[padb-users] Trouble running padb with intelmpi

[Ashley Pittman]

On 12 May 2014, at 11:55, Duncan .H <harris.duncan at gmail.com> wrote:

> Hi,
> We're having problem running padb with intelMPI (4.x) on our systems
> and were hoping for some advice on tracking down the problem.
> 
> We keep getting these errors:
> 
> --------
> host3:~> padb --show-jobs
> 33224
> host3:~> padb -tx 33224
> No MPIR_proctable_size symbol found, cannot continue
> No suitable backend found (perhaps try installing pdsh or clush ?)!
> Fatal problem setting up the resource manager: mpirun
> ————

I think what is happening here is that the mpirun code is failing to find the process table and then the backend code is looking for a backend that can run on zero hosts, as such the pdsh message is a being erroneously reported.  I can fix that specific but obviously it won’t help you with the underlying problem here.

Mpich2 and hydra need to be built with the --enable-debuginfo configure flag to enable both the message queue support and also the MPI_proctable interface which allows debugger attach, I’m assuming the hydra you’re using is provided as part of intelMPI, in which case it would be worth asking them if they build with that option enabled.

Ashley,