[mpich-discuss] hydra and SIGCONT

Ashley Pittman ashley at pittman.co.uk
Mon Nov 30 11:37:58 CST 2009


In looking at getting hydra working with padb I've discovered that
mpiexec.hydra exits abnormally if you send it SIGCONT.  The default
action for this signal is to ignore it however hydra exits claiming to
have been killed by SIGTERM.

Would it make sense to remove this signal handler and leave the default
action in place?

In the past I've found that if you uncleanly detach from a process when
ptracing it the process can end up stopped in kernel (T state) and
SIGCONT is needed to wake it up again, to ensure this doesn't happen
padb always sends SIGCONT after detaching from a process which in this
case it causing it to exit.



Ashley Pittman, Bath, UK.

Padb - A parallel job inspection tool for cluster computing

More information about the mpich-discuss mailing list