[slurm-dev] test15.17 - Test the launch of a batch job within an existing job allocation

[Steven McDougall]

This test crashes slurmctld.
We're running on a stock x86 system.
You don't need the test to reproduce the failure; you can do it right on
the command line.
test17.19 does the same thing.

smcdougall@ssp015:~/SLURM/slurm-1.3.1/testsuite/expect>cat test15.17.input
#!/bin/bash
hostname

smcdougall@ssp015:~/SLURM/slurm-1.3.1/testsuite/expect>salloc bash
salloc: Granted job allocation 1121

smcdougall@ssp015:~/SLURM/slurm-1.3.1/testsuite/expect>sbatch
--jobid=1121 test15.17.input
sbatch: error: slurm_receive_msg: Zero Bytes were transmitted or received
sbatch: error: Batch job submission failed: Zero Bytes were transmitted
or received

smcdougall@ssp015:~/SLURM/slurm-1.3.1/testsuite/expect>sinfo
slurm_load_part: Unable to contact slurm controller (connect failure)

ssp015 smcdougall # ps u -C slurmctld
USER PID %CPU %MEM VSZ RSS TTY STAT START TIME COMMAND
ssp015 smcdougall # /etc/init.d/slurmctld status
* status: stopped

ssp015 smcdougall # tail -100 /var/log/messages-20080501 | grep slurm
2008-05-01T13:03:00-04:00 <kern:info> ssp015 slurmctld[10690]: segfault
at 0000000000000001 rip 00002ac0aa816ce0 rsp 0000000040b09ab8 error 4

[jet...@llnl.gov]

We can't reproduce this. Could you send me (not the mailing list)
your slurm.conf file and a backtrace plus log of the slurmctld.
Try running slurmctld in the foreground if you want:
/usr/sbin/slurmctld -Dvvvv