[ROCm][CI] pass merge-base to container for python-only wheel metadata (#45869)

Signed-off-by: Divakar Verma <divakar.verma@amd.com>
Co-authored-by: Andreas Karatzas <akaratza@amd.com>

.buildkite/scripts/hardware_ci/run-amd-test.sh

@@ -367,6 +367,20 @@ remove_docker_container() {
 }
 trap remove_docker_container EXIT
 
+# python_only_compile.sh runs `python setup.py develop` and needs the full repo tree
+# under /vllm-workspace (Dockerfile.rocm test stage: mkdir src && mv vllm).
+# The ROCm wheel artifact tarball only ships a thin tree (tests, etc.), so
+# artifact images cannot satisfy that test — use the full rocm/vllm-ci image.
+_cmd_probe="${VLLM_TEST_COMMANDS:-}"
+if [[ -z "${_cmd_probe}" ]]; then
+  _cmd_probe="$*"
+fi
+if [[ "${VLLM_CI_USE_ARTIFACTS:-0}" == "1" && "${_cmd_probe}" == *python_only_compile.sh* ]]; then
+  echo "INFO: disabling VLLM_CI_USE_ARTIFACTS for python_only_compile (requires full /vllm-workspace tree)"
+  export VLLM_CI_USE_ARTIFACTS=0
+fi
+unset -v _cmd_probe
+
 if ! prepare_artifact_image; then
   echo "Using full ROCm CI image: ${image_name}"
   docker pull "${image_name}" || exit 1
@@ -426,6 +440,24 @@ fi
 
 echo "Final commands: $commands"
 
+# The ROCm test image often ships /vllm-workspace without .git (artifact tarball unpack).
+# tests/standalone_tests/python_only_compile.sh uses merge-base(HEAD, origin/main) for
+# wheels.vllm.ai; compute on the agent (full git checkout) and pass into the container.
+vllm_standalone_merge_base=""
+checkout="${BUILDKITE_BUILD_CHECKOUT_PATH:-}"
+if [[ -z "${checkout}" || ! -d "${checkout}" ]]; then
+  checkout="."
+fi
+if git -C "${checkout}" rev-parse --is-inside-work-tree >/dev/null 2>&1; then
+  vllm_standalone_merge_base="$(
+    git -C "${checkout}" merge-base HEAD origin/main 2>/dev/null || true
+  )"
+fi
+if [[ -z "${vllm_standalone_merge_base}" ]]; then
+  vllm_standalone_merge_base="${BUILDKITE_COMMIT:-}"
+fi
+echo "INFO: passing VLLM_STANDALONE_MERGE_BASE into container: ${vllm_standalone_merge_base}"
+
 MYPYTHONPATH="/vllm-workspace"
 
 container_job_id="${BUILDKITE_JOB_ID:-${BUILDKITE_PARALLEL_JOB:-0}}"
@@ -525,6 +557,7 @@ else
     -e "VLLM_CACHE_ROOT=${CONTAINER_CACHE_ROOT}/vllm" \
     -e "XDG_CACHE_HOME=${CONTAINER_CACHE_ROOT}/xdg" \
     -e "PYTORCH_ROCM_ARCH=" \
+    -e "VLLM_STANDALONE_MERGE_BASE=${vllm_standalone_merge_base}" \
     --name "${container_name}" \
     "${image_name}" \
     /bin/bash -c "${CONTAINER_PREFLIGHT} && ${commands}"

.buildkite/test_areas/misc.yaml

@@ -262,6 +262,16 @@ steps:
   - setup.py
   commands:
   - bash standalone_tests/python_only_compile.sh
+  mirror:
+    amd:
+      device: mi325_1
+      timeout_in_minutes: 20
+      depends_on:
+      - image-build-amd
+      source_file_dependencies:
+      - tests/standalone_tests/python_only_compile.sh
+      - setup.py
+      - vllm/platforms/rocm.py
 
 - label: Async Engine, Inputs, Utils, Worker
   device: h200_35gb

tests/standalone_tests/python_only_compile.sh

@@ -4,9 +4,26 @@
 
 set -e
 
-merge_base_commit=$(git merge-base HEAD origin/main)
+# ROCm CI runs this script inside `run-amd-test.sh` where /vllm-workspace often has no .git
+# (wheel artifact layout). The wrapper passes VLLM_STANDALONE_MERGE_BASE from the agent checkout.
+merge_base_commit=""
+if [[ -n "${VLLM_STANDALONE_MERGE_BASE:-}" ]]; then
+    merge_base_commit="${VLLM_STANDALONE_MERGE_BASE}"
+elif merge_base_commit="$(git -C /vllm-workspace merge-base HEAD origin/main 2>/dev/null)"; then
+    :
+elif merge_base_commit="$(git merge-base HEAD origin/main 2>/dev/null)"; then
+    :
+else
+    echo "ERROR: need a git checkout or VLLM_STANDALONE_MERGE_BASE to resolve wheels.vllm.ai commit." >&2
+    exit 1
+fi
+
 echo "INFO: current merge base commit with main: $merge_base_commit"
-git show --oneline -s "$merge_base_commit"
+if git show --oneline -s "$merge_base_commit" 2>/dev/null; then
+    :
+else
+    echo "INFO: git show unavailable in this environment; using SHA above for precompiled metadata."
+fi
 
 # test whether the metadata.json url is valid, retry each 3 minutes up to 5 times
 # this avoids cumbersome error messages & manual retries in case the precompiled wheel
@@ -59,7 +76,12 @@ cd /vllm-workspace/
 # uninstall vllm
 pip3 uninstall -y vllm
 # restore the original files
-mv src/vllm ./vllm
+if [[ -d src/vllm ]]; then
+    mv src/vllm ./vllm
+elif [[ ! -d vllm ]]; then
+    echo "ERROR: expected vllm package at /vllm-workspace/src/vllm or /vllm-workspace/vllm" >&2
+    exit 1
+fi
 
 # remove all compilers
 apt remove --purge build-essential -y
@@ -67,7 +89,14 @@ apt autoremove -y
 
 echo 'import os; os.system("touch /tmp/changed.file")' >> vllm/__init__.py
 
-VLLM_PRECOMPILED_WHEEL_COMMIT=$merge_base_commit VLLM_USE_PRECOMPILED=1 pip3 install -vvv -e .
+# ROCm CI uses setuptools develop for editable installs (see Dockerfile.rocm and run-amd-test.sh).
+_vllm_target_lower="$(printf '%s' "${VLLM_TARGET_DEVICE:-}" | tr '[:upper:]' '[:lower:]')"
+if [[ "${_vllm_target_lower}" == "rocm" ]]; then
+  VLLM_PRECOMPILED_WHEEL_COMMIT=$merge_base_commit VLLM_USE_PRECOMPILED=1 python3 setup.py develop
+else
+  VLLM_PRECOMPILED_WHEEL_COMMIT=$merge_base_commit VLLM_USE_PRECOMPILED=1 pip3 install -vvv -e .
+fi
+unset -v _vllm_target_lower
 # Run the script
 python3 -c 'import vllm'