fix(python-backend): make JIT subprocesses work on hosts of any size (#9679)

Two related runtime fixes for Python backends that JIT-compile CUDA
kernels at first model load (FlashInfer, PyTorch inductor, triton):

1. libbackend.sh: replace `source ${EDIR}/venv/bin/activate` with a
   minimal manual setup (_activateVenv: export VIRTUAL_ENV, prepend
   PATH, unset PYTHONHOME) computed from $EDIR at runtime. `uv venv`
   and `python -m venv` both bake the create-time absolute path into
   bin/activate (e.g. VIRTUAL_ENV='/vllm/venv' from the Docker build
   stage), so sourcing activate on a relocated venv — copied out of
   the build container and unpacked at an arbitrary backend dir —
   prepends a stale, non-existent path to $PATH. Pip-installed CLI
   tools (e.g. ninja, used by FlashInfer's NVFP4 GEMM JIT) are then
   never found and the load aborts with FileNotFoundError. Doing the
   env setup ourselves matches what `uv run` does internally and
   sidesteps the relocation problem entirely. Generic — every Python
   backend benefits.

2. vllm/run.sh: replace ninja's default -j$(nproc)+2 with an adaptive
   MAX_JOBS = min(nproc, (MemAvailable-4)/4). Each concurrent
   nvcc/cudafe++ peaks at multiple GiB; the default OOM-kills on
   memory-tight hosts (e.g. a 16 GiB desktop loading a 27B NVFP4
   model) but underutilises 100-core / 1 TB boxes. User-set MAX_JOBS
   still wins. Also pin NVCC_THREADS=2 unless overridden.

Refs: vllm-project/vllm#20079

Assisted-by: Claude:claude-opus-4-7 [Edit] [Bash]

Author: richiejp

backend/python/common/libbackend.sh

@@ -318,6 +318,21 @@ _makeVenvPortable() {
 }
 
 
+# Apply the venv to the current process: VIRTUAL_ENV, PATH, PYTHONHOME hygiene.
+# Equivalent to the runtime portion of `source bin/activate`, but computed from
+# $EDIR (resolved at runtime via realpath) instead of the path baked into
+# bin/activate at venv-create time. `uv venv` (and `python -m venv`) both bake
+# the create-time absolute path in, so sourcing activate on a relocated venv —
+# e.g. one built at /vllm/venv inside a Docker stage and unpacked under
+# /backends/cuda13-vllm-development/venv at runtime — silently prepends a
+# stale, non-existent path to $PATH. Doing the setup ourselves sidesteps that;
+# this is the same approach `uv run` takes internally.
+_activateVenv() {
+    export VIRTUAL_ENV="${EDIR}/venv"
+    export PATH="${EDIR}/venv/bin:${PATH}"
+    unset PYTHONHOME
+}
+
 # ensureVenv makes sure that the venv for the backend both exists, and is activated.
 #
 # This function is idempotent, so you can call it as many times as you want and it will
@@ -354,7 +369,7 @@ function ensureVenv() {
                 venv_args="--copies"
             fi
             "${interpreter}" -m venv ${venv_args} "${EDIR}/venv"
-            source "${EDIR}/venv/bin/activate"
+            _activateVenv
             "${interpreter}" -m pip install --upgrade pip
         else
             if [ "x${PORTABLE_PYTHON}" == "xtrue" ]; then
@@ -375,7 +390,7 @@ function ensureVenv() {
     fi
 
     if [ "x${VIRTUAL_ENV:-}" != "x${EDIR}/venv" ]; then
-        source "${EDIR}/venv/bin/activate"
+        _activateVenv
     fi
 }
 

backend/python/vllm/run.sh

@@ -3,6 +3,30 @@ set -x
 
 backend_dir=$(dirname $0)
 
+# FlashInfer / PyTorch JIT-compile CUDA kernels at first model load (e.g.
+# the NVFP4 GEMM kernel for Blackwell SM120). Each concurrent nvcc /
+# cudafe++ peaks at multiple GiB during compilation; ninja's default
+# (-j$(nproc)+2) OOM-kills on memory-tight hosts but underutilises
+# 100-core / 1 TB boxes. Default MAX_JOBS to the smaller of the CPU count
+# and an available-memory budget at ~4 GiB per job. User-set MAX_JOBS in
+# the environment wins.
+# https://github.com/vllm-project/vllm/issues/20079
+if [ -z "${MAX_JOBS:-}" ]; then
+    _ncpus=$(nproc 2>/dev/null || echo 1)
+    _mem_avail_kb=$(awk '/^MemAvailable:/ {print $2; exit}' /proc/meminfo 2>/dev/null || echo 0)
+    _mem_avail_gb=$(( _mem_avail_kb / 1024 / 1024 ))
+    # Reserve ~4 GiB for the rest of the system; budget ~4 GiB per job.
+    if [ "${_mem_avail_gb}" -gt 8 ]; then
+        _mem_jobs=$(( (_mem_avail_gb - 4) / 4 ))
+    else
+        _mem_jobs=1
+    fi
+    [ "${_mem_jobs}" -lt 1 ] && _mem_jobs=1
+    [ "${_mem_jobs}" -gt "${_ncpus}" ] && _mem_jobs=${_ncpus}
+    export MAX_JOBS="${_mem_jobs}"
+fi
+export NVCC_THREADS="${NVCC_THREADS:-2}"
+
 if [ -d $backend_dir/common ]; then
     source $backend_dir/common/libbackend.sh
 else