Install procps so Nextflow's task wrapper can run

The pipeline still failed after the system-prefix fix, with exit 1, empty
stdout and no traceback. The image was correct -- exit 1 (not 127/126) proves
the adapter script was found and its shebang resolved, so bin/ injection and
the interpreter were both working.

The cause was the one real line in the task log:

    Command 'ps' required by nextflow to collect task metrics cannot be found

Nextflow's task wrapper shells out to `ps`, so without procps the task dies
before the process script ever runs. With PYTHONUNBUFFERED=1 set, a payload
crash would have printed something; nothing did, because Python never started.

nf-core requires procps in every container for exactly this reason.
python:3.11-slim omits it, and the pytorch base image this Dockerfile replaced
happened to include it, so the loss went unnoticed at the base-image swap.

Covered by a build-time check and a CI smoke test so it cannot regress.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

Author: logan-blaine

.github/workflows/docker.yaml

@@ -82,6 +82,9 @@ jobs:
           # Default CMD is ["perturbo", "--help"].
           docker run --rm perturbo:ci
           docker run --rm perturbo:ci perturbo --help
+          # Nextflow's task wrapper shells out to `ps`; without it the task dies
+          # before the process script runs, which looks like a silent payload crash.
+          docker run --rm perturbo:ci ps -o pid,comm
           # How the pipeline's bin/*.py scripts (`#!/usr/bin/env python`) enter
           # Python. A venv symlink or wrapper at /usr/local/bin/python breaks
           # exactly this while leaving `perturbo` itself working.

Dockerfile

@@ -10,8 +10,15 @@ ENV PIP_DISABLE_PIP_VERSION_CHECK=1 \
     PYTHONUNBUFFERED=1 \
     XLA_PYTHON_CLIENT_PREALLOCATE=false
 
+# procps supplies `ps`, which Nextflow's task wrapper shells out to in order to
+# collect task metrics. Without it the wrapper reports "Command 'ps' required by
+# nextflow to collect task metrics cannot be found" and the task dies BEFORE the
+# process script runs -- exit 1, empty stdout, no traceback, which reads exactly
+# like a silent crash in the payload. nf-core requires procps in every container
+# for this reason. python:3.11-slim omits it; the pytorch base image this
+# Dockerfile replaced happened to include it, so the loss went unnoticed.
 RUN apt-get update \
-    && apt-get install -y --no-install-recommends ca-certificates \
+    && apt-get install -y --no-install-recommends ca-certificates procps \
     && rm -rf /var/lib/apt/lists/*
 
 # uv lives OUTSIDE /usr/local so the sync below (which prunes its target
@@ -51,6 +58,7 @@ RUN uv sync --locked --no-dev --no-editable --extra cuda
 # actually enter Python, and it is what the venv layouts kept breaking.
 RUN set -eu \
     && test ! -e /app/.venv || { echo "FATAL: a venv exists; deps must be in the system prefix"; exit 1; } \
+    && command -v ps >/dev/null || { echo "FATAL: no ps; Nextflow's task wrapper needs it"; exit 1; } \
     && python  -c 'import sys, perturbo; print("python  ->", sys.executable, "| perturbo", getattr(perturbo, "__version__", "?"))' \
     && python3 -c 'import perturbo' \
     && /usr/bin/env python  -c 'import perturbo' \