Fix MPI on candide (OpenMPI 5 image + latent code bug); containerize & SLURM-ify candide scripts#737

Open

cailmdaley wants to merge 21 commits into

developfrom

cleanup/candide-scripts-container

Commits on May 30, 2026

Run candide PBS job scripts through the container instead of conda

cailmdaley
and
claude
committed

Commits on May 31, 2026

felt: close cleanup-rhostats-jobscripts (D1 stale premise, D2 shipped as #737)

cailmdaley
and
claude
committed
Build OpenMPI 5.0.x in the image; SLURM-ify candide job scripts

cailmdaley
and
claude
committed
ci: publish images on every branch push, not just integration branches

cailmdaley
and
claude
committed
felt: scrub personal wayfinding from public shapepipe store

cailmdaley
and
claude
committed
Fix MPI path: thread module_config_sec through to WorkerHandler.worker

cailmdaley
and
claude
committed
chore: gitignore felt index WAL sidecars (index.db-shm/-wal)

cailmdaley
and
claude
committed
felt: correct mpi-hybrid close (two-layer bug); add exec-modes-schedulers

cailmdaley
and
claude
committed
Merge remote-tracking branch 'origin/develop' into cleanup/candide-scripts-container
cailmdaley
committed
Fix stale module names in example/pbs/config_mpi.ini

cailmdaley
and
claude
committed
felt: mpi-hybrid — record Layer 3 (stale config) + final e2e verification

cailmdaley
and
claude
committed
felt: temper MPI claims to observed-vs-inferred (canfar run history unknown)

cailmdaley
and
claude
committed
felt: record SMP==MPI same-worker finding; sharpen MPI question to Martin

cailmdaley
and
claude
committed
Propagate shapepipe_run's exit code (main must return run's value)

cailmdaley
and
claude
committed
Add MPI world-size preflight check: fail loudly on "rank 0 of N singletons"

cailmdaley
and
claude
committed
felt: record Layer 4 hardening (singleton guard + exit-code fix)

cailmdaley
and
claude
committed
Remove MPI singleton preflight guard from this PR (defer pending decision)

cailmdaley
and
claude
committed
felt: park the singleton guard as follow-up; sharpen Martin question

cailmdaley
and
claude
committed
docs(README): add verified candide container quickstart

cailmdaley
and
claude
committed
docs: make README a general front door; move candide detail to container.md

cailmdaley
and
claude
committed
Move user-facing docs to the docs-rework PR (#739)

cailmdaley
and
claude
committed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Fix MPI on candide (OpenMPI 5 image + latent code bug); containerize & SLURM-ify candide scripts#737

Fix MPI on candide (OpenMPI 5 image + latent code bug); containerize & SLURM-ify candide scripts#737
cailmdaley wants to merge 21 commits into
developfrom
cleanup/candide-scripts-container

Commits on May 30, 2026

Run candide PBS job scripts through the container instead of conda

Commits on May 31, 2026

felt: close cleanup-rhostats-jobscripts (D1 stale premise, D2 shipped as #737)

Build OpenMPI 5.0.x in the image; SLURM-ify candide job scripts

ci: publish images on every branch push, not just integration branches

felt: scrub personal wayfinding from public shapepipe store

Fix MPI path: thread module_config_sec through to WorkerHandler.worker

chore: gitignore felt index WAL sidecars (index.db-shm/-wal)

felt: correct mpi-hybrid close (two-layer bug); add exec-modes-schedulers

Merge remote-tracking branch 'origin/develop' into cleanup/candide-scripts-container

Fix stale module names in example/pbs/config_mpi.ini

felt: mpi-hybrid — record Layer 3 (stale config) + final e2e verification

felt: temper MPI claims to observed-vs-inferred (canfar run history unknown)

felt: record SMP==MPI same-worker finding; sharpen MPI question to Martin

Propagate shapepipe_run's exit code (main must return run's value)

Add MPI world-size preflight check: fail loudly on "rank 0 of N singletons"

felt: record Layer 4 hardening (singleton guard + exit-code fix)

Remove MPI singleton preflight guard from this PR (defer pending decision)

felt: park the singleton guard as follow-up; sharpen Martin question

docs(README): add verified candide container quickstart

docs: make README a general front door; move candide detail to container.md

Move user-facing docs to the docs-rework PR (#739)