[None][chore] Optimize perf for the RPC executor and add some profile utilities to llm-api #8415
Conversation
Superjomn
force-pushed
the
perf-tuning
branch
from
931d8b0 to
77d7d74
Compare
📝 WalkthroughWalkthroughThis PR introduces environment variable override support for orchestrator type configuration, refactors executor creation into factory methods for improved modularity, and enables two previously skipped RPC-related tests. The changes consolidate executor instantiation logic and allow orchestrator type to be set via the Changes
Estimated code review effort🎯 3 (Moderate) | ⏱️ ~20 minutes The changes involve mixed concerns across multiple files: understanding the factory method extraction pattern and its integration with existing executor paths, following the environment variable override flow through the validator, and contextualizing why the previously skipped tests are now enabled. While individual changes are straightforward, coherent review requires tracing the orchestrator type flow and validating the factory methods properly replace inline instantiation logic. Pre-merge checks and finishing touches❌ Failed checks (3 warnings)
✨ Finishing touches
🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 0
🧹 Nitpick comments (1)
tensorrt_llm/llmapi/utils.py (1)
359-361: Add a docstring for better documentation.For consistency with similar utility functions in this file (e.g.,
enable_llmapi_debug()at lines 351-356), consider adding a docstring to document the purpose and behavior of this function.Apply this diff to add a docstring:
def orchestrator_type_env() -> Optional[str]: + """Read orchestrator type from the TLLM_ORCHESTRATOR_TYPE environment variable. + + Returns: + Optional[str]: The orchestrator type ('rpc' or 'ray') if set, None otherwise. + """ return os.environ.get("TLLM_ORCHESTRATOR_TYPE", None)
📜 Review details
Configuration used: Path: .coderabbit.yaml
Review profile: CHILL
Plan: Pro
📒 Files selected for processing (5)
tensorrt_llm/executor/executor.py(5 hunks)tensorrt_llm/llmapi/llm_args.py(2 hunks)tensorrt_llm/llmapi/utils.py(1 hunks)tests/unittest/llmapi/test_llm_multi_gpu_pytorch.py(0 hunks)tests/unittest/llmapi/test_llm_pytorch.py(0 hunks)
💤 Files with no reviewable changes (2)
- tests/unittest/llmapi/test_llm_pytorch.py
- tests/unittest/llmapi/test_llm_multi_gpu_pytorch.py
🧰 Additional context used
📓 Path-based instructions (3)
**/*.{h,hpp,hh,hxx,cpp,cxx,cc,cu,cuh,py}
📄 CodeRabbit inference engine (CODING_GUIDELINES.md)
Use only spaces, no tabs; indent with 4 spaces.
Files:
tensorrt_llm/llmapi/utils.pytensorrt_llm/llmapi/llm_args.pytensorrt_llm/executor/executor.py
**/*.py
📄 CodeRabbit inference engine (CODING_GUIDELINES.md)
**/*.py: Python code must target Python 3.8+.
Indent Python code with 4 spaces; do not use tabs.
Maintain module namespace when importing; prefer 'from package.subpackage import foo' then 'foo.SomeClass()' instead of importing the class directly.
Python filenames should be snake_case (e.g., some_file.py).
Python classes use PascalCase names.
Functions and methods use snake_case names.
Local variables use snake_case; prefix 'k' for variables that start with a number (e.g., k_99th_percentile).
Global variables use upper SNAKE_CASE prefixed with 'G' (e.g., G_MY_GLOBAL).
Constants use upper SNAKE_CASE (e.g., MY_CONSTANT).
Avoid shadowing variables from an outer scope.
Initialize all externally visible members of a class in the constructor.
Prefer docstrings for interfaces that may be used outside a file; comments for in-function or file-local interfaces.
Use Google-style docstrings for classes and functions (Sphinx-parsable).
Document attributes and variables inline so they render under the class/function docstring.
Avoid reflection when a simpler, explicit approach suffices (e.g., avoid dict(**locals()) patterns).
In try/except, catch the most specific exceptions possible.
For duck-typing try/except, keep the try body minimal and use else for the main logic.
Files:
tensorrt_llm/llmapi/utils.pytensorrt_llm/llmapi/llm_args.pytensorrt_llm/executor/executor.py
**/*.{cpp,cxx,cc,h,hpp,hh,hxx,cu,cuh,py}
📄 CodeRabbit inference engine (CODING_GUIDELINES.md)
Prepend the NVIDIA Apache-2.0 copyright header with current year to the top of all source files (e.g., .cpp, .h, .cu, .py).
Files:
tensorrt_llm/llmapi/utils.pytensorrt_llm/llmapi/llm_args.pytensorrt_llm/executor/executor.py
🧬 Code graph analysis (2)
tensorrt_llm/llmapi/llm_args.py (1)
tensorrt_llm/llmapi/utils.py (1)
orchestrator_type_env(359-360)
tensorrt_llm/executor/executor.py (7)
tensorrt_llm/llmapi/mpi_session.py (1)
MpiSession(84-129)tensorrt_llm/executor/postproc_worker.py (1)
PostprocWorkerConfig(42-49)tensorrt_llm/llmapi/llm_args.py (1)
KvCacheConnectorConfig(454-466)tensorrt_llm/executor/rpc_proxy.py (1)
GenerationExecutorRpcProxy(24-375)tensorrt_llm/executor/worker.py (1)
GenerationExecutorWorker(41-229)tensorrt_llm/executor/proxy.py (1)
GenerationExecutorProxy(37-454)tensorrt_llm/executor/utils.py (1)
ProcessPoolExecutorSession(80-104)
🪛 Ruff (0.14.0)
tensorrt_llm/llmapi/llm_args.py
2036-2038: Avoid specifying long messages outside the exception class
(TRY003)
⏰ Context from checks skipped due to timeout of 90000ms. You can increase the timeout in your CodeRabbit configuration to a maximum of 15 minutes (900000ms). (1)
- GitHub Check: Pre-commit Check
🔇 Additional comments (9)
tensorrt_llm/llmapi/llm_args.py (2)
24-24: LGTM!The import is correctly placed and necessary for the new validator below.
2029-2041: Missing@classmethoddecorator causes validator to fail.The field validator is missing the
@classmethoddecorator, which is required by Pydantic. Without it, the validator will receive incorrect arguments and fail at runtime.Apply this diff to fix the issue:
@field_validator('orchestrator_type', mode='before') +@classmethod -def validate_orchestrator_config(v): +def validate_orchestrator_config(cls, v): # The environment variable will override the orchestrator_type field. # TODO: remove the environment variable after RPC path is stable, then # there will be only two stable options: None(RPC) and 'ray'. if (ev := orchestrator_type_env()) is not None: if ev not in ['rpc', 'ray']: raise ValueError( f"Invalid orchestrator type: {ev}. Please set orchestrator_type to 'rpc' or 'ray'." ) v = ev return vNote: The static analysis hint about the long error message (TRY003) is a minor style concern and can be addressed separately if desired.
Likely an incorrect or invalid review comment.
tensorrt_llm/executor/executor.py (7)
378-395: LGTM! Clean refactoring into factory method.The
_create_rpc_executorstatic method consolidates RPC executor instantiation logic, improving code maintainability by eliminating duplicate creation code across multiple call sites.
396-426: LGTM! Well-designed factory method with clear branching.The
_create_ipc_executorstatic method effectively consolidates IPC-based executor creation with a clearuse_workerparameter to toggle between single-process Worker and multi-process Proxy modes. The docstring clearly explains the parameter's behavior.
485-487: LGTM! Correctly excludes "rpc" from unsupported orchestrators.The condition now explicitly allows
orchestrator_type == "rpc"to bypass the unsupported orchestrator error, preserving the RPC execution path as intended by this PR.
493-506: LGTM! Clear routing logic with improved readability.The
orchestrator_is_rpcboolean flag improves code clarity by consolidating the orchestrator type check. The conditional routing correctly calls the appropriate factory method based on the orchestrator type.
525-541: LGTM! Correct factory method usage for single-process path.The refactored code correctly routes to RPC executor when needed, and uses
_create_ipc_executorwithuse_worker=Trueto create a single-processGenerationExecutorWorkerfor the TP1 optimization path. The pattern is consistent and correct.
548-564: LGTM! Consistent factory method usage for non-Windows streaming path.The refactored code maintains consistent routing patterns. The
use_worker=Falseparameter correctly creates a multi-processGenerationExecutorProxyfor streaming performance, andmpi_session=Noneallows the executor to use mpi4py internally.
571-578: LGTM! Windows-specific executor creation with known limitation.The refactored code correctly uses
_create_ipc_executorwithuse_worker=FalseandProcessPoolExecutorSessionfor Windows compatibility (since mpi4py cannot be used). Note the TODO comment at line 570 indicating that RPC worker support on Windows is planned future work.
Superjomn
force-pushed
the
perf-tuning
branch
2 times, most recently
from
14303d0 to
3e6d5a5
Compare
Superjomn
force-pushed
the
perf-tuning
branch
4 times, most recently
from
26477c1 to
2bcafb3
Compare
|
/bot run |
Superjomn
changed the title
|
PR_Github #22522 [ run ] triggered by Bot. Commit: |
|
PR_Github #22522 [ run ] completed with state |
Superjomn
force-pushed
the
perf-tuning
branch
from
a4ee0a9 to
6b56dac
Compare
|
/bot run |
|
PR_Github #22623 [ run ] triggered by Bot. Commit: |
|
PR_Github #22623 [ run ] completed with state |
|
/bot run |
|
PR_Github #22708 [ run ] triggered by Bot. Commit: |
Signed-off-by: Superjomn <[email protected]>
Signed-off-by: Superjomn <[email protected]>
Signed-off-by: Superjomn <[email protected]>
Signed-off-by: Superjomn <[email protected]>
Signed-off-by: Superjomn <[email protected]>
Signed-off-by: Superjomn <[email protected]>
Signed-off-by: Superjomn <[email protected]>
Signed-off-by: Superjomn <[email protected]>
|
/bot run |
|
/bot kill |
|
/bot reuse-pipeline |
|
PR_Github #23415 [ run ] triggered by Bot. Commit: |
|
PR_Github #23416 [ kill ] triggered by Bot. Commit: |
|
PR_Github #23415 [ run ] completed with state |
|
PR_Github #23417 [ reuse-pipeline ] triggered by Bot. Commit: |
|
PR_Github #23416 [ kill ] completed with state |
|
PR_Github #23417 [ reuse-pipeline ] completed with state |
|
/bot run |
|
PR_Github #23423 [ run ] triggered by Bot. Commit: |
|
/bot skip --comment "the CI has passed #8415 (comment)" |
|
PR_Github #23439 [ skip ] triggered by Bot. Commit: |
|
PR_Github #23423 [ run ] completed with state |
|
PR_Github #23439 [ skip ] completed with state |
… utilities to llm-api (NVIDIA#8415) Signed-off-by: Superjomn <[email protected]> Signed-off-by: FredricZ-2007 <[email protected]>
6 participants
Summary
tensorrt_llm._utilsand create a tracer in LLM class to track GC in LLM proxyThe GC tracer works in each process:
Summary by CodeRabbit
New Features
TLLM_ORCHESTRATOR_TYPE) for orchestrator type configuration, enabling dynamic selection between RPC and Ray orchestrators.Tests
Description
Test Coverage
PR Checklist
Please review the following before submitting your PR:
PR description clearly explains what and why. If using CodeRabbit's summary, please make sure it makes sense.
PR Follows TRT-LLM CODING GUIDELINES to the best of your knowledge.
Test cases are provided for new code paths (see test instructions)
Any new dependencies have been scanned for license and vulnerabilities
CODEOWNERS updated if ownership changes
Documentation updated as needed
The reviewers assigned automatically/manually are appropriate for the PR.
Please check this after reviewing the above items as appropriate for this PR.
GitHub Bot Help
/bot [-h] ['run', 'kill', 'skip', 'reuse-pipeline'] ...Provide a user friendly way for developers to interact with a Jenkins server.
Run
/bot [-h|--help]to print this help message.See details below for each supported subcommand.
run [--reuse-test (optional)pipeline-id --disable-fail-fast --skip-test --stage-list "A10-PyTorch-1, xxx" --gpu-type "A30, H100_PCIe" --test-backend "pytorch, cpp" --add-multi-gpu-test --only-multi-gpu-test --disable-multi-gpu-test --post-merge --extra-stage "H100_PCIe-TensorRT-Post-Merge-1, xxx" --detailed-log --debug(experimental)]Launch build/test pipelines. All previously running jobs will be killed.
--reuse-test (optional)pipeline-id(OPTIONAL) : Allow the new pipeline to reuse build artifacts and skip successful test stages from a specified pipeline or the last pipeline if no pipeline-id is indicated. If the Git commit ID has changed, this option will be always ignored. The DEFAULT behavior of the bot is to reuse build artifacts and successful test results from the last pipeline.--disable-reuse-test(OPTIONAL) : Explicitly prevent the pipeline from reusing build artifacts and skipping successful test stages from a previous pipeline. Ensure that all builds and tests are run regardless of previous successes.--disable-fail-fast(OPTIONAL) : Disable fail fast on build/tests/infra failures.--skip-test(OPTIONAL) : Skip all test stages, but still run build stages, package stages and sanity check stages. Note: Does NOT update GitHub check status.--stage-list "A10-PyTorch-1, xxx"(OPTIONAL) : Only run the specified test stages. Examples: "A10-PyTorch-1, xxx". Note: Does NOT update GitHub check status.--gpu-type "A30, H100_PCIe"(OPTIONAL) : Only run the test stages on the specified GPU types. Examples: "A30, H100_PCIe". Note: Does NOT update GitHub check status.--test-backend "pytorch, cpp"(OPTIONAL) : Skip test stages which don't match the specified backends. Only support [pytorch, cpp, tensorrt, triton]. Examples: "pytorch, cpp" (does not run test stages with tensorrt or triton backend). Note: Does NOT update GitHub pipeline status.--only-multi-gpu-test(OPTIONAL) : Only run the multi-GPU tests. Note: Does NOT update GitHub check status.--disable-multi-gpu-test(OPTIONAL) : Disable the multi-GPU tests. Note: Does NOT update GitHub check status.--add-multi-gpu-test(OPTIONAL) : Force run the multi-GPU tests in addition to running L0 pre-merge pipeline.--post-merge(OPTIONAL) : Run the L0 post-merge pipeline instead of the ordinary L0 pre-merge pipeline.--extra-stage "H100_PCIe-TensorRT-Post-Merge-1, xxx"(OPTIONAL) : Run the ordinary L0 pre-merge pipeline and specified test stages. Examples: --extra-stage "H100_PCIe-TensorRT-Post-Merge-1, xxx".--detailed-log(OPTIONAL) : Enable flushing out all logs to the Jenkins console. This will significantly increase the log volume and may slow down the job.--debug(OPTIONAL) : Experimental feature. Enable access to the CI container for debugging purpose. Note: Specify exactly one stage in thestage-listparameter to access the appropriate container environment. Note: Does NOT update GitHub check status.For guidance on mapping tests to stage names, see
docs/source/reference/ci-overview.mdand the
scripts/test_to_stage_mapping.pyhelper.kill
killKill all running builds associated with pull request.
skip
skip --comment COMMENTSkip testing for latest commit on pull request.
--comment "Reason for skipping build/test"is required. IMPORTANT NOTE: This is dangerous since lack of user care and validation can cause top of tree to break.reuse-pipeline
reuse-pipelineReuse a previous pipeline to validate current commit. This action will also kill all currently running builds associated with the pull request. IMPORTANT NOTE: This is dangerous since lack of user care and validation can cause top of tree to break.