[mpich-discuss] CLOG write failure

[chong tan]

Yes, we did check.  We tried setting env var TMPDIR, we got pretty much
the same error, with the tmp being in the TMPDIR.  We also checked the
temp files, and the file that caused the error was there, 360M of it.
 
I read the code of clog_buffer.c, the ierr returned by the write is int, but block size
is unsigned.  I don't know if this is the problem.  I modified the error message
to print more info.  But the peroson running my application is on vacation for a 
week.  I will not be able to run it to see more info.
 
tan


--- On Fri, 2/27/09, Anthony Chan <chan at mcs.anl.gov> wrote:

From: Anthony Chan <chan at mcs.anl.gov>
Subject: Re: [mpich-discuss] CLOG write failure
To: "chong guan tan" <chong_guan_tan at yahoo.com>, mpich-discuss at mcs.anl.gov
Date: Friday, February 27, 2009, 9:56 AM

If TMPDIR is not set or it is not read by MPE for some
reasons, /tmp will be used for local storage.
The error message below shows that /tmp_proj is used
for local clog2 file, can you check if you have write
permission to /tmp_proj ?

A.Chan

----- "chong tan" <chong_guan_tan at yahoo.com> wrote:

> we ran our application, after a few hours, MPICH run into file write
> error and aborted. THe error :
>  
> clog_buffer.c:CLOG_Buffer_localIO_write() - 
>         Fail to write to the temporary logfile
> /tmp_proj/clog2_BgxoHx_taskID=000003.
>         Check if the filesystem where the logfile resides is full.
> If so, set environmental variable TMPDIR to a bigger filesystem.
> 
> tried setting TMPDIR to a disk with more than 20G of free space, still
> get the same error. 
> When it failed, the temporary log file was about 360M.
>  
> we ran 'limit' and filesize was set to unlimited.
>  
>  
> any suggestion ?
>  
> thanks
> tan



      
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/mpich-discuss/attachments/20090227/9a7dd0d2/attachment.htm>