[CI][XPU] Work around intermittent segfault in Intel XPU CI with VLLM_DISABLE_COMPILE_CACHE=1 (#51337)
Signed-off-by: Chaojun Zhang <chaojun.zhang@intel.com>
Chaojun Zhang committed
Aug 7, 2026 at 13:18 UTC
b706fd1628b06c216a945176a9fedfa808324803
5 files changed
+10
-5
.buildkite/intel_jobs/basic_correctness.yaml
+2
-1
@@ -18,10 +18,11 @@ steps:
18
source_file_dependencies:
19
- vllm/
20
- tests/basic_correctness/test_cumem.py
21
+ # Remove VLLM_DISABLE_COMPILE_CACHE=1 once intel/intel-xpu-backend-for-triton#7682 is fixed
22
commands:
23
- >-
24
bash .buildkite/scripts/hardware_ci/run-intel-test.sh
25
'cd tests &&
26
export VLLM_WORKER_MULTIPROC_METHOD=spawn &&
26
- pytest -v -s basic_correctness/test_cpu_offload.py &&
27
+ VLLM_DISABLE_COMPILE_CACHE=1 pytest -v -s basic_correctness/test_cpu_offload.py &&
28
pytest -v -s basic_correctness/test_mem.py::test_end_to_end'
.buildkite/intel_jobs/model_executor_intel.yaml
+2
-1
@@ -21,6 +21,7 @@ steps:
21
- vllm/config/model.py
22
- vllm/model_executor
23
- tests/model_executor
24
+ # Remove VLLM_DISABLE_COMPILE_CACHE=1 once intel/intel-xpu-backend-for-triton#7682 is fixed
25
commands:
26
- >-
27
bash .buildkite/scripts/hardware_ci/run-intel-test.sh
@@ -30,4 +31,4 @@ steps:
31
export VLLM_WORKER_MULTIPROC_METHOD=spawn &&
32
export PYTHONFAULTHANDLER=1 &&
33
cd tests &&
33
- pytest -v -s model_executor -m "not slow_test" --ignore="model_executor/layers/test_rocm_unquantized_gemm.py" --deselect="tests/model_executor/model_loader/test_reload.py::test_kv_scale_reload"'
34
+ VLLM_DISABLE_COMPILE_CACHE=1 pytest -v -s model_executor -m "not slow_test" --ignore="model_executor/layers/test_rocm_unquantized_gemm.py" --deselect="tests/model_executor/model_loader/test_reload.py::test_kv_scale_reload"'
.buildkite/intel_jobs/model_runner_v2_intel.yaml
+2
-1
@@ -22,13 +22,14 @@ steps:
22
- vllm/v1/attention/
23
- tests/v1/engine/test_llm_engine.py
24
- tests/v1/e2e/
25
+ # Remove VLLM_DISABLE_COMPILE_CACHE=1 once intel/intel-xpu-backend-for-triton#7682 is fixed
26
commands:
27
- >-
28
bash .buildkite/scripts/hardware_ci/run-intel-test.sh
29
'export VLLM_USE_V2_MODEL_RUNNER=1 &&
30
cd tests &&
31
pytest -v -s v1/engine/test_llm_engine.py -k "not test_engine_metrics" &&
31
- pytest -v -s v1/e2e/general/test_context_length.py &&
32
+ VLLM_DISABLE_COMPILE_CACHE=1 pytest -v -s v1/e2e/general/test_context_length.py &&
33
ENFORCE_EAGER=1 pytest -v -s v1/e2e/general/test_async_scheduling.py -k "not ngram" &&
34
pytest -v -s entrypoints/llm/test_struct_output_generate.py -k "xgrammar and not speculative_config6 and not speculative_config7 and not speculative_config8 and not speculative_config0" &&
35
pytest -v -s v1/e2e/general/test_min_tokens.py'
.buildkite/intel_jobs/models_distributed_intel.yaml
+2
-1
@@ -20,8 +20,9 @@ steps:
20
- vllm/model_executor/model_loader/sharded_state_loader.py
21
- vllm/model_executor/models/
22
- tests/model_executor/model_loader/test_sharded_state_loader.py
23
+ # Remove VLLM_DISABLE_COMPILE_CACHE=1 once intel/intel-xpu-backend-for-triton#7682 is fixed
24
commands:
25
- >-
26
bash .buildkite/scripts/hardware_ci/run-intel-test.sh
27
'cd tests &&
27
- pytest -v -s model_executor/model_loader/test_sharded_state_loader.py -m "not slow_test"'
28
+ VLLM_DISABLE_COMPILE_CACHE=1 pytest -v -s model_executor/model_loader/test_sharded_state_loader.py -m "not slow_test"'
.buildkite/intel_jobs/samplers_intel.yaml
+2
-1
@@ -22,8 +22,9 @@ steps:
22
- tests/samplers
23
- tests/conftest.py
24
- vllm/entrypoints/generate/beam_search
25
+ # Remove VLLM_DISABLE_COMPILE_CACHE=1 once intel/intel-xpu-backend-for-triton#7682 is fixed
26
commands:
27
- >-
28
bash .buildkite/scripts/hardware_ci/run-intel-test.sh
29
'cd tests &&
29
- VLLM_XPU_USE_SAMPLER_KERNEL=1 pytest -v -s samplers'
30
+ VLLM_XPU_USE_SAMPLER_KERNEL=1 VLLM_DISABLE_COMPILE_CACHE=1 pytest -v -s samplers'