[Swift-devel] Lammps on BGQ: task completes but status shows active

[Mihael Hategan]

I would put an strace in _swiftwrap around the executable to see what
keeps it from completing.

Mihael

On Mon, 2014-12-08 at 10:29 -0600, Ketan Maheshwari wrote:
> Hi Mihael, All,
> 
> Can you help debugging this issue.
> 
> On BG/Q (cetus), running lammps with provider coaster.
> 
> The symptom is that the lammps task completes but Swift still thinks it is
> running and continues to show "Active" status. Worker logs also show that
> the task is running. The _wrapperlog is stalled in EXECUTE stage.
> 
> The script (bg.sh) running in the qsub "script" mode invokes runjob and it
> seems that the line after runjob is not reached, meaning runjob does not
> return.
> 
> The same configuration (qsub in script mode and runjob with same
> parameters) , when run outside of Swift seems to be working (ie. script
> exits on completion).
> 
> Attaching the bg.sh, and a tarball with Swift run dir and worker log in
> DEBUG mode.
> 
> Thanks for any help further debugging this.
> 
> Best,
> Ketan