A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 3, 03:00 UTC
[XPU][CI]Fix Docker image pull-to-run race in Intel GPU CI (#43266)
Signed-off-by: zengxian <xiangdong.zeng@intel.com> Co-authored-by: Kunshang Ji <kunshang.ji@intel.com>
xiangdong committed
May 21, 2026 at 18:41 UTC
5ecd8e9c708821916323d25d5f7beddb7f41d22b
1 file changed
+22
-8
.buildkite/scripts/hardware_ci/run-intel-test.sh
+22
-8
@@ -352,17 +352,31 @@ if [[ -z "${ZE_AFFINITY_MASK:-}" ]]; then
352
echo "Warning: ZE_AFFINITY_MASK is not set. Proceeding without device affinity." >&2
353
fi
354
355
-docker run \
355
+export CMDS="${commands}"
356
+export HF_TOKEN ZE_AFFINITY_MASK
357
+
358
+{
359
+ flock 9
360
+ if ! docker image inspect "${IMAGE}" >/dev/null 2>&1; then
361
+ echo 'Image missing before container creation, pulling again...'
362
+ timeout 900 docker pull "${IMAGE}"
363
+ fi
364
+
365
+ docker create \
366
--device /dev/dri:/dev/dri \
367
--net=host \
368
--ipc=host \
369
--privileged \
370
-v /dev/dri/by-path:/dev/dri/by-path \
361
- -v ${HOME}/.cache/huggingface:/root/.cache/huggingface \
362
- --entrypoint="" \
363
- -e "HF_TOKEN=${HF_TOKEN:-}" \
364
- -e "ZE_AFFINITY_MASK=${ZE_AFFINITY_MASK:-}" \
365
- -e "CMDS=${commands}" \
371
+ -v "${HOME}/.cache/huggingface:/root/.cache/huggingface" \
372
+ --entrypoint='' \
373
+ -e HF_TOKEN \
374
+ -e ZE_AFFINITY_MASK \
375
+ -e CMDS \
376
--name "${container_name}" \
367
- "${image_name}" \
368
- bash -c 'set -e; echo "ZE_AFFINITY_MASK is ${ZE_AFFINITY_MASK:-}"; eval "$CMDS"'
377
+ "${IMAGE}" \
378
+ bash -c 'set -e; echo "ZE_AFFINITY_MASK is ${ZE_AFFINITY_MASK:-}"; eval "$CMDS"' \
379
+ >/dev/null
380
+} 9>/tmp/docker-pull.lock
381
+
382
+docker start -a "${container_name}"