Why not abort when posix_fallocate() failed?

Why not abort when posix_fallocate() failed?

[xan.peng]

Hi all,

I'm recently reading fio source code and running it for test purpose.
A case bothered me which is fio keep going after posix_fallocate()
failed, I'm wondering why, shouldn't it abort immediately?

Information may be useful followed:

==part of the output==
fio: set debug option file
libaio.dat: (g=0): rw=randrw, bs=8K-8K/8K-8K/8K-8K, ioengine=libaio, iodepth=16
libaio.dat: Laying out IO file(s) (1 file(s) / 2048MB)
fio: posix_fallocate fails: No space left on device
fio: ENOSPC on laying out file, stopping
fio: io_u error on file /mnt/nfs/libaio.dat.1.0: No space left on device
     write offset=1398325248, buflen=8192
fio: io_u error on file /mnt/nfs/libaio.dat.1.0: No space left on device
     write offset=1994014720, buflen=8192
fio: pid=7152, err=28/file:io_u.c:1377, func=io_u error, error=No
space left on device

==reproduce==
dd if=/dev/zero of=/tmp/1gdisk bs=1M count=1024
losetup /dev/loop0 /tmp/1gdisk
mkfs.ext4 /dev/loop0
mount /dev/loop0 /mnt/loopfs
cat aio-bench # simplified here
// [global] ioengine=libaio,direct=1,rw=randrw,bs=8k,directory=/mnt/loopfs,time_based,runtime=10
// [libaio.dat] size=2g,iodepth=16
fio aio-bench

-- 
- Thanks.
- xan.peng

Re: Why not abort when posix_fallocate() failed?

[Jens Axboe]

On Mon, Aug 12 2013, xan.peng wrote:
> Hi all,
> 
> I'm recently reading fio source code and running it for test purpose.
> A case bothered me which is fio keep going after posix_fallocate()
> failed, I'm wondering why, shouldn't it abort immediately?

Fio treats it as advisory, like fadvise. Back in the day, fallocate
would often not be available. So it would be hard to treat it as a full
error.

-- 
Jens Axboe

Re: Why not abort when posix_fallocate() failed?

[xan.peng]

On Tue, Aug 13, 2013 at 10:43 PM, Jens Axboe <axboe@kernel.dk> wrote:
>> I'm recently reading fio source code and running it for test purpose.
>> A case bothered me which is fio keep going after posix_fallocate()
>> failed, I'm wondering why, shouldn't it abort immediately?
>
> Fio treats it as advisory, like fadvise. Back in the day, fallocate
> would often not be available. So it would be hard to treat it as a full
> error.

Thanks. This sounds very reasonable to me.
-- 
- Thanks.
- xan.peng

FIO-2.1.2 Windows Binaries Availability

[Ritchie Babaylan]

Hi,

I'd like to know if someone already made a Windows binaries for x32 and x64 platform for FIO-2.1.2. I check http://www.bluestop.org/fio/, they don't have yet.

�
Thanks,
�
Ritchie Babaylan
Advance CORE Engineering
BiTMICRO Networks, Inc.

Encountered Mix Access R/W Problem

[Ritchie Babaylan]

Hi,

I use FIO-2.1.1 in one of my system running a simple workload with the following details.
	64KB; 60% Sequential 60% Read; 1-Depth / 1-Job
	64KB; 30% Sequential 60% Read; 1-Depth / 1-Job

The result of the metrics for Read/Write access are zero as reflected on the command line in Windows 7 32-bit. I'd like to seek help why I am getting a zero metrics on Read/Write access.

�
Thanks,
�
Ritchie Babaylan
Advance CORE Engineering
BiTMICRO Networks, Inc.

Re: Encountered Mix Access R/W Problem

[xan.peng]

On Fri, Aug 16, 2013 at 1:44 PM, Ritchie Babaylan
<Ritchie.Babaylan@bitmicro.com> wrote:
> Hi,
>
> I use FIO-2.1.1 in one of my system running a simple workload with the following details.
>         64KB; 60% Sequential 60% Read; 1-Depth / 1-Job
>         64KB; 30% Sequential 60% Read; 1-Depth / 1-Job
>
> The result of the metrics for Read/Write access are zero as reflected on the command line in Windows 7 32-bit. I'd like to seek help why I am getting a zero metrics on Read/Write access.

I'm a rookie to fio :). And I have seen zero metics when I used loop
device on Ubuntu. Did you use a pseudo-device?
Can you post the whole fio script content?
-- 
-Thanks.
- xan.peng