[Labs-l] job won't stop; state "dr"

[Marc A. Pelletier]

On 10/22/2014 06:12 AM, Amir Ladsgroup wrote:
> For me doesn't matter the job is once or continuous. it won't stop with
> qdel and it become annoying since sometimes I'm killing a malfunctioned
> task and it counties and I can do nothing.

It would seem that the new gentler way of killing jobs to be deleted -
which was put in place to help one problem - is causing others.

For the record, qdel now tries to kill jobs by sending SIGINT to the
entire process group, where it previously sent SIGKILL to the process
group leader only (that default method may be reliable, but it's also
very disruptive).

Apparently, gridengine isn't smart enough to use progressive killing on
its own; I'll replace its kill method with a script sometime during the
day so that it can fall back to SIGKILL when the process doesn't react
to gentler nudging.

-- Marc