Global filename/job filename interaction

Global filename/job filename interaction

[Phillip Chen]

Hello,
I was surprised to find that global filenames are added to job filenames rather than being replaced by job filenames. For instance, running
fio --ioengine=libaio --io_size=40k --readwrite=randread --filename=/dev/sdd --direct=1 --exitall --debug=io --name=sdb_randread_1_4k --filename=/dev/sdb
with fio-3.30 will show reads to both /dev/sdb and /dev/sdd. Is this expected behaviour? And are there other parameters that work in this way? I couldn't find anything in the job file parameters documentation mentioning this behaviour, but perhaps I was looking in the wrong place or missed it.
Thank you for your attention,
Phillip Chen

Seagate Internal

RE: Global filename/job filename interaction

[Vincent Fu]

> -----Original Message-----
> From: Phillip Chen [mailto:phillip.a.chen@seagate.com]
> Sent: Thursday, May 26, 2022 4:50 PM
> To: fio@vger.kernel.org
> Subject: Global filename/job filename interaction
> 
> Hello,
> I was surprised to find that global filenames are added to job filenames
> rather than being replaced by job filenames. For instance, running
> fio --ioengine=libaio --io_size=40k --readwrite=randread --
> filename=/dev/sdd --direct=1 --exitall --debug=io --
> name=sdb_randread_1_4k --filename=/dev/sdb
> with fio-3.30 will show reads to both /dev/sdb and /dev/sdd. Is this
> expected behaviour? And are there other parameters that work in this
> way? I couldn't find anything in the job file parameters documentation
> mentioning this behaviour, but perhaps I was looking in the wrong place
> or missed it.
> Thank you for your attention,
> Phillip Chen
> 
> Seagate Internal

options.c:str_filename_cb parses the filename option. Since fio allows
users to specify multiple files by separating names with a colon, for
each filename specified the function calls add_file() to add the new
file to the job's file array.

I believe what is happening is that sdd was added to the job's file
array when the [global] section was parsed and then sdb was added to the
file array when the [sdb_randread_1_4k] section was parsed. So the job
ran as if you had specified filename=/dev/sdd:/dev/sdb.

Unless someone wants to change this behavior perhaps the thing to do is
to document it.