Behavior of checkpoint_method on job termination

[wish]

If a job is terminated (with ENABLE_ADDGRP_KILL=true )while it is being checkpointed the ckpt_command is not killed by grid engine. This can cause issues with some checkpointing tools (eg ompi-checkpoint command from openmpi when used with blcr) which don't terminate if you kill the processes it is trying to checkpoint. This isn't too hard to work around but should be documented.

Possibly one could delay termination of a job until after ckpt_command has finished running.

[dlove]

Is the additional group not actually added to the command currently?

That would include the hook commands in the accounting. It's not clear
to me if that's appropriate but it seems reasonable. WDYT?

[Dave Love <d.love@…>]

In 4490/sge:

Fix #1449: add supplementary group to async-started commands
Enables them to be killed, but adds them to accounting -- arguably the
right thing anyway (e.g. expensive checkpointing).