[None][chore] share input_ids buffers among different cuda graphs #7236
Conversation
Signed-off-by: junq <[email protected]> Signed-off-by: junq <[email protected]> Signed-off-by: junq <[email protected]> Signed-off-by: junq <[email protected]>
Signed-off-by: junq <[email protected]>
📝 WalkthroughWalkthroughReplaces per-(batch,draft_len) static CUDA tensor allocations with a single preallocated Changes
Sequence Diagram(s)sequenceDiagram
autonumber
participant App
participant Engine as PyTorchModelEngine
participant Runner as CUDAGraphRunner
participant Pool as shared_static_tensors
App->>Engine: instantiate(...)
Engine->>Runner: instantiate(..., cuda_graphs enabled)
activate Runner
Runner->>Runner: compute max_possible_draft_len
Runner->>Runner: _create_shared_static_tensors()
Runner->>Pool: allocate big tensors (input_ids, position_ids, [mrope_position_deltas])
deactivate Runner
App->>Runner: capture(initial_inputs)
activate Runner
Runner->>Pool: slice views for batch/beam/token range (sliced_static_tensors)
Runner->>Runner: build capture_inputs using sliced_static_tensors
Runner-->>App: return graph_metadata (includes spec_metadata)
deactivate Runner
App->>Runner: replay(current_inputs)
activate Runner
Runner->>Pool: copy current inputs into appropriate slice
Runner->>Runner: launch CUDA graph replay
Runner-->>App: outputs
deactivate Runner
Estimated code review effort🎯 4 (Complex) | ⏱️ ~45–75 minutes Possibly related PRs
Suggested labels
Suggested reviewers
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
|
|
/bot run |
|
PR_Github #16499 [ run ] triggered by Bot |
![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
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (2)
tensorrt_llm/_torch/pyexecutor/cuda_graph_runner.py (2)
1-1: Add the required 2025 NVIDIA copyright header at file topPer repo guidelines, every Python source must prepend the NVIDIA copyright header (current year). This file is missing it.
Apply:
+# Copyright (c) 2025, NVIDIA CORPORATION & AFFILIATES. All rights reserved. + import bisect import contextlib import weakref
281-287: Invokeengine.use_mrope()instead of referencing the bound methodThe grep run confirmed two spots where
engine.use_mropeis used without parentheses, causing a bound-method object (always truthy) to be passed or tested instead of its boolean return value.· tensorrt_llm/_torch/pyexecutor/cuda_graph_runner.py:72
· tensorrt_llm/_torch/pyexecutor/cuda_graph_runner.py:285Please apply the following fixes:
--- a/tensorrt_llm/_torch/pyexecutor/cuda_graph_runner.py @@ -69,7 +69,7 @@ } - if engine.use_mrope: + if engine.use_mrope(): self.shared_static_tensors["mrope_position_deltas"] = torch.zeros( (self.max_supported_batch_size, 1), ) @@ -285,7 +285,7 @@ self.padding_dummy_request = kv_cache_manager.add_dummy_requests( [CUDA_GRAPH_DUMMY_REQUEST_ID], is_gen=True, - max_num_draft_tokens=engine.max_draft_len, - use_mrope=engine.use_mrope, + max_num_draft_tokens=engine.max_draft_len, + use_mrope=engine.use_mrope(), max_beam_width=engine.max_beam_width)[0] self.padding_dummy_request.is_cuda_graph_dummy = TrueThese changes ensure that the method’s boolean result is used, not the method object itself.
🧹 Nitpick comments (8)
tensorrt_llm/_torch/pyexecutor/cuda_graph_runner.py (8)
43-45: Avoid duplicating draft length state; use the existing draft_len propertymax_possible_draft_len duplicates the semantics of draft_len. Keeping both invites drift. Use draft_len everywhere.
Apply:
- self.max_possible_draft_len = (self.spec_config.max_draft_len - if self.enable_spec_decode else 0)And below (see other diffs in Lines 61-64 and 173-176) replace self.max_possible_draft_len with self.draft_len.
53-56: Lazily allocate and make release explicit (large buffers)Creation is unconditional when enabled. Consider:
- Allocate only if supported_batch_sizes is non-empty.
- Also free these buffers in clear() to release VRAM under pressure.
Apply in init:
- if self.enabled: + if self.enabled and self.supported_batch_sizes: self._create_shared_static_tensors()And in clear() (see Lines 320-330):
self.graph_metadata.clear() del self.memory_pool self.memory_pool = None + self.shared_static_tensors.clear() torch.cuda.empty_cache()
61-64: Use draft_len property to compute token_per_requestKeeps a single source of truth for the effective draft length.
Apply:
- token_per_request = self.max_possible_draft_len + 1 + token_per_request = self.draft_len + 1
177-187: Good: reuse via sliced views; minor improvementThe capture uses views; mirror this pattern in replay for clarity and to protect against accidental over/under-copy if invariants change.
Apply in replay (see Lines 222-235):
- static_tensors = self.shared_static_tensors + token_per_request = self.draft_len + 1 + expected_tokens = batch_size * self.max_beam_width * token_per_request + static_tensors = { + "input_ids": self.shared_static_tensors["input_ids"][:expected_tokens], + "position_ids": self.shared_static_tensors["position_ids"][:, :expected_tokens], + } + if "mrope_position_deltas" in self.shared_static_tensors: + static_tensors["mrope_position_deltas"] = \ + self.shared_static_tensors["mrope_position_deltas"][:batch_size]
222-235: Safer copying and shape assertion during replayReplay currently copies directly into base tensors; if invariants drift, this can silently misalign with captured slices. Add a sanity check or slice as in capture.
Apply:
- static_tensors = self.shared_static_tensors - - input_ids = current_inputs["input_ids"] - seqlen = input_ids.shape[0] - static_tensors["input_ids"][:seqlen].copy_(input_ids) - - position_ids = current_inputs["position_ids"] - static_tensors["position_ids"][:, :seqlen].copy_(position_ids) + token_per_request = self.draft_len + 1 + expected = batch_size * self.max_beam_width * token_per_request + input_ids = current_inputs["input_ids"] + seqlen = input_ids.shape[0] + assert seqlen == expected, ( + f"Unexpected input_ids length {seqlen}, expected {expected} for CUDA graph replay." + ) + self.shared_static_tensors["input_ids"][:seqlen].copy_(input_ids) + + position_ids = current_inputs["position_ids"] + self.shared_static_tensors["position_ids"][:, :seqlen].copy_(position_ids) if "mrope_position_deltas" in current_inputs: - assert "mrope_position_deltas" in static_tensors - static_tensors["mrope_position_deltas"][:batch_size].copy_( + assert "mrope_position_deltas" in self.shared_static_tensors + self.shared_static_tensors["mrope_position_deltas"][:batch_size].copy_( current_inputs["mrope_position_deltas"])If you prefer not to assert in production, guard it with a debug flag.
320-330: Consider freeing shared_static_tensors in clear()clear() releases graphs/mempool but leaves the shared tensor pool allocated. If engines are long-lived, this is fine; if clear() is used for reclamation, also clear the pool.
Apply:
self.graph_metadata.clear() del self.memory_pool self.memory_pool = None + self.shared_static_tensors.clear() torch.cuda.empty_cache()If we should keep the pool across captures, we can add a separate release_static_pool() method instead. Preference?
127-137: Variable naming is confusing in TP allgather sectionNot a bug, but readability suffers (e.g., is_all_gen_only vs all_can_graph). Consider renaming locals for clarity.
No code change required; optional cleanup for future readers.
166-171: Add a brief Google-style docstring for capture() params/returnsPublic API, frequently touched. Minimal docstring with Args/Returns improves maintainability.
I can draft it if you’d like.
📜 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 (2)
tensorrt_llm/_torch/pyexecutor/cuda_graph_runner.py(3 hunks)tensorrt_llm/_torch/pyexecutor/model_engine.py(0 hunks)
💤 Files with no reviewable changes (1)
- tensorrt_llm/_torch/pyexecutor/model_engine.py
🧰 Additional context used
📓 Path-based instructions (2)
**/*.py
📄 CodeRabbit inference engine (CODING_GUIDELINES.md)
**/*.py: Code must target Python 3.8+
Indent Python code with 4 spaces; do not use tabs
Preserve module namespaces when importing; import modules/packages and access members via the module (e.g., from package.subpackage import foo; foo.SomeClass())
Python file names should be snake_case
Python class names should be PascalCase
Python functions/methods and local variables should be snake_case; variables beginning with a number should be prefixed with k_ (e.g., k_99th_percentile)
Global variables should be UPPER_SNAKE_CASE prefixed with G_ (e.g., G_MY_GLOBAL); constants should be UPPER_SNAKE_CASE
Avoid shadowing variables from outer scopes; initialize all externally visible members in init
Prefer docstrings for interfaces used outside a file; comments should be reserved for in-function or file-local interfaces
Use Google-style docstrings for classes and functions; attributes and variables may be documented inline with trailing string literals
Avoid reflection when simpler, explicit code suffices (e.g., avoid dict(**locals()) patterns)
In try/except, catch the narrowest exceptions possible
For duck-typing patterns, keep the try body minimal and move logic to else to avoid masking unrelated failures
Files:
tensorrt_llm/_torch/pyexecutor/cuda_graph_runner.py
**/*.{c,cc,cpp,cxx,h,hh,hpp,hxx,cu,cuh,py}
📄 CodeRabbit inference engine (CODING_GUIDELINES.md)
Prepend the NVIDIA copyright header (current year) to all source files (.cpp, .h, .cu, .py, etc.)
Files:
tensorrt_llm/_torch/pyexecutor/cuda_graph_runner.py
🧬 Code graph analysis (1)
tensorrt_llm/_torch/pyexecutor/cuda_graph_runner.py (1)
tensorrt_llm/_torch/pyexecutor/model_engine.py (1)
use_mrope(480-488)
🪛 Ruff (0.12.2)
tensorrt_llm/_torch/pyexecutor/cuda_graph_runner.py
51-51: Undefined name Request
(F821)
⏰ 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 (5)
tensorrt_llm/_torch/pyexecutor/cuda_graph_runner.py (5)
173-176: LGTM: token accounting for captureDeriving num_tokens_for_capture from batch_size, beam width, and token_per_request is correct for prealloc slicing. Just align token_per_request to use draft_len per earlier comment.
Apply:
- token_per_request = self.max_possible_draft_len + 1 + token_per_request = self.draft_len + 1 num_tokens_for_capture = (batch_size * self.max_beam_width * token_per_request)
188-191: LGTM: minimal mutation of capture_inputsCopy-on-write of initial_inputs and updating just the shared slices is clean and avoids reallocations.
194-195: LGTM: spec_metadata sourcing is saferPulling spec_metadata from initial_inputs avoids stale local state.
1-4: The scripts are running to locate any Python‐sideRequestdefinitions and list the files in thepyexecutordirectory. I’ll follow up as soon as we have those results.
236-239: Return the Tensor, not the weak-ref thunk incuda_graph_runner.replayReplace the current return of the weak-ref callable with a call to it so that
replay()returns atorch.Tensoras typed:• File: tensorrt_llm/_torch/pyexecutor/cuda_graph_runner.py
Lines 236–239Suggested patch:
- self.graphs[key].replay() - output_ref = self.graph_outputs[key] - - return output_ref + self.graphs[key].replay() + output_ref = self.graph_outputs[key] + return output_ref()All existing call sites (in
model_engine.pyand the various unit tests undertests/unittest/_torch/…) treat the result ofreplay(...)as aTensor, so this change restores the intended API contract.
|
PR_Github #16499 [ run ] completed with state |
|
/bot run |
|
PR_Github #16569 [ run ] triggered by Bot |
|
PR_Github #16569 [ run ] completed with state |
Signed-off-by: junq <[email protected]>
|
/bot run |
|
PR_Github #16679 [ run ] triggered by Bot |
There was a problem hiding this comment.
Actionable comments posted: 1
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (1)
tensorrt_llm/_torch/pyexecutor/cuda_graph_runner.py (1)
14-16: Fix Ruff F821 for forward-declared Request.Type is only referenced in quotes but Ruff still flags it. Import under TYPE_CHECKING.
Apply:
if TYPE_CHECKING: - from .model_engine import PyTorchModelEngine + from .model_engine import PyTorchModelEngine + from .scheduler import Request
♻️ Duplicate comments (2)
tensorrt_llm/_torch/pyexecutor/cuda_graph_runner.py (2)
286-287: Call engine.use_mrope() when adding dummy requests.Same method/attribute bug as above; otherwise mrope is always treated as True here.
Apply:
- use_mrope=engine.use_mrope, + use_mrope=engine.use_mrope(),
57-78: Call engine.use_mrope() and avoid large device memsets for shared buffers.
- Bug: engine.use_mrope is a method; attribute check is always truthy and over-allocates mrope buffers. Same issue reappears later.
- Perf: use torch.empty for big static buffers; you overwrite before use during capture/replay anyway.
Apply:
self.shared_static_tensors = { - "input_ids": - torch.ones((max_total_tokens, ), device="cuda", dtype=torch.int32), - "position_ids": - torch.zeros((1, max_total_tokens), device="cuda", - dtype=torch.int32), + "input_ids": torch.empty((max_total_tokens,), device="cuda", dtype=torch.int32), + "position_ids": torch.empty((1, max_total_tokens), device="cuda", dtype=torch.int32), } - if engine.use_mrope: + if engine.use_mrope(): self.shared_static_tensors["mrope_position_deltas"] = torch.zeros( (self.max_supported_batch_size, 1), device="cuda", dtype=torch.int32)
🧹 Nitpick comments (5)
tensorrt_llm/_torch/pyexecutor/cuda_graph_runner.py (5)
1-1: Add NVIDIA copyright header.Per repo guidelines, prepend the current-year NVIDIA header to this source file.
Example:
+# Copyright (c) 2025, NVIDIA CORPORATION. All rights reserved.
223-236: Optional: assert replay sizes against preallocated capacity.Adds an early, clear error if a future change violates sizing invariants.
Apply:
static_tensors = self.shared_static_tensors input_ids = current_inputs["input_ids"] seqlen = input_ids.shape[0] + assert seqlen <= static_tensors["input_ids"].numel(), \ + "Replay seqlen exceeds preallocated input_ids capacity" static_tensors["input_ids"][:seqlen].copy_(input_ids) position_ids = current_inputs["position_ids"] + assert seqlen <= static_tensors["position_ids"].shape[1], \ + "Replay seqlen exceeds preallocated position_ids capacity" static_tensors["position_ids"][:, :seqlen].copy_(position_ids)
99-101: Harden destructor to avoid teardown-time CUDA errors.Guard clear() in del to prevent noisy exceptions during interpreter shutdown.
Apply:
- def __del__(self): - self.clear() + def __del__(self): + try: + self.clear() + except Exception: + pass
66-72: Optional: document why int32 is used for IDs.Quick docstring or comment prevents regressions to wrong dtypes later.
121-136: Nit: variable names are misleading.is_all_gen_only actually checks can_run_cuda_graph; consider renaming for clarity.
📜 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/pyexecutor/cuda_graph_runner.py(3 hunks)
🧰 Additional context used
📓 Path-based instructions (2)
**/*.py
📄 CodeRabbit inference engine (CODING_GUIDELINES.md)
**/*.py: Code must target Python 3.8+
Indent Python code with 4 spaces; do not use tabs
Preserve module namespaces when importing; import modules/packages and access members via the module (e.g., from package.subpackage import foo; foo.SomeClass())
Python file names should be snake_case
Python class names should be PascalCase
Python functions/methods and local variables should be snake_case; variables beginning with a number should be prefixed with k_ (e.g., k_99th_percentile)
Global variables should be UPPER_SNAKE_CASE prefixed with G_ (e.g., G_MY_GLOBAL); constants should be UPPER_SNAKE_CASE
Avoid shadowing variables from outer scopes; initialize all externally visible members in init
Prefer docstrings for interfaces used outside a file; comments should be reserved for in-function or file-local interfaces
Use Google-style docstrings for classes and functions; attributes and variables may be documented inline with trailing string literals
Avoid reflection when simpler, explicit code suffices (e.g., avoid dict(**locals()) patterns)
In try/except, catch the narrowest exceptions possible
For duck-typing patterns, keep the try body minimal and move logic to else to avoid masking unrelated failures
Files:
tensorrt_llm/_torch/pyexecutor/cuda_graph_runner.py
**/*.{c,cc,cpp,cxx,h,hh,hpp,hxx,cu,cuh,py}
📄 CodeRabbit inference engine (CODING_GUIDELINES.md)
Prepend the NVIDIA copyright header (current year) to all source files (.cpp, .h, .cu, .py, etc.)
Files:
tensorrt_llm/_torch/pyexecutor/cuda_graph_runner.py
🧬 Code graph analysis (1)
tensorrt_llm/_torch/pyexecutor/cuda_graph_runner.py (1)
tensorrt_llm/_torch/pyexecutor/model_engine.py (1)
use_mrope(481-489)
🪛 Ruff (0.12.2)
tensorrt_llm/_torch/pyexecutor/cuda_graph_runner.py
51-51: Undefined name Request
(F821)
⏰ 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 (4)
tensorrt_llm/_torch/pyexecutor/cuda_graph_runner.py (4)
43-45: LGTM: clear max_possible_draft_len calculation.
53-56: LGTM: allocate shared pool only when enabled.
61-65: LGTM: capping by engine.max_num_tokens.Good safeguard against over-allocation.
73-77: No action needed –use_mropeis correctly used as a @PropertyAll occurrences of
use_mropewere reviewed across the repository. Since it’s implemented as a Python@property, accessing it without parentheses is intentional and correct. No fixes are required.
|
PR_Github #16679 [ run ] completed with state |
|
/bot run |
|
PR_Github #16690 [ run ] triggered by Bot |
|
PR_Github #16690 [ run ] completed with state |
|
/bot run |
|
PR_Github #17254 [ run ] triggered by Bot |
|
PR_Github #17254 [ run ] completed with state |
|
/bot run |
|
PR_Github #17281 [ run ] triggered by Bot |
|
PR_Github #17281 [ run ] completed with state |
|
/bot run |
|
PR_Github #17411 [ run ] triggered by Bot |
|
PR_Github #17411 [ run ] completed with state |
|
/bot run |
|
PR_Github #17689 [ run ] triggered by Bot |
|
PR_Github #17689 [ run ] completed with state |
|
/bot run |
|
PR_Github #17799 [ run ] triggered by Bot |
|
PR_Github #17799 [ run ] completed with state |
|
/bot run |
|
PR_Github #17826 [ run ] triggered by Bot |
|
PR_Github #17826 [ run ] completed with state |
|
/bot run |
|
PR_Github #17852 [ run ] triggered by Bot |
|
PR_Github #17852 [ run ] completed with state |
|
/bot run |
|
PR_Github #17886 [ run ] triggered by Bot |
|
PR_Github #17886 [ run ] completed with state |
…IDIA#7236) Signed-off-by: junq <[email protected]>
4 participants
Summary by CodeRabbit
Refactor
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.