Bad verify header

Bad verify header

[Ricardo M. Correia]

(Please CC me in the reply, I'm not subscribed)

Hi,

I've justed observed another type of verification failure, using the following
configuration:

--------------------
[global]
ioengine=libaio
iodepth=30
rw=randwrite
bsrange=512-128k
direct=1
verify=sha512
verify_async=8

[/dev/sda2]
--------------------

The output is:

-------------------------
wizeman@cfslap:~/src/fio$ sudo ./fio verify.conf
/dev/sda2: (g=0): rw=randwrite, bs=512-128K/512-128K, ioengine=libaio, iodepth=30
Starting 1 process                                                               
Bad verify header 11a201ef at 7151022087K /s] [0/172 iops] [eta 03m:11s]         
fio: pid=30053, err=84/file:verify.c:840, func=async_verify, error=Invalid or incomplete multibyte or wide character
Bad verify header 713f2224 at 60533760                                                                              
Bad verify header 7df21ab6 at 44630528          
Bad verify header 1124297c at 50690048                                                                              
Bad verify header 6fc5d15f at 102814208                                                                             
Bad verify header 63469875 at 139774464                                                                             
Bad verify header 1743d35f at 164766720                                                                             
Bad verify header 29cbf9d7 at 173868544      
...
-------------------------

This happens on both RHEL5.4 and Ubuntu 9.10.
fio was compiled from source from the latest git commit.

Any ideas of what might be wrong?

Thanks in advance,
Ricardo

Re: Bad verify header

[Ricardo M. Correia]

Ok, this one appears to be because of the 'bsrange' option.

It works fine if I replace that option by 'bs=512' or 'bs=128k'.

Any chance of this problem being fixed, or is the 'verify' option
conceptually incompatible with 'bsrange'?

Cheers,
Ricardo

On Qui, 2010-02-18 at 01:48 +0000, Ricardo M. Correia wrote:
> (Please CC me in the reply, I'm not subscribed)
> 
> Hi,
> 
> I've justed observed another type of verification failure, using the following
> configuration:
> 
> --------------------
> [global]
> ioengine=libaio
> iodepth=30
> rw=randwrite
> bsrange=512-128k
> direct=1
> verify=sha512
> verify_async=8
> 
> [/dev/sda2]
> --------------------
> 
> The output is:
> 
> -------------------------
> wizeman@cfslap:~/src/fio$ sudo ./fio verify.conf
> /dev/sda2: (g=0): rw=randwrite, bs=512-128K/512-128K, ioengine=libaio, iodepth=30
> Starting 1 process                                                               
> Bad verify header 11a201ef at 7151022087K /s] [0/172 iops] [eta 03m:11s]         
> fio: pid=30053, err=84/file:verify.c:840, func=async_verify, error=Invalid or incomplete multibyte or wide character
> Bad verify header 713f2224 at 60533760                                                                              
> Bad verify header 7df21ab6 at 44630528          
> Bad verify header 1124297c at 50690048                                                                              
> Bad verify header 6fc5d15f at 102814208                                                                             
> Bad verify header 63469875 at 139774464                                                                             
> Bad verify header 1743d35f at 164766720                                                                             
> Bad verify header 29cbf9d7 at 173868544      
> ...
> -------------------------
> 
> This happens on both RHEL5.4 and Ubuntu 9.10.
> fio was compiled from source from the latest git commit.
> 
> Any ideas of what might be wrong?
> 
> Thanks in advance,
> Ricardo
> 
> 

Re: Bad verify header

[Jens Axboe]

On Thu, Feb 18 2010, Ricardo M. Correia wrote:
> Ok, this one appears to be because of the 'bsrange' option.
> 
> It works fine if I replace that option by 'bs=512' or 'bs=128k'.
> 
> Any chance of this problem being fixed, or is the 'verify' option
> conceptually incompatible with 'bsrange'?

If I recall correctly, it's due to overlaps of the ios screwing up the
verify headers. It's not impossible to fix, but you have to either relax
the IO coverage or make the random block generator more clever (so eg
once a 512b block has been allocated, you can only fit < 128KB blocks
into the remaining offsets).

-- 
Jens Axboe