On 10/7/22 21:05, Pankaj Raghav wrote: > Expand the reasoning behind using QD1 for zoned devices with io_uring_cmd > engine. > > Signed-off-by: Pankaj Raghav <p.raghav@xxxxxxxxxxx> > --- > examples/uring-cmd-zoned.fio | 8 ++++++-- > 1 file changed, 6 insertions(+), 2 deletions(-) > > diff --git a/examples/uring-cmd-zoned.fio b/examples/uring-cmd-zoned.fio > index 58e8f79e..89be61be 100644 > --- a/examples/uring-cmd-zoned.fio > +++ b/examples/uring-cmd-zoned.fio > @@ -1,7 +1,11 @@ > # io_uring_cmd I/O engine for nvme-ns generic zoned character device > # > -# NOTE: with write workload iodepth must be set to 1 as there is no IO > -# scheduler. > +# NOTE: > +# Regular writes against a zone should be limited to QD1, as the device can > +# reorder the requests. > +# > +# As the passthrough path do not use an IO scheduler (such as mq-deadline), > +# the queue depth should be limited to 1 to avoid zone invalid writes. Since all the writes issued by fio are "valid" in the first place, I would have phrased this: "...to avoid reordering of sequential write commands in the IO stack." > > [global] > filename=/dev/ng0n1 -- Damien Le Moal Western Digital Research
