ci: add --bind-to none to Phoenix syscheck smoke-test (fixes GPU-node MPI bind failure)

[sbryngelson]

Symptom

Post-merge, Phoenix Test Suite jobs began failing (runs 926a3a5, 5829daf) with:

hwloc_set_cpubind returned "Error" for bitmap "0"   (orte/mca/rtc/hwloc/rtc_hwloc.c:382)
Open MPI tried to bind a new process ... process was killed without launching
Application name: build/install/gpu-acc-*/bin/syscheck

Root cause

build.sh's post-build syscheck smoke-test launched mpirun -np 1 "$syscheck_bin" without --bind-to none. On a busy GPU node SLURM hands the job an offset/partial cpuset (e.g. cores 32-47), and Open MPI's default --bind-to core then fails at hwloc_set_cpubind before the binary starts.

This was a latent bug: every other MPI launch already passes --bind-to none — the .mako run templates, coverage_build.py, and preflight.sh. This smoke-test was the lone omission. It stayed hidden while test jobs landed on idle nodes; #1833's partition-list change let them backfill onto shared GPU nodes with awkward cpusets, exposing it.

Evidence

The failing CI job already contains the perfect A/B — same node, same binary, -np 1:

Launch	Command	Result
preflight.sh	mpirun --bind-to none -np 1 syscheck	passed
build.sh smoke-test	mpirun -np 1 syscheck	failed (hwloc_set_cpubind)

Reproduced directly on multiple, distinct GPU nodes (h100-HGX atl1-1-03-008-22-0, h200 atl1-1-02-012-23-0 — neither is a CI node), confirming it is not node-specific:

mpirun -np 2/4 hostname            (default)   -> rc=131 "binding generic error"
mpirun --bind-to none -np 2/4 …    (suppressed) -> rc=0

A CPU node with an equally fragmented cpuset did not reproduce — it is specific to the GPU-node topology.

Fix

Add --bind-to none to the smoke-test mpirun, matching every other launch in the repo. One line.

Verification

• bash -n and python3 toolchain/mfc/lint_source.py pass locally.
• Grep confirms this was the only bare mpirun launch left in .github/.

[Copilot]

🟢 Approval recommended

The change is minimal, CI-scoped, and matches existing MPI launch behavior to address a clearly identified failure mode.

Pull request overview

This PR fixes Phoenix CI syscheck smoke-test failures on shared GPU nodes by aligning the mpirun invocation in the build workflow with the repository’s existing MPI launch pattern (--bind-to none), preventing Open MPI from failing early during CPU binding on fragmented SLURM cpusets.

Changes:

• Add --bind-to none to the Phoenix post-build syscheck smoke-test mpirun command.
• Expand the surrounding comment to document why binding must be disabled on busy GPU nodes.

File summaries

File	Description
.github/workflows/common/build.sh	Updates the Phoenix syscheck smoke-test to use mpirun --bind-to none to avoid hwloc/CPU-bind failures under SLURM cpusets.

Review details

• Files reviewed: 1/1 changed files
• Comments generated: 0
• Review effort level: Lite

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 62.35%. Comparing base (b44c811) to head (1bcf036).

Additional details and impacted files

@@           Coverage Diff           @@
##           master    #1834   +/-   ##
=======================================
  Coverage   62.35%   62.35%           
=======================================
  Files          84       84           
  Lines       21621    21621           
  Branches     3199     3199           
=======================================
  Hits        13482    13482           
  Misses       5939     5939           
  Partials     2200     2200           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.