Compare commits

...
Author SHA1 Message Date
khluu 8798dacaab Add tasks/version/beam_search/logging_utils to entrypoints deps
Address claude-bot review on #42219. Four additional paths are
direct first-order imports of `vllm/entrypoints/` code or its
tests, missed in earlier scans:

- `vllm/tasks.py`: imported by 14 entrypoints files including
  `api_server.py`, `llm.py`, and every pooling/scoring factory.
- `vllm/version.py`: imported by `api_server.py`, `openai/api_server.py`,
  `openai/run_batch.py`, `grpc_server.py`, `cli/serve.py`,
  `serve/instrumentator/basic.py`.
- `vllm/beam_search.py`: imported by `llm.py` and
  `openai/engine/serving.py`.
- `vllm/logging_utils/`: imported by `openai/server_utils.py`
  (`create_uvicorn_log_config`).

These are first-order imports, not transitive utilities, so they fit
the principle established earlier in this PR: include if directly
imported by tests or by entrypoint code under test.

Signed-off-by: Kevin H. Luu <kevin@anyscale.com>

Signed-off-by: khluu <khluu000@gmail.com>
2026-05-10 03:39:32 -07:00
khluu a5b814b2c9 Add _aiter_ops/sequence to entrypoints integration deps
Address review on #42219. Two additions:

- `vllm/_aiter_ops.py`: directly imported by
  `tests/entrypoints/openai/chat_completion/test_serving_chat.py`
  (`is_aiter_found_and_supported`). The chat-completion test runs
  in API Server openai - Part 1, but applying it consistently
  across all integration jobs to keep the lists uniform.
- `vllm/sequence.py`: defines `IntermediateTensors` and other core
  types in the engine API surface; an incompatible change can
  reach the integration paths even without a direct test import.

Skipping `vllm/_custom_ops.py` and `vllm/forward_context.py` per
gemini's suggestion: those are kernel-level/internal-plumbing
modules with their own dedicated test coverage (kernel tests,
compile tests). Letting kernel/compile tests catch regressions
there avoids triggering full entrypoints integration runs on
low-level changes.

Signed-off-by: Kevin H. Luu <kevin@anyscale.com>

Signed-off-by: khluu <khluu000@gmail.com>
2026-05-10 03:33:38 -07:00
khluu 636597c23c [CI] Narrow entrypoints.yaml source dependencies
Seven Entrypoints Integration jobs in `.buildkite/test_areas/entrypoints.yaml`
listed `vllm/` as a source dependency, causing them to run on any change
anywhere under `vllm/`. Replace with the inference-stack modules these
LLM/API-server tests actually import.

Affected jobs:
- Entrypoints Integration (LLM)
- Entrypoints Integration (API Server openai - Part 1, 2, 3)
- Entrypoints Integration (API Server 2)
- Entrypoints Integration (Pooling)
- Entrypoints Integration (Responses API)

Excludes paths that have their own dedicated test coverage and are
not directly imported by these tests: `vllm/compilation/`,
`vllm/kernels/`, `vllm/ir/`, `vllm/plugins/`, `vllm/triton_utils/`,
`vllm/forward_context.py`, `vllm/sequence.py`, `vllm/_aiter_ops.py`,
`vllm/_custom_ops.py`, `vllm/benchmarks/`, `vllm/profiler/`,
`vllm/tracing/`, `vllm/usage/`.

The `Entrypoints Unit Tests` job already had a narrow
`vllm/entrypoints` dep and is unchanged. Likewise `OpenAI API
Correctness` already had a narrow scope and is unchanged.

Signed-off-by: Kevin H. Luu <kevin@anyscale.com>

Signed-off-by: khluu <khluu000@gmail.com>
2026-05-10 03:26:21 -07:00
+217 -7
View File
@@ -18,7 +18,37 @@ steps:
timeout_in_minutes: 40
working_dir: "/vllm-workspace/tests"
source_file_dependencies:
- vllm/
- vllm/_aiter_ops.py
- vllm/assets/
- vllm/beam_search.py
- vllm/config/
- vllm/distributed/
- vllm/engine/
- vllm/entrypoints/
- vllm/envs.py
- vllm/exceptions.py
- vllm/inputs/
- vllm/logger.py
- vllm/logging_utils/
- vllm/logprobs.py
- vllm/lora/
- vllm/model_executor/
- vllm/multimodal/
- vllm/outputs.py
- vllm/parser/
- vllm/platforms/
- vllm/pooling_params.py
- vllm/reasoning/
- vllm/renderers/
- vllm/sampling_params.py
- vllm/sequence.py
- vllm/tasks.py
- vllm/tokenizers/
- vllm/tool_parsers/
- vllm/transformers_utils/
- vllm/utils/
- vllm/v1/
- vllm/version.py
- tests/entrypoints/llm
- tests/entrypoints/offline_mode
commands:
@@ -37,7 +67,37 @@ steps:
timeout_in_minutes: 50
working_dir: "/vllm-workspace/tests"
source_file_dependencies:
- vllm/
- vllm/_aiter_ops.py
- vllm/assets/
- vllm/beam_search.py
- vllm/config/
- vllm/distributed/
- vllm/engine/
- vllm/entrypoints/
- vllm/envs.py
- vllm/exceptions.py
- vllm/inputs/
- vllm/logger.py
- vllm/logging_utils/
- vllm/logprobs.py
- vllm/lora/
- vllm/model_executor/
- vllm/multimodal/
- vllm/outputs.py
- vllm/parser/
- vllm/platforms/
- vllm/pooling_params.py
- vllm/reasoning/
- vllm/renderers/
- vllm/sampling_params.py
- vllm/sequence.py
- vllm/tasks.py
- vllm/tokenizers/
- vllm/tool_parsers/
- vllm/transformers_utils/
- vllm/utils/
- vllm/v1/
- vllm/version.py
- tests/entrypoints/openai
- tests/entrypoints/test_chat_utils
commands:
@@ -50,7 +110,37 @@ steps:
timeout_in_minutes: 50
working_dir: "/vllm-workspace/tests"
source_file_dependencies:
- vllm/
- vllm/_aiter_ops.py
- vllm/assets/
- vllm/beam_search.py
- vllm/config/
- vllm/distributed/
- vllm/engine/
- vllm/entrypoints/
- vllm/envs.py
- vllm/exceptions.py
- vllm/inputs/
- vllm/logger.py
- vllm/logging_utils/
- vllm/logprobs.py
- vllm/lora/
- vllm/model_executor/
- vllm/multimodal/
- vllm/outputs.py
- vllm/parser/
- vllm/platforms/
- vllm/pooling_params.py
- vllm/reasoning/
- vllm/renderers/
- vllm/sampling_params.py
- vllm/sequence.py
- vllm/tasks.py
- vllm/tokenizers/
- vllm/tool_parsers/
- vllm/transformers_utils/
- vllm/utils/
- vllm/v1/
- vllm/version.py
- tests/entrypoints/openai
- tests/entrypoints/test_chat_utils
commands:
@@ -64,7 +154,37 @@ steps:
device: h200_18gb
working_dir: "/vllm-workspace/tests"
source_file_dependencies:
- vllm/
- vllm/_aiter_ops.py
- vllm/assets/
- vllm/beam_search.py
- vllm/config/
- vllm/distributed/
- vllm/engine/
- vllm/entrypoints/
- vllm/envs.py
- vllm/exceptions.py
- vllm/inputs/
- vllm/logger.py
- vllm/logging_utils/
- vllm/logprobs.py
- vllm/lora/
- vllm/model_executor/
- vllm/multimodal/
- vllm/outputs.py
- vllm/parser/
- vllm/platforms/
- vllm/pooling_params.py
- vllm/reasoning/
- vllm/renderers/
- vllm/sampling_params.py
- vllm/sequence.py
- vllm/tasks.py
- vllm/tokenizers/
- vllm/tool_parsers/
- vllm/transformers_utils/
- vllm/utils/
- vllm/v1/
- vllm/version.py
- tests/entrypoints/openai
- tests/entrypoints/test_chat_utils
commands:
@@ -76,7 +196,37 @@ steps:
timeout_in_minutes: 130
working_dir: "/vllm-workspace/tests"
source_file_dependencies:
- vllm/
- vllm/_aiter_ops.py
- vllm/assets/
- vllm/beam_search.py
- vllm/config/
- vllm/distributed/
- vllm/engine/
- vllm/entrypoints/
- vllm/envs.py
- vllm/exceptions.py
- vllm/inputs/
- vllm/logger.py
- vllm/logging_utils/
- vllm/logprobs.py
- vllm/lora/
- vllm/model_executor/
- vllm/multimodal/
- vllm/outputs.py
- vllm/parser/
- vllm/platforms/
- vllm/pooling_params.py
- vllm/reasoning/
- vllm/renderers/
- vllm/sampling_params.py
- vllm/sequence.py
- vllm/tasks.py
- vllm/tokenizers/
- vllm/tool_parsers/
- vllm/transformers_utils/
- vllm/utils/
- vllm/v1/
- vllm/version.py
- tests/entrypoints/rpc
- tests/entrypoints/serve/instrumentator
- tests/tool_use
@@ -91,7 +241,37 @@ steps:
timeout_in_minutes: 50
working_dir: "/vllm-workspace/tests"
source_file_dependencies:
- vllm/
- vllm/_aiter_ops.py
- vllm/assets/
- vllm/beam_search.py
- vllm/config/
- vllm/distributed/
- vllm/engine/
- vllm/entrypoints/
- vllm/envs.py
- vllm/exceptions.py
- vllm/inputs/
- vllm/logger.py
- vllm/logging_utils/
- vllm/logprobs.py
- vllm/lora/
- vllm/model_executor/
- vllm/multimodal/
- vllm/outputs.py
- vllm/parser/
- vllm/platforms/
- vllm/pooling_params.py
- vllm/reasoning/
- vllm/renderers/
- vllm/sampling_params.py
- vllm/sequence.py
- vllm/tasks.py
- vllm/tokenizers/
- vllm/tool_parsers/
- vllm/transformers_utils/
- vllm/utils/
- vllm/v1/
- vllm/version.py
- tests/entrypoints/pooling
commands:
- export VLLM_WORKER_MULTIPROC_METHOD=spawn
@@ -102,7 +282,37 @@ steps:
timeout_in_minutes: 50
working_dir: "/vllm-workspace/tests"
source_file_dependencies:
- vllm/
- vllm/_aiter_ops.py
- vllm/assets/
- vllm/beam_search.py
- vllm/config/
- vllm/distributed/
- vllm/engine/
- vllm/entrypoints/
- vllm/envs.py
- vllm/exceptions.py
- vllm/inputs/
- vllm/logger.py
- vllm/logging_utils/
- vllm/logprobs.py
- vllm/lora/
- vllm/model_executor/
- vllm/multimodal/
- vllm/outputs.py
- vllm/parser/
- vllm/platforms/
- vllm/pooling_params.py
- vllm/reasoning/
- vllm/renderers/
- vllm/sampling_params.py
- vllm/sequence.py
- vllm/tasks.py
- vllm/tokenizers/
- vllm/tool_parsers/
- vllm/transformers_utils/
- vllm/utils/
- vllm/v1/
- vllm/version.py
- tests/entrypoints/openai/responses
commands:
- pytest -v -s entrypoints/openai/responses