Seems to me that we should just reset back to zero, and not worry about the io_size at all. This is what is screwing things up, and I can't think of why we would attempt to be more clever in wrapping around. The below should do the trick. David, once appveyor finishes, you should be able to download the build here and re-test: https://ci.appveyor.com/project/axboe/fio/build/job/p0misp4ehe96ti7d/artifacts diff --git a/io_u.c b/io_u.c index 4246edff0b2f..aac74bf6f7ad 100644 --- a/io_u.c +++ b/io_u.c @@ -363,14 +363,7 @@ static int get_next_seq_offset(struct thread_data *td, struct fio_file *f, if (f->last_pos[ddir] >= f->io_size + get_start_offset(td, f) && o->time_based) { - struct thread_options *o = &td->o; - uint64_t io_size = f->io_size + (f->io_size % o->min_bs[ddir]); - - if (io_size > f->last_pos[ddir]) - f->last_pos[ddir] = 0; - else - f->last_pos[ddir] = f->last_pos[ddir] - io_size; - + f->last_pos[ddir] = 0; loop_cache_invalidate(td, f); } On 11/01/2017 10:44 AM, Jens Axboe wrote: > I just tried to reproduce on Linux, and I can. If the files don't exist and > I run this job: > > [global] > blocksize=64k > direct=1 > group_reporting > rw=read > size=512m > > [asdf] > filename=testfile1:testfile2:testfile3 > > it works fine. Then I change the size to be 250m AND add time_based > and a longer runtime to ensure we hit the roll-around, and it fails: > > fio win.fio > asdf: (g=0): rw=read, bs=(R) 64.0KiB-64.0KiB, (W) 64.0KiB-64.0KiB, (T) 64.0KiB-64.0KiB, ioengine=psync, iodepth=1 > fio-3.1-76-g08bd-dirty > Starting 1 process > fio: io_u error on file testfile1: Invalid argument: read offset=21846, buflen=65536 > fio: pid=18198, err=22/file:io_u.c:1770, func=io_u error, error=Invalid argument > > So looks like a bug in the reset part. I'll take a look. > > > On 10/31/2017 05:06 PM, Sitsofe Wheeler wrote: >> OK I see the problem too when using this on Windows: >> $ ./fio --size=1g --filename >> fio1.tmp:fio2.tmp:fio3.tmp:fio4.tmp:fio5.tmp:fio6.tmp:fio7.tmp:fio8.tmp:fio9.tmp >> --name=go --create_only=1 >> $ ./fio --size=250m --filename fio1.tmp:fio2.tmp:fio3.tmp --name=go >> --direct=1 --bs=64k --runtime=1m --time_based --thread >> >> Using --debug=all shows the following: >> io 2796 getevents: 1 >> io 2796 io complete: io_u 0000000007BD8F00: >> off=87359488/len=65536/ddir=0io 2796 /fio3.tmpio 2796 >> file 2796 put file fio3.tmp, ref=2 >> file 2796 trying file fio1.tmp 11 >> file 2796 goodf=1, badf=2, ff=11 >> file 2796 get_next_file_rr: 00007FF5FEEA4610 >> file 2796 get_next_file: 00007FF5FEEA4610 [fio1.tmp] >> file 2796 get file fio1.tmp, ref=1 >> io 2796 fill_io_u: io_u 0000000007BD8F00: >> off=21846/len=65536/ddir=0io 2796 /fio1.tmpio 2796 >> io 2796 prep: io_u 0000000007BD8F00: >> off=21846/len=65536/ddir=0io 2796 /fio1.tmpio 2796 >> io 2796 queue: io_u 0000000007BD8F00: >> off=21846/len=65536/ddir=0io 2796 /fio1.tmpio >> helperthread 2796 >> 2796 since_ss: 0, next_ss: 1000, next_log: 495, msec_to_next_event: 136 >> file 2796 put file fio1.tmp, ref=2 >> io 2796 io_u_queued_complete: min=0 >> io 2796 getevents: 0 >> process 2796 pid=2936: runstate RUNNING -> FINISHING >> fio: pid=2936, err=22/file:ioengines.c:335, func=td_io_queue, >> error=Invalid argument >> >> off=21846 is not a multiple of 512. >> >> $ du -b fio*tmp >> 119304647 fio1.tmp >> 119304647 fio2.tmp >> 119304647 fio3.tmp >> 119304647 fio4.tmp >> 119304647 fio5.tmp >> 119304647 fio6.tmp >> 119304647 fio7.tmp >> 119304647 fio8.tmp >> 119304647 fio9.tmp >> >> I think this is all hinting at a bug in fio (perhaps when working with >> multiple pre-existing files that are bigger than size specified) but I >> think I'm done for the day... >> >> On 31 October 2017 at 22:51, David Hare <david.hare@xxxxxxxxxxxxxxx> wrote: >>> >>> Not sure what you mean by one file? The drives are local NTFS formatted with >>> Allocation unit size of 64k, they range in size from with are 186gb - 372gb. >> > > -- Jens Axboe -- 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
