v3.1.x: Update to PMIx v2.2.3 pre-release candidate

[rhc54]

Will update to official release after passing MTT

Refs #6763

Signed-off-by: Ralph Castain [email protected]

[rhc54]

@artpol84 Is this some hiccup on the Mellanox Jenkins? I'm not quite sure I see how PMIx would have caused this failure, but I'm not ruling it out - just need guidance.

21:50:43 + taskset -c 2,3 timeout -s SIGSEGV 17m /var/lib/jenkins/jobs/gh-ompi-master-pr/workspace/ompi_install1/bin/mpirun -np 8 -bind-to none -mca orte_tmpdir_base /tmp/tmp.sBsKcd8v7T --report-state-on-timeout --get-stack-traces --timeout 900 -mca coll '^hcoll' -mca btl_openib_if_include mlx4_0:1 -x UCX_NET_DEVICES=mlx4_0:1 -mca btl_openib_allow_ib true -mca pml ob1 -mca btl self,openib taskset -c 2,3 /var/lib/jenkins/jobs/gh-ompi-master-pr/workspace/ompi_install1/examples/hello_c
21:50:43 --------------------------------------------------------------------------
21:50:43 WARNING: There is at least non-excluded one OpenFabrics device found,
21:50:43 but there are no active ports detected (or Open MPI was unable to use
21:50:43 them).  This is most certainly not what you wanted.  Check your
21:50:43 cables, subnet manager configuration, etc.  The openib BTL will be
21:50:43 ignored for this job.
21:50:43 
21:50:43   Local host: jenkins03
21:50:43 --------------------------------------------------------------------------
21:50:43 [1561575043.381564] [jenkins03:22333:0]    ucp_context.c:601  UCX  WARN  device 'mlx4_0:1' is not available
21:50:43 [1561575043.381933] [jenkins03:22289:0]    ucp_context.c:601  UCX  WARN  device 'mlx4_0:1' is not available
21:50:43 --------------------------------------------------------------------------
21:50:43 At least one pair of MPI processes are unable to reach each other for
21:50:43 MPI communications.  This means that no Open MPI device has indicated
21:50:43 that it can be used to communicate between these processes.  This is
21:50:43 an error; Open MPI requires that all MPI processes be able to reach
21:50:43 each other.  This error can sometimes be the result of forgetting to
21:50:43 specify the "self" BTL.
21:50:43 
21:50:43   Process 1 ([[27929,1],4]) is on host: jenkins03
21:50:43   Process 2 ([[27929,1],0]) is on host: jenkins03
21:50:43   BTLs attempted: self
21:50:43 
21:50:43 Your MPI job is now going to abort; sorry.
21:50:43 --------------------------------------------------------------------------
21:50:43 [1561575043.382579] [jenkins03:22294:0]    ucp_context.c:601  UCX  WARN  device 'mlx4_0:1' is not available
21:50:43 [1561575043.382703] [jenkins03:22374:0]    ucp_context.c:601  UCX  WARN  device 'mlx4_0:1' is not available
21:50:43 --------------------------------------------------------------------------
21:50:43 MPI_INIT has failed because at least one MPI process is unreachable
21:50:43 from another.  This *usually* means that an underlying communication
21:50:43 plugin -- such as a BTL or an MTL -- has either not loaded or not
21:50:43 allowed itself to be used.  Your MPI job will now abort.
21:50:43 
21:50:43 You may wish to try to narrow down the problem;
21:50:43 
21:50:43  * Check the output of ompi_info to see which BTL/MTL plugins are
21:50:43    available.
21:50:43  * Run your application with MPI_THREAD_SINGLE.
21:50:43  * Set the MCA parameter btl_base_verbose to 100 (or mtl_base_verbose,
21:50:43    if using MTL-based communications) to see exactly which
21:50:43    communication plugins were considered and/or discarded.
21:50:43 --------------------------------------------------------------------------
21:50:43 [jenkins03:22333] *** An error occurred in MPI_Init
21:50:43 [jenkins03:22333] *** reported by process [140735023742977,140733193388036]
21:50:43 [jenkins03:22333] *** on a NULL communicator
21:50:43 [jenkins03:22333] *** Unknown error
21:50:43 [jenkins03:22333] *** MPI_ERRORS_ARE_FATAL (processes in this communicator will now abort,
21:50:43 [jenkins03:22333] ***    and potentially your MPI job)

[rhc54]

Looking at the output from Mellanox Jenkins, this is clearly some problem with the openib btl - the exact same program ran successfully right before it with vader instead of openib.

[artpol84]

@karasevb please have a look

[jsquyres]

Same Mellanox CI error is occurring on #6779.

[jsquyres]

bot:mellanox:retest

[artpol84]

@jsquyres I've checked with @karasevb.
This seems like a HW issue on the jenkins node. It is still being fixed, so I expect this retest to fail, but hopefully, we will get it working by tomorrow.

[jsquyres]

Oh, ok. The retest on another PR just succeeded. Shrug. Thanks for the heads up.

[artpol84]

Looks like it went fixed recently.
All good

[jsquyres]

Once this PR is merged, we'll watch MTT. If it's all good, there will be another PR to get the final final final PMIx 2.2.3 release.