Hi, On 25 August 2017 at 23:11, Matthew Eaton <m.eaton82@xxxxxxxxx> wrote: > Hello fio group. > > I have a volume with 3573 files and I am trying to write to them 10 at > a time in sequential order so I have: > > [job] > direct=1 > bs=128k > rw=write > refill_buffers > opendir=/media/sped/rmfs > openfiles=10 > file_service_type=sequential > > The above is close to what I want. Fio sees all 3573 files but only > services 1 file at a time. > > If I remove file_service_type=sequential fio services 10 files at a > time but using lsof I see only the last file incrementing: > > fio 6003 root 3u REG 8,17 > 134217728 84934 /media/sped/rmfs/0001 > fio 6003 root 4u REG 8,17 > 134217728 84935 /media/sped/rmfs/0002 > fio 6003 root 5u REG 8,17 > 134217728 84936 /media/sped/rmfs/0003 > fio 6003 root 6u REG 8,17 > 134217728 84937 /media/sped/rmfs/0004 > fio 6003 root 7u REG 8,17 > 134217728 84938 /media/sped/rmfs/0005 > fio 6003 root 8u REG 8,17 > 134217728 84939 /media/sped/rmfs/0006 > fio 6003 root 9u REG 8,17 > 134217728 84940 /media/sped/rmfs/0007 > fio 6003 root 10u REG 8,17 > 134217728 84941 /media/sped/rmfs/0008 > fio 6003 root 11u REG 8,17 > 134217728 84942 /media/sped/rmfs/0009 > fio 6003 root 12u REG 8,17 > 134217728 90617 /media/sped/rmfs/0058 > > Fio version is latest from git: fio-3.0-9-gb427 > > Is this the correct behavior? Not having file_service_type seems to workforme: rm -rf /tmp/fio/ mkdir /tmp/fio/ for i in $(seq -w 1 20); do dd if=/dev/zero of=/tmp/fio/fiofile.$i bs=16k count=1; done fio --number_ios=5 --bs=4 --rw=write --refill_buffers --opendir=/tmp/fio/ --openfiles=10 --buffer_pattern='"fio "' --name=rr grep fio /tmp/fio/fiofile.* Binary file /tmp/fio/fiofile.01 matches Binary file /tmp/fio/fiofile.02 matches Binary file /tmp/fio/fiofile.03 matches Binary file /tmp/fio/fiofile.04 matches Binary file /tmp/fio/fiofile.05 matches Reading http://fio.readthedocs.io/en/latest/fio_doc.html#cmdoption-arg-file-service-type gives a few clues and tells us about the :postfix to get fio to "pick" a different file after postfix I/Os. It's kind of subtle but when you don't set the postfix it actually defaults to 1 so when file_service_type=roundrobin (the default) you should have gotten the behaviour you desired because behind the scenes it is actually behaving as if file_service_type=roundrobin:1 . The difference comes when you use file_service_type=sequential (which using the above logic will be quietly converted to file_service_type=sequential:1) which is something you can't really arrange with roundrobin. file_service_type=sequential is actually interpreted as "ignore the postfix" so you always stay on the current file until it is finished (i.e. fully written). Here the help is misleading as it implies the postfix is used with file_service_type=sequential when it never is. I can definitely see a case for the help to be clarified and if you have a patch to do this I'm sure it would be considered! The above guesswork was based on reading the HOWTO and https://github.com/axboe/fio/blob/2f282cecc27c58ed3aa1204f5bda98433502d7ff/io_u.c#L1269 . If you use --debug=file are there any more clues? -- 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
