If loops= is specified, zone state is unexpectedly retained across iterations.

If loops= is specified, zone state is unexpectedly retained across iterations.

[Akash Verma]

I'm wondering if this is desired behavior or accidental. If I use this
control file:

[sdb-zoned-and-looped]
rw=write
bs=4k
size=24k
zonesize=12k
zonerange=512k
zoneskip=512k
ioengine=sync
direct=1
thread=1
loops=10
filename=/dev/sdb
write_iolog=sdb_iolog.txt

sdb_iolog.txt shows:
fio version 2 iolog
/dev/sdb add
/dev/sdb open
/dev/sdb write 0 4096
/dev/sdb write 4096 4096
/dev/sdb write 8192 4096
/dev/sdb write 1048576 4096
/dev/sdb write 1052672 4096
/dev/sdb write 1056768 4096
/dev/sdb close
/dev/sdb open
/dev/sdb write 1048576 4096
/dev/sdb write 1052672 4096
/dev/sdb write 1056768 4096
/dev/sdb write 2097152 4096
/dev/sdb write 2101248 4096
/dev/sdb write 2105344 4096
...

I'd expect each iteration of the loop to be identical, rather than
resuming from the zone offset of the previous iteration. Would it be
okay to change this behavior?

Re: If loops= is specified, zone state is unexpectedly retained across iterations.

[Akash Verma]

Adding Jens on cc.

On Tue, Sep 23, 2014 at 6:02 PM, Akash Verma <akashv@google.com> wrote:
> I'm wondering if this is desired behavior or accidental. If I use this
> control file:
>
> [sdb-zoned-and-looped]
> rw=write
> bs=4k
> size=24k
> zonesize=12k
> zonerange=512k
> zoneskip=512k
> ioengine=sync
> direct=1
> thread=1
> loops=10
> filename=/dev/sdb
> write_iolog=sdb_iolog.txt
>
> sdb_iolog.txt shows:
> fio version 2 iolog
> /dev/sdb add
> /dev/sdb open
> /dev/sdb write 0 4096
> /dev/sdb write 4096 4096
> /dev/sdb write 8192 4096
> /dev/sdb write 1048576 4096
> /dev/sdb write 1052672 4096
> /dev/sdb write 1056768 4096
> /dev/sdb close
> /dev/sdb open
> /dev/sdb write 1048576 4096
> /dev/sdb write 1052672 4096
> /dev/sdb write 1056768 4096
> /dev/sdb write 2097152 4096
> /dev/sdb write 2101248 4096
> /dev/sdb write 2105344 4096
> ...
>
> I'd expect each iteration of the loop to be identical, rather than
> resuming from the zone offset of the previous iteration. Would it be
> okay to change this behavior?

Re: If loops= is specified, zone state is unexpectedly retained across iterations.

[Jens Axboe]

On 2014-09-25 11:36, Akash Verma wrote:
> Adding Jens on cc.
>
> On Tue, Sep 23, 2014 at 6:02 PM, Akash Verma <akashv@google.com> wrote:
>> I'm wondering if this is desired behavior or accidental. If I use this
>> control file:
>>
>> [sdb-zoned-and-looped]
>> rw=write
>> bs=4k
>> size=24k
>> zonesize=12k
>> zonerange=512k
>> zoneskip=512k
>> ioengine=sync
>> direct=1
>> thread=1
>> loops=10
>> filename=/dev/sdb
>> write_iolog=sdb_iolog.txt
>>
>> sdb_iolog.txt shows:
>> fio version 2 iolog
>> /dev/sdb add
>> /dev/sdb open
>> /dev/sdb write 0 4096
>> /dev/sdb write 4096 4096
>> /dev/sdb write 8192 4096
>> /dev/sdb write 1048576 4096
>> /dev/sdb write 1052672 4096
>> /dev/sdb write 1056768 4096
>> /dev/sdb close
>> /dev/sdb open
>> /dev/sdb write 1048576 4096
>> /dev/sdb write 1052672 4096
>> /dev/sdb write 1056768 4096
>> /dev/sdb write 2097152 4096
>> /dev/sdb write 2101248 4096
>> /dev/sdb write 2105344 4096
>> ...
>>
>> I'd expect each iteration of the loop to be identical, rather than
>> resuming from the zone offset of the previous iteration. Would it be
>> okay to change this behavior?

Yeah I think that would be OK, basically all state is cleared between 
loops, so I'd consider that a bug. The whole point of a loop is that 
they are supposed to be identical.

-- 
Jens Axboe