal in case you would like to take a look.

[kawasaki]

Thanks Bart. I found the relevant discussion in the linux-block list. I still not sure if it is agreed that libaio should be used instead of i_uring for QD>2 case. Will wait hoping the dicussion gets concluded.

[bvanassche]

From https://lkml.kernel.org/linux-block/20241218065859.GA25215@lst.de/:

On Tue, Dec 17, 2024 at 12:59:00PM -0800, Damien Le Moal wrote:
> Thanks for the pointer. Will have a look. It may be as simple as always using
> the io-wq worker for zone writes and have these ordered (__WQ_ORDERED). Maybe.

No.  Don't force ordering on people for absolutely no reason.  Use the
new append writes that return the written offsets we've talked about.
Enforcing ordering on I/O submissions/commands/requests always ends up
badly.

Does this mean that io_uring won't support preserving the I/O submission order in the near future? Does that answer mean that there won't be a good alternative to libaio for zoned writes with QD > 1? Is this sufficient to have another look at this pull request?

[damien-lemoal]

Yes and yes is my answer to your first 2 questions. So I think that switching the test to using libaio is enough and makes the test reliable and valid.

[kawasaki]

@bvanassche I applied the patch as the commit 827924b. I took liberty to do some minor changes.
Anyway, thanks for the work!