[GRASS-dev] Re: superfluous G_legal_filename etc calls

[Glynn Clements]

Markus Neteler wrote:

> Apparently I have even more (one of many Grid Engine job outputs):
> 
> cat launch_SGE_grassjob_rsun_energy.sh.e33397
> doy: 160 (month: 06)
> Processing day 160...
> Mode 2: integrated daily irradiation
> 0..Illegal filename.  Cannot be '.' or 'NULL'
> Illegal filename.  Cannot be '.' or 'NULL'
> Illegal filename.  Cannot be '.' or 'NULL'

...

> Illegal filename.  Cannot be '.' or 'NULL'
> /home/neteler/binaries/bin/rsun_photoperiod_PAT_utm32.sh: line 46:
> 7878 Killed                  r.sun -s --o $DEM day=$d
> linkein=linke_turbidity$month numpartitions=64 horizonstep=30
> horizon=pat5m_horangle insol_time=pat_insol_time$D
> beam_rad=pat_beam_rad$D diff_rad=pat_diff_rad$D
> refl_rad=pat_refl_rad$D glob_rad=pat_glob_rad$D
> ERROR: Raster map <pat_insol_time160> not found
> 
> Is there a possibility to find out why the job was killed?

"Killed" means SIGKILL. This can come from exceeding resource limits,
e.g. stack size (ulimit -s) or CPU time (ulimit -t), but most other
resource limits either can't be exceeded (e.g. RSS, ulimit -m), or
simply result in the relevant system call failing (e.g. data segment
size, ulimit -d).

Exceeding total virtual memory (ulimit -v) can result in SIGKILL if
the limit is exceeded due to stack extension rather than heap
extension (malloc).

And the OOM killer uses SIGKILL.

> In /var/log/messages there is no trace (no OOM).

Is that the appropriate log file? On my system, kernel messages never
go to that file:

	*.=info;*.=notice;*.=warn;\
	        auth,authpriv.none;\
	        cron,daemon.none;\
	        mail,news.none;\
	        kern.none               -/var/log/messages

kern.info and above go to /var/log/kern.log, while kern.debug is
ignored.

But there won't normally be anything in the logs if the process is
killed due to exceeding its own resource limits.

> I had declared that
> each job gets 11GB RAM in Grid Engine. Other jobs continue (same
> script, just another day of the year).

> GE is supposed to send me an email in case of error but nothing
> happens.

I suspect that "error" would mean that the script itself returned
non-zero or terminated due to a signal.

-- 
Glynn Clements <glynn at gclements.plus.com>