On 23 March 2018 at 16:08, Kris Davis <Kris.Davis@xxxxxxx> wrote:
> In fio version 2.20 a test seems to have been added to backend.c which "may" falsely report that no I/O has occurred.
> Currently, at line 1856 of backend.c there is the following test block, which doesn't appear to exist in versions prior to 2.20, unless the equivalent resided elsewhere.
>
> /*
> * If td ended up with no I/O when it should have had,
> * then something went wrong unless FIO_NOIO or FIO_DISKLESSIO.
> * (Are we not missing other flags that can be ignored ?)
> */
> if ((td->o.size || td->o.io_size) && !ddir_rw_sum(bytes_done) &&
> !did_some_io && !td->o.create_only &&
> !(td_ioengine_flagged(td, FIO_NOIO) ||
> td_ioengine_flagged(td, FIO_DISKLESSIO)))
> log_err("%s: No I/O performed by %s, "
> "perhaps try --debug=io option for details?\n",
> td->o.name, td->io_ops->name);
>
>
> In the following fio test, the error message is logged, even though it appears that everything has run ok (fio built from main below, but results appear to be the same with at least fio version 2.21)
> (It's also interesting to note that the estimated "eta" is also grossly inaccurate [captured while running below] - but that is a different issue I think).
> You can see the " Random-read-4K-QD1: No I/O performed by libaio, perhaps try --debug=io option for details?" error message below, which "appears" to be false.
>
> $ fio --ioengine=libaio --loops=32 --direct=1 --numjobs=1--norandommap --randrepeat=0 --size=16GB --filename=/dev/sdb --name=Random-read-4K-QD1 --rw=randread --bs=4K --iodepth=1 --number_ios=8192
> Random-read-4K-QD1: (g=0): rw=randread, bs=(R) 4096B-4096B, (W) 4096B-4096B, (T) 4096B-4096B, ioengine=libaio, iodepth=1
> fio-3.5-60-g69c5
> Starting 1 process
> Jobs: 1 (f=1): [r(1)][0.2%][r=30.9MiB/s,w=0KiB/s][r=7898,w=0 IOPS][eta 04h:45m:35s]
>
> Random-read-4K-QD1: No I/O performed by libaio, perhaps try --debug=io option for details?
>
[...]
>
> I added a print message which showed that both ddir_rw_sum(bytes_done) and did_some_io values were zero at that point in the code.
> I didn't try to dig into the operation of the bytes_done and did_some_io values further, yet.
>
>
> Also, I noticed that if option "--io_size=32M" was used instead of "number_ios=8192", then the eta appeared to be accurate - so the eta issue does "appear" to be independent - I can open a separate discussion or issue if desired.
Just quickly looking I think this is down to number_ios=<val> being
paired with loops=32 but I haven't had time to dig further...
--
Sitsofe | http://sucs.org/~sits/
--
To unsubscribe from this list: send the line "unsubscribe fio" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
