On Wed, 15 Aug 2018 at 18:05, Gnana Sekhar <kgsgnana2020@xxxxxxxxx> wrote: > > Thanks for the detailed explanation Sitsofe. > > So if I have --percentage_random flag, fio doesn't fill 5% of the > drive or may exit without filling 5% of the drive? At least 5% of the size you specified won't be touched but because the distribution of regions used is no longer uniform you could end up with it being more than 5%. > If so is there a way to fill 5% partly sequential and partially > random. I mean, in the same job to have combination of both sequential > and random writes Not within the same job. If you wanted to do this you would split it into two jobs and have one job work on a region sequentially and the other job work on a region randomly. It's possible to tether jobs together using flow (https://fio.readthedocs.io/en/latest/fio_doc.html#cmdoption-arg-flow ) but it will complicate things a bit. > Sorry for side track, one another question, is there a way you know to > exit fio completely > There are some instances where drive dropped off from the system due > to timeout but fio process still exists. > With the parameters I shared in the thread before, fio seems to be > running. Tried with --exitall_on_error and --continue_on_error flags I know what you're talking about but no, there's very little fio can do. On Linux SCSI error handling should kick in when an I/O takes too long and eventually force the I/O to fail. However, if you're suppressing that by quietly holding the I/O forever (people using multipath are known to configure their systems in this fashion) userspace will never be told the I/O finished in any way and the knock on of this is that fio will fail to exit forever. See https://github.com/axboe/fio/issues/550#issuecomment-370127049 for a similar case with multipath . Additionally if you're trying to fake infinite I/O suspends the commands mentioned in https://www.spinics.net/lists/fio/msg05213.html can be useful. -- Sitsofe | http://sucs.org/~sits/
