slurm-llnl cannot execve scripts

[sysadmn]

Binary package hint: slurm-llnl

Description: Ubuntu 9.04
Release: 9.04
slurm-llnl:
  Installed: 1.3.13-1
  Candidate: 1.3.13-1
  Version table:
 *** 1.3.13-1 0
        500 http://us.archive.ubuntu.com jaunty/universe Packages
        100 /var/lib/dpkg/status

Any attempt to run a job gives the following slurm-nn.out:
slurmd[hostname]: execve(): /var/lib/slurm-llnl/slurmd/job00002/slurm_script: Bad address

It appears slurmctld creates a "job.nn" directory, but slurmstepd looks for a "jobs000nn" directory:
# strings /usr/sbin/slurmctld | grep /job
/job.%d
/job.%d/script
/job.%d/environment
/job_state.old
/job_state
/job_state.new
/job.%u
# strings /usr/sbin/slurmctld | grep /job
/job.%d
/job.%d/script
/job.%d/environment
/job_state.old
/job_state
/job_state.new
/job.%u

This occurs regardless of the value for SpoolDir in the config file.

[sysadmn (paul-joslin)]

Sorry, last command should have been:

# strings /usr/sbin/slurmstepd | grep /job
%s/job%05u
%s/job%05u.%05u

[moe jette (jette1)]

This problem is fixed in Slurm version 2.0.4 or you can apply the patch below.
The problem is caused by the sbatch command reading input from stdin and
failing to NULL terminate the argv argument to the execve() function. As a
work-around you can supply a batch script file as an argument to sbatch
rather than as stdin (e.g. "sbatch my.script"). Here's the patch:

Index: src/slurmd/slurmstepd/slurmstepd_job.c
===================================================================
--- src/slurmd/slurmstepd/slurmstepd_job.c (revision 18091)
+++ src/slurmd/slurmstepd/slurmstepd_job.c (revision 18092)
@@ -382,7 +382,7 @@
                /* job script has not yet been written out to disk --
                 * argv will be filled in later by _make_batch_script()
                 */
- job->argv = (char **) xmalloc(sizeof(char *));
+ job->argv = (char **) xmalloc(2 * sizeof(char *));
        }

        job->task = (slurmd_task_info_t **)

[Fabrice Coutadeur (fabricesp)]

Hi,

As lucid has version 2.0.5, I'll close this bug report as fixed. If you feel it's not the case, please reopen it by changing the status back to New.

Thanks,
Fabrice