Behaviour of fio with iodepth > 1 and write, and fsync=1

Travis Downs <[email protected]> Fri, 23 May 2025 11:05:46 -0400
Newsgroups org.kernel.vger.fio
Message-ID <CAAWAFP2JMTEUe_XUwW0xuGd2CODg2qPTC=DmY0qfik=U9XpDRQ@mail.gmail.com>
Given a fio benchmark with num_jobs=1, rw=write and iodepth=4 (or
whatever value > 1), nrfiles=1 and fysnc=1, as well as default
arguments for all the submission (`iodepth_batch_submit` et al) how
will the IOs and fsync be issued? If we number the blocks 0, 1, 2,
etc, will the initial submission be like:

write 0
fsync (file)
write 1
fsync (file)

I.e., writes to the first two blocks and (redundant) fsyncs to that
file, all in parallel? Then when any of those completes, write 2, then
another fsync and so on?

Thanks,
Travis