[None][fix] Fix possible mpi broadcast and gather issue on large object #7507
Conversation
Signed-off-by: Dongxu Yang <[email protected]>
📝 WalkthroughWalkthroughImplements chunked, MPI-based broadcast and gather for large Python objects with new safe_broadcast and safe_gather helpers. MPIDist.broadcast/tp_broadcast/tp_gather now delegate to these helpers and accept a chunk_size parameter. TorchDist initializes additional TP/CP process groups. Imports updated to include mpi4py.MPI, pickle, math, and BuildInfo.ENABLE_MULTI_DEVICE. Changes
Sequence Diagram(s)sequenceDiagram
autonumber
participant Root as Root Rank
participant Peers as Peer Ranks
participant MPI as MPI Comm
Note over Root,Peers: Chunked broadcast of a Python object
Root->>Root: pickle.dumps(obj) -> bytes
Root->>MPI: Bcast header {total_size, n_chunks, ok_flag}
Peers->>MPI: Receive header
alt ok_flag == true
loop For each chunk
Root->>MPI: Bcast chunk[i]
Peers->>MPI: Receive chunk[i]
end
Peers->>Peers: Reassemble bytes, pickle.loads -> obj
else ok_flag == false
Note over Root,Peers: Abort/recover per header signal
end
sequenceDiagram
autonumber
participant All as All Ranks
participant Root as Root Rank
participant MPI as MPI Comm
Note over All: Chunked gather of Python objects
All->>All: pickle.dumps(local_obj) -> local_bytes
All->>MPI: Allgather lengths
Root->>MPI: Gatherv chunked payloads using displacements
Root->>Root: Reconstruct list by slicing payloads and pickle.loads
All-->>All: Non-root returns None
Estimated code review effort🎯 3 (Moderate) | ⏱️ ~25 minutes ✨ 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: 4
🧹 Nitpick comments (5)
tensorrt_llm/_torch/distributed/communicator.py (5)
12-15: Prefer module namespace imports for utils per guidelinesGuidelines say: “Maintain module namespace on import.” Consider
import tensorrt_llm._utils as _utilsand use_utils.mpi_comm()etc. This reduces symbol leakage and clarifies provenance.
120-133: Unify header broadcast path (optional)The root rank Bcasts the header inside the exception path and other ranks Bcast once later. It works, but it’s brittle. Prefer setting
ok_flag=0and falling through to a single commonBcastto keep collective structure obvious.Also applies to: 139-144
254-274: Guard against 32-bit displacements overflow in GathervMany MPI impls require 32-bit counts/displacements. With very large aggregated payloads,
displs32can overflow silently. Add an explicit check to fail fast with a clear message.counts32 = counts64.astype(np.int32) - displs32 = round_displs64.astype(np.int32) + displs32 = round_displs64.astype(np.int32) + if (round_displs64 > np.iinfo(np.int32).max).any(): + raise OverflowError( + "Gatherv displacements exceed 32-bit range; aggregated payload per round is too large. " + "Consider reducing per-rank object sizes or switching to a point-to-point gather." + )If you want, I can prototype a point-to-point gather variant that avoids Gatherv’s 32-bit displacement limit.
211-214: Docstring: clarify return in non-MPI modeFunction returns
[obj]whenENABLE_MULTI_DEVICEis False. Docstring currently says “On non-root: None” only.- Returns: - On root: list of deserialized objects (len == comm.size) - On non-root: None + Returns: + On root: list of deserialized objects (len == comm.size). + On non-root: None. + If ENABLE_MULTI_DEVICE is False, returns [obj] on the single process.
317-320: Pass-through looks good; ensure interface consistencyThis delegates to
safe_broadcastcorrectly. However, the abstractDistributed.broadcastsignature does not includechunk_size, which may confuse users and static checkers.Proposed update (outside this hunk) to the abstract base:
# In Distributed @abstractmethod def broadcast(self, obj, root=0, chunk_size: int = 4 * 1024 * 1024): """Broadcast object; subclasses may ignore chunk_size if not applicable.""" passAnd mirror the optional parameter in
TorchDist.broadcastsignature for consistency, even if unused.
📜 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 (1)
tensorrt_llm/_torch/distributed/communicator.py(3 hunks)
🧰 Additional context used
📓 Path-based instructions (4)
**/*
📄 CodeRabbit inference engine (CODING_GUIDELINES.md)
Filenames compiled into a target must be case-insensitively unique
Files:
tensorrt_llm/_torch/distributed/communicator.py
**/*.{h,hpp,hh,hxx,cc,cpp,cxx,cu,cuh,py}
📄 CodeRabbit inference engine (CODING_GUIDELINES.md)
Use spaces, not tabs; indent 4 spaces
Files:
tensorrt_llm/_torch/distributed/communicator.py
**/*.py
📄 CodeRabbit inference engine (CODING_GUIDELINES.md)
**/*.py: Code must target Python 3.8+
Indent with 4 spaces; do not use tabs (Python)
Maintain module namespace on import: prefer from package.subpackage import foo; use foo.Symbol()
Python filenames use snake_case
Python class names use PascalCase
Python functions and methods use snake_case
Python local variables use snake_case; if starting with a number concept, prefix with k (e.g., k_99th_percentile)
Python global variables use G_ prefix with UPPER_SNAKE_CASE
Python constants use UPPER_SNAKE_CASE
Avoid shadowing variables from outer scopes
Initialize all externally visible class members in init
For public interfaces, prefer docstrings over comments; comments should be for in-function or file-local interfaces
Use Google-style docstrings for classes and functions (Sphinx-parsable)
Document attributes and variables inline with docstrings immediately after assignment
Avoid reflection when a non-reflective approach suffices
Limit except clauses to specific exceptions where possible
When using try/except for duck-typing, keep try body minimal and move logic to else
Files:
tensorrt_llm/_torch/distributed/communicator.py
**/*.{cpp,cc,cxx,h,hpp,hh,hxx,cu,cuh,py}
📄 CodeRabbit inference engine (CODING_GUIDELINES.md)
Prepend NVIDIA copyright header (current year) to all source files
Files:
tensorrt_llm/_torch/distributed/communicator.py
🧬 Code graph analysis (1)
tensorrt_llm/_torch/distributed/communicator.py (1)
tensorrt_llm/_utils.py (9)
mpi_allgather(533-534)mpi_barrier(519-521)mpi_comm(475-476)mpi_isend(537-542)mpi_isend_object(565-568)mpi_recv(553-557)mpi_recv_object(571-574)mpi_send(545-550)mpi_send_object(560-562)
⏰ 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 (2)
tensorrt_llm/_torch/distributed/communicator.py (2)
355-357: tp_gather: good change; aligns with chunked gather helperDelegation and signature are appropriate after adopting
safe_gather.
359-361: tp_broadcast: good change; aligns with chunked broadcast helperDelegation and signature are appropriate after adopting
safe_broadcast.
Signed-off-by: Dongxu Yang <[email protected]>
dongxuy04
requested review from
QiJune,
dc3671,
kaiyux and
qiaoxj07
and removed request for
nv-yilinf and
yilin-void
|
/bot run --disable-fail-fast |
|
PR_Github #17534 [ run ] triggered by Bot |
|
PR_Github #17534 [ run ] completed with state |
…ct (NVIDIA#7507) Signed-off-by: Dongxu Yang <[email protected]>
…ct (NVIDIA#7507) Signed-off-by: Dongxu Yang <[email protected]>
4 participants
Summary by CodeRabbit
New Features
Improvements
Description
Fix possible MPI issues on large object.
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.