[https://nvbugs/5480289][fix] release slot manager in mtp MTPHiddenStatesManager #7340
Conversation
📝 WalkthroughWalkthroughAdds a SlotManager.shutdown() method to forcibly release all slots, wires MTPHiddenStatesManager.shutdown() to call it, and updates CUDAGraphRunner.clear() to also reset padding_dummy_request to None. Changes
Sequence Diagram(s)sequenceDiagram
autonumber
actor Client
participant MTPHiddenStatesManager as MTPHiddenStatesManager
participant SlotManager as SlotManager
participant Slots as Slots (internal)
Client->>MTPHiddenStatesManager: shutdown()
MTPHiddenStatesManager->>SlotManager: shutdown()
rect rgb(235, 245, 255)
note right of SlotManager: Iterate current request→slot map
loop For each active slot
SlotManager->>Slots: remove_slot(request_id)
Slots-->>SlotManager: slot freed
end
end
SlotManager-->>MTPHiddenStatesManager: all slots released (asserts pass)
MTPHiddenStatesManager-->>Client: shutdown complete
Estimated code review effort🎯 2 (Simple) | ⏱️ ~10 minutes Tip 🔌 Remote MCP (Model Context Protocol) integration is now available!Pro plan users can now connect to remote MCP servers from the Integrations page. Connect with popular remote MCPs such as Notion and Linear to add more context to your reviews and chats. ✨ Finishing Touches
🧪 Generate unit tests
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. 🪧 TipsChatThere are 3 ways to chat with CodeRabbit:
SupportNeed help? Create a ticket on our support page for assistance with any issues or questions. CodeRabbit Commands (Invoked using PR/Issue comments)Type Other keywords and placeholders
Status, Documentation and Community
|
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 1
🧹 Nitpick comments (6)
tensorrt_llm/_torch/pyexecutor/cuda_graph_runner.py (2)
49-50: Fix type annotation: use LlmRequest, not an undefined "Request".“Request” isn’t defined/imported; use LlmRequest under TYPE_CHECKING to satisfy static checkers.
if TYPE_CHECKING: - from .model_engine import PyTorchModelEngine + from .model_engine import PyTorchModelEngine + from .llm_request import LlmRequest @@ - self.padding_dummy_request: Optional["Request"] = None + self.padding_dummy_request: Optional["LlmRequest"] = NoneYou can run mypy/pyright locally to ensure no remaining annotation issues.
Also applies to: 14-16
1-1: Add 2025 NVIDIA copyright header.Required by the repo’s coding guidelines.
+# Copyright (c) 2025, NVIDIA CORPORATION. All rights reserved.tensorrt_llm/_torch/speculative/mtp.py (2)
88-88: Release MTP GPU pools too, and document shutdown.Good to shut down the SlotManager. Also drop large CUDA tensors to free memory deterministically on shutdown.
def shutdown(self): - self.slot_manager.shutdown() + """Release all MTP resources (slots and GPU pools).""" + self.slot_manager.shutdown() + # Drop large buffers to free device memory promptly + self.mtp_past_hidden_states_pool = None + self.mtp_past_tokens_pool = None + if self.use_relaxed_acceptance_for_thinking: + self.mtp_relaxed_delta_pool = None
1-1: Add 2025 NVIDIA copyright header.+# Copyright (c) 2025, NVIDIA CORPORATION. All rights reserved.tensorrt_llm/_torch/pyexecutor/resource_manager.py (2)
981-987: Add a docstring and log invariant failures rather than assert in prod.Asserts may be stripped in optimized runs; prefer a clear exception or logging.
- def shutdown(self): - req_ids_list = list(self.slot_mapping.keys()) - for rid in req_ids_list: - self.remove_slot(rid) - assert len(self.slot_mapping) == 0 and len( - self.free_slots) == self.max_num_requests + def shutdown(self): + """Force-release all active slots and reset to the initial state.""" + req_ids_list = list(self.slot_mapping.keys()) + for rid in req_ids_list: + self.remove_slot(rid) + ok = (len(self.slot_mapping) == 0 + and len(self.free_slots) == self.max_num_requests) + if not ok: + raise RuntimeError( + f"SlotManager shutdown invariant failed: " + f"slot_mapping={len(self.slot_mapping)}, " + f"free_slots={len(self.free_slots)}/{self.max_num_requests}" + )
1-1: Add 2025 NVIDIA copyright header.+# Copyright (c) 2025, NVIDIA CORPORATION. All rights reserved.
📜 Review details
Configuration used: Path: .coderabbit.yaml
Review profile: CHILL
Plan: Pro
💡 Knowledge Base configuration:
- MCP integration is disabled by default for public repositories
- Jira integration is disabled by default for public repositories
- Linear integration is disabled by default for public repositories
You can enable these sources in your CodeRabbit configuration.
📒 Files selected for processing (3)
tensorrt_llm/_torch/pyexecutor/cuda_graph_runner.py(1 hunks)tensorrt_llm/_torch/pyexecutor/resource_manager.py(1 hunks)tensorrt_llm/_torch/speculative/mtp.py(1 hunks)
🧰 Additional context used
📓 Path-based instructions (2)
**/*.py
📄 CodeRabbit inference engine (CODING_GUIDELINES.md)
**/*.py: Code must target Python 3.8+
Indent with 4 spaces; do not use tabs
Preserve module namespaces in imports: import the subpackage/module, not the symbol (from package.subpackage import foo; foo.SomeClass())
Naming: files snake_case; classes PascalCase; functions/methods snake_case; local variables snake_case (k_ prefix if starting with a number); globals G_ + UPPER_SNAKE_CASE; constants UPPER_SNAKE_CASE
Avoid shadowing outer-scope variables; initialize all externally visible members in init
Prefer docstrings for interfaces used outside a file; reserve comments for function-internal or file-local interfaces
Use Google-style docstrings for classes and functions; inline docstrings for attributes/variables are allowed
Avoid reflection when straightforward code suffices (e.g., prefer explicit parameters over dict(**locals()))
Use narrow except clauses (e.g., catch FileNotFoundError instead of bare except)
For duck-typing try/except, keep try body minimal and use else for the main logic
Files:
tensorrt_llm/_torch/speculative/mtp.pytensorrt_llm/_torch/pyexecutor/cuda_graph_runner.pytensorrt_llm/_torch/pyexecutor/resource_manager.py
**/*.{cpp,cc,cxx,cu,h,hpp,hh,hxx,cuh,py}
📄 CodeRabbit inference engine (CODING_GUIDELINES.md)
Prepend NVIDIA copyright header with current year to all source files
Files:
tensorrt_llm/_torch/speculative/mtp.pytensorrt_llm/_torch/pyexecutor/cuda_graph_runner.pytensorrt_llm/_torch/pyexecutor/resource_manager.py
🧬 Code graph analysis (2)
tensorrt_llm/_torch/speculative/mtp.py (2)
tensorrt_llm/_torch/pyexecutor/resource_manager.py (4)
shutdown(80-81)shutdown(355-356)shutdown(981-986)shutdown(1132-1133)tensorrt_llm/_torch/pyexecutor/py_executor.py (1)
shutdown(353-366)
tensorrt_llm/_torch/pyexecutor/resource_manager.py (2)
tensorrt_llm/_torch/speculative/mtp.py (1)
shutdown(87-88)tensorrt_llm/_torch/pyexecutor/py_executor.py (1)
shutdown(353-366)
⏰ 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 (1)
tensorrt_llm/_torch/pyexecutor/resource_manager.py (1)
981-987: LGTM: Idempotent bulk release for SlotManager.Iterating over a snapshot of keys and asserting invariants is correct and safe.
|
/bot run |
|
PR_Github #16868 [ run ] triggered by Bot |
yweng0828
changed the title
|
PR_Github #16868 [ run ] completed with state |
yweng0828
force-pushed
the
yweng/fix_release_slot_manager_in_mtp
branch
from
07ae634 to
ab6cc5b
Compare
|
/bot run |
|
PR_Github #16961 [ run ] triggered by Bot |
|
PR_Github #16961 [ run ] completed with state |
|
/bot run |
|
PR_Github #16987 [ run ] triggered by Bot |
|
PR_Github #16987 [ run ] completed with state |
yweng0828
force-pushed
the
yweng/fix_release_slot_manager_in_mtp
branch
from
ab6cc5b to
9bc5578
Compare
|
/bot run |
|
PR_Github #16994 [ run ] triggered by Bot |
|
PR_Github #16994 [ run ] completed with state |
yweng0828
force-pushed
the
yweng/fix_release_slot_manager_in_mtp
branch
from
9bc5578 to
c2a90e7
Compare
|
/bot run |
|
PR_Github #17041 [ run ] triggered by Bot |
|
PR_Github #17041 [ run ] completed with state |
|
/bot run --disable-fail-fast |
|
PR_Github #17049 [ run ] triggered by Bot |
|
PR_Github #17049 [ run ] completed with state |
|
/bot run --disable-fail-fast |
|
PR_Github #17105 [ run ] triggered by Bot |
|
PR_Github #17105 [ run ] completed with state |
yweng0828
force-pushed
the
yweng/fix_release_slot_manager_in_mtp
branch
from
c2a90e7 to
127e421
Compare
|
/bot run --disable-fail-fast |
|
PR_Github #17188 [ run ] triggered by Bot |
|
PR_Github #17188 [ run ] completed with state |
|
/bot run --disable-fail-fast |
|
PR_Github #17277 [ run ] triggered by Bot |
Signed-off-by: Yue Weng <25103990+yweng0828@users.noreply.github.com>
yweng0828
force-pushed
the
yweng/fix_release_slot_manager_in_mtp
branch
from
127e421 to
6ac55ac
Compare
|
/bot run --disable-fail-fast |
|
PR_Github #17296 [ run ] triggered by Bot |
|
PR_Github #17277 [ run ] completed with state |
|
PR_Github #17296 [ run ] completed with state |
|
/bot run |
|
PR_Github #17358 [ run ] triggered by Bot |
|
PR_Github #17358 [ run ] completed with state |
4 participants
Summary by CodeRabbit
Description
The root cause of this issue is that we didn't release the dummy request and the occupied slot after the KV cache estimation. This caused the number of slots required to exceed
max_num_requestsduring the CUDA graph warmup.Fix Step 1: After completing the KV cache estimation, we need to release the resources occupied by the model_engine and the dummy requests.
Fix Step 2: Implement the
shutdown()function of the MTP'sMTPHiddenStatesManager. We need to properly release the slot (which is used by the dummy request) in the slot manager during_release_cuda_graphs().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.