[slurm-users] PMIX and slurm failure (and fix).
Bill Broadley
bill at cse.ucdavis.edu
Thu May 17 05:26:10 MDT 2018
Greetings all,
Just wanted to mention I build building the newest slurm on Ubuntu 18.04.
Gcc-7.3 is the default compiler, which means that the various dependencies
(munge, libevent, hwloc, netloc, pmix, etc) are already available and built with
gcc-7.3.
I carefully built slurm-17.11.6 + openmpi-3.1 to use the same pmix-2.1.1. But
at runtime get a pmix failure, and a warning to use --mpi=list to see what was
supported... which included pmix_v2.
I very carefully reviewed the final report of slurm's ./configure, nothing
there. I grepped the config.log for any mention of pmix, which looked good.
Similar for building openmpi.
Turns out in the 1000s of line of output there was a line that mentions openssl
support was missing (I hadn't installed libssl-dev). Turns out that srun+pmix
depends on openssl and gives a failure message that looks just like pmix is failing.
So definitely grep your config.log for WARNING and keep an eye out for anything
important missing. My final fix to get rid of most of the warnings was to
install the following packages (in puppet, but it's easy to read):
package { [libmysqld-dev,man2html,"libgtk2.0-dev","librrd-dev",
"libudev-dev","libjansson-dev","libssl-dev","libpmix-dev"]: ensure => latest }
Was hoping that this might prevent someone else from having the same frustration.
More information about the slurm-users
mailing list