Re: unable to replay a trace generated w/ fio

[Jens Axboe <axboe@kernel.dk>]

On 2014-06-05 13:02, Allen Schade wrote:
> Jens Axboe <axboe@...> writes:
>
>>
>> On 2014-05-18 04:01, Alket Memushaj wrote:
>>> I am not having luck with 2.1.7 either, but anything up to 2.0.11
>>> works. It seems to terminate as soon as it completes reading the trace
>>> file - at the end of which I have /dev/sdb close as that's what's
>>> generated by fio itself when I set write_iolog=<file>
>>
>> I'll take a look, I bet it's an issue from around the time the file
>> opening logic was added.
>>
>> Can you send the exact job you used to create and replay files? Will
>> make my life easier.
>>
>
> Hi, I'm am running into this issue too.
>
> Here are simple instructions to replicate:
>
> [hdc-write]
> filename=/dev/hdc
> ioengine=sync
> iodepth=1
> direct=1
> thread=1
> continue_on_error=1
> read_iolog=trace.log
>
>
> trace.log:
> /dev/hdc add
> /dev/hdc open
> /dev/hdc write 2045557202944 8192
> /dev/hdc close
>
> ./fio trace.fio
> hdc-write: (g=0): rw=read, bs=4K-4K/4K-4K/4K-4K, ioengine=sync, iodepth=1
> fio-2.1.7
> Starting 1 thread
> Segmentation fault (core dumped)

OK, so the replay bug here is that you added a filename=. Due to a bug 
in the freeing of files logic (fio does do that for iolog loads), we 
don't reset this when freeing files. So fio attempts to close 2 files on 
exit, boom.

This is fixed in -git as of now. Replay of that file still looks weird, 
doesn't look like it actually does that IO. Will debug next...

-- 
Jens Axboe