[ROCm][CI] Stabilize Cargo cache and pre-test image checks (#43815)

Signed-off-by: Andreas Karatzas <akaratza@amd.com>

.buildkite/hardware_tests/amd.yaml

@@ -17,6 +17,26 @@ steps:
       --target test
       --no-cache
       --progress plain .
+    - |
+      docker run --rm --network=none --entrypoint /bin/bash "rocm/vllm-ci:${BUILDKITE_COMMIT}" -ec '
+        if [ ! -d /vllm-workspace ]; then echo Missing directory: /vllm-workspace >&2; exit 1; fi
+        if [ ! -d /vllm-workspace/tests ]; then echo Missing directory: /vllm-workspace/tests >&2; exit 1; fi
+        if [ ! -d /vllm-workspace/src/vllm ]; then echo Missing directory: /vllm-workspace/src/vllm >&2; exit 1; fi
+        if [ ! -x /vllm-workspace/src/vllm/vllm-rs ]; then echo Missing executable: /vllm-workspace/src/vllm/vllm-rs >&2; exit 1; fi
+        command -v python3
+        command -v uv
+        command -v pytest
+        if ! command -v amd-smi >/dev/null 2>&1 && ! command -v rocminfo >/dev/null 2>&1; then
+          echo No ROCm CLI found in image >&2
+          exit 1
+        fi
+        python3 - <<PY
+      import torch, vllm
+      print(torch.__version__)
+      print(vllm.__version__)
+      PY
+        echo AMD image smoke OK
+      '
     - docker push "rocm/vllm-ci:${BUILDKITE_COMMIT}"
     env:
       DOCKER_BUILDKIT: "1"

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

@@ -35,25 +35,9 @@ export PYTHONPATH=".."
 # Helper Functions
 ###############################################################################
 
-cleanup_docker() {
-  # Get Docker's root directory
-  docker_root=$(docker info -f '{{.DockerRootDir}}')
-  if [ -z "$docker_root" ]; then
-    echo "Failed to determine Docker root directory."
-    exit 1
-  fi
-  echo "Docker root directory: $docker_root"
-
-  disk_usage=$(df "$docker_root" | tail -1 | awk '{print $5}' | sed 's/%//')
-  threshold=70
-  if [ "$disk_usage" -gt "$threshold" ]; then
-    echo "Disk usage is above $threshold%. Cleaning up Docker images and volumes..."
-    docker image prune -f
-    docker volume prune -f && docker system prune --force --filter "until=72h" --all
-    echo "Docker images and volumes cleanup completed."
-  else
-    echo "Disk usage is below $threshold%. No cleanup needed."
-  fi
+report_docker_usage() {
+  echo "--- Docker usage"
+  docker system df || true
 }
 
 cleanup_network() {
@@ -254,8 +238,8 @@ re_quote_pytest_markers() {
 echo "--- ROCm info"
 rocminfo
 
-# --- Docker housekeeping ---
-cleanup_docker
+# --- Docker status ---
+report_docker_usage
 
 # --- Pull test image ---
 echo "--- Pulling container"
@@ -264,9 +248,17 @@ container_name="rocm_${BUILDKITE_COMMIT}_$(tr -dc A-Za-z0-9 < /dev/urandom | hea
 docker pull "${image_name}"
 
 remove_docker_container() {
-  docker rm -f "${container_name}" || docker image rm -f "${image_name}" || true
+  # docker run uses --rm, so the container is normally already gone when the
+  # EXIT trap runs. Cleanup is best-effort and must not affect the test result.
+  docker rm -f "${container_name}" >/dev/null 2>&1 || true
 }
-trap remove_docker_container EXIT
+
+on_exit() {
+  local exit_code=$?
+  remove_docker_container
+  exit "$exit_code"
+}
+trap on_exit EXIT
 
 # --- Prepare commands ---
 echo "--- Running container"

docker/Dockerfile.rocm

@@ -135,13 +135,18 @@ ENV PATH="/root/.cargo/bin:${PATH}"
 # Cap cargo parallelism to avoid exhausting the AMD CI host's open-file limit
 # (rustc spawns enough concurrent processes to hit RLIMIT_NOFILE otherwise).
 ENV CARGO_BUILD_JOBS=4
+ENV CARGO_NET_RETRY=10
+ENV RUSTUP_MAX_RETRIES=10
 
-# Build the release binary. Cache cargo registry/git, and copy the binary out
-# so it persists into the image layer for later COPY --from=rust-build.
-RUN --mount=type=cache,target=/root/.cargo/registry \
-    --mount=type=cache,target=/root/.cargo/git \
+# Build the release binary. Cargo's registry/git caches can be written by
+# concurrent BuildKit jobs on shared workers, so lock those cache mounts while
+# keeping the cache benefit. Copy the binary out so it persists into the image
+# layer for later COPY --from=rust-build.
+RUN --mount=type=cache,id=vllm-rocm-cargo-registry,target=/root/.cargo/registry,sharing=locked \
+    --mount=type=cache,id=vllm-rocm-cargo-git,target=/root/.cargo/git,sharing=locked \
     cd ${COMMON_WORKDIR}/vllm \
-    && VLLM_RS_TARGET_PATH=/tmp/vllm-rs bash build_rust.sh
+    && VLLM_RS_TARGET_PATH=/tmp/vllm-rs bash build_rust.sh \
+    && test -x /tmp/vllm-rs
 
 # -----------------------
 # vLLM build stages