A little more info for you:

When I constrain 'srun' with 'ntasks-per-node', it seems to do OK . . .?

I'm forwarding this to slurm-dev.


SD: I'm having a problem using srun -N 2 -n 8, as you can see below . . .


-------- Forwarded Message --------
Subject:        segfault on certain configuration
Date:   Wed, 24 May 2017 16:28:21 -0700
From:   Geof Sawaya <[email protected]><mailto:[email protected]>
To:     
[email protected]<mailto:[email protected]> 
<[email protected]><mailto:[email protected]>



Hi MVAPICH people, I have another problem.

I am using the GDR (2.2) version with Slurm 17.


I have this simple MPI program that segfaults occasionally (depending
upon the number of processes per node).

Here is a small case, where I use srun -N 2 -n 8 (2 nodes, 8 tasks) and get:

2 procs on node 2, and 6 on node 1.   Rank 5 crashes.


Here's the trace:


Program received signal SIGSEGV, Segmentation fault.
0x00007f5c28128a12 in post_srq_send () from
/opt/mvapich2/gdr/2.2/cuda8.0/gnu/lib64/libmpi.so.12
(gdb) bt
#0  0x00007f5c28128a12 in post_srq_send () from
/opt/mvapich2/gdr/2.2/cuda8.0/gnu/lib64/libmpi.so.12
#1  0x00007f5c2812d532 in vbuf_address_send () from
/opt/mvapich2/gdr/2.2/cuda8.0/gnu/lib64/libmpi.so.12
#2  0x00007f5c2812e8f9 in MPIDI_CH3I_MRAIL_Parse_header () from
/opt/mvapich2/gdr/2.2/cuda8.0/gnu/lib64/libmpi.so.12
#3  0x00007f5c28114a2a in handle_read () from
/opt/mvapich2/gdr/2.2/cuda8.0/gnu/lib64/libmpi.so.12
#4  0x00007f5c2811519b in MPIDI_CH3I_Progress () from
/opt/mvapich2/gdr/2.2/cuda8.0/gnu/lib64/libmpi.so.12
#5  0x00007f5c280be09d in MPIC_Wait () from
/opt/mvapich2/gdr/2.2/cuda8.0/gnu/lib64/libmpi.so.12
#6  0x00007f5c280be976 in MPIC_Sendrecv () from
/opt/mvapich2/gdr/2.2/cuda8.0/gnu/lib64/libmpi.so.12
#7  0x00007f5c27e8c95c in MPIR_Pairwise_Barrier_MV2 () from
/opt/mvapich2/gdr/2.2/cuda8.0/gnu/lib64/libmpi.so.12
#8  0x00007f5c27e8caa1 in MPIR_Barrier_intra_MV2 () from
/opt/mvapich2/gdr/2.2/cuda8.0/gnu/lib64/libmpi.so.12
#9  0x00007f5c27e8cc89 in MPIR_Barrier_MV2 () from
/opt/mvapich2/gdr/2.2/cuda8.0/gnu/lib64/libmpi.so.12
#10 0x00007f5c27e2fc0f in MPIR_Barrier_impl () from
/opt/mvapich2/gdr/2.2/cuda8.0/gnu/lib64/libmpi.so.12
#11 0x00007f5c28087f05 in PMPI_Finalize () from
/opt/mvapich2/gdr/2.2/cuda8.0/gnu/lib64/libmpi.so.12
#12 0x0000000000401065 in main (argv=1, argc=0x7ffeb2d7dd78) at test.cpp:49
(gdb)


Does this give you enough to go on to (hopefully) tell me / guess how
I've misconfigured?


Many thanks -- Geof





Reply via email to