feat(cuda.core): add cluster scheduling policy to LaunchConfig by atiaomar1978-hub · Pull Request #2703 · NVIDIA/cuda-python

atiaomar1978-hub · 2026-08-26T00:17:31Z

Description

Add cluster_scheduling_policy_preference to LaunchConfig, mapping to
CU_LAUNCH_ATTRIBUTE_CLUSTER_SCHEDULING_POLICY_PREFERENCE via
ClusterSchedulingPolicyType (DEFAULT, SPREAD, LOAD_BALANCING).

Unlike stream synchronization policy (#2628), cluster scheduling policy is
documented as valid for graph nodes and kernel launches, so LaunchConfig is
the correct surface.

This is a signed re-issue of #2702 (same change, linear history on current
main). Commits are SSH-signed. GitHub would not reopen #2668.

Checklist

New or existing tests cover these changes.
The documentation is up to date with these changes.

Test plan

Native attribute mapping for DEFAULT / SPREAD / LOAD_BALANCING
Getter/setter round-trip for each policy
Driver enum acceptance; invalid value rejection; CC < 9.0 gate
Combined cluster + policy attributes
Hopper+ launch() smoke for all three policies
RunPod H100 80GB HBM3 / driver 580 / CTK 13.3 — 13 passed, 0 skipped
pre-commit.ci stubgen-pyx stubs regenerated

copy-pr-bot · 2026-08-26T00:17:35Z

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

Expose CU_LAUNCH_ATTRIBUTE_CLUSTER_SCHEDULING_POLICY_PREFERENCE on LaunchConfig via ClusterSchedulingPolicyType, with validation, CC >= 9.0 gating, and tests mapping to the native launch attribute. Closes NVIDIA#2629 Co-authored-by: Cursor <cursoragent@cursor.com> Signed-off-by: Omar Atie <atiaomar1978@gmail.com>

Assert LaunchConfig emits both CU_LAUNCH_ATTRIBUTE_CLUSTER_DIMENSION and CU_LAUNCH_ATTRIBUTE_CLUSTER_SCHEDULING_POLICY_PREFERENCE when set. Co-authored-by: Cursor <cursoragent@cursor.com> Signed-off-by: Omar Atie <atiaomar1978@gmail.com>

Use init_cuda with an inline noop kernel so test_launcher.py does not depend on the get_saxpy_kernel_cubin fixture from test_module.py. Co-authored-by: Cursor <cursoragent@cursor.com> Signed-off-by: Omar Atie <atiaomar1978@gmail.com>

…rough Align NVIDIA#2629 tests with reviewer guidance: round-trip each policy on LaunchConfig and exercise launch() for DEFAULT/SPREAD/LOAD_BALANCING. Co-authored-by: Cursor <cursoragent@cursor.com> Signed-off-by: Omar Atie <atiaomar1978@gmail.com>

atiaomar1978-hub · 2026-08-26T00:19:41Z

Hi @lijinf2 — when you have a moment, could you please review this PR?

This is the signed re-issue of #2702 (GitHub would not reopen #2668). It implements the plan we aligned on in #2629: LaunchConfig.cluster_scheduling_policy_preference maps to CU_LAUNCH_ATTRIBUTE_CLUSTER_SCHEDULING_POLICY_PREFERENCE (DEFAULT / SPREAD / LOAD_BALANCING), with CC >= 9.0 gating.

Tests follow your guidance (getter/setter plus a launch() run-through; no attempt to observe the live scheduler). On RunPod H100 80GB HBM3 (CC 9.0, driver 580, CTK 13.3) the cluster-policy suite is 13 passed, 0 skipped.

I do not have permission to set assignee or milestone on this account; cuda.core 1.3.0 would match the related work if that is still the right milestone.

Sorry for the force-push on this branch. I know rewriting published history is against the project rules. I did it only so the commits would be SSH-signed and show as Verified on GitHub. I will not force-push again.

lijinf2 · 2026-08-27T21:26:43Z

/ok to test c6c283b

github-actions · 2026-08-27T21:44:52Z

Doc Preview CI
🚀 View preview at https://nvidia.github.io/cuda-python/pr-preview/pr-2703/
https://nvidia.github.io/cuda-python/pr-preview/pr-2703/cuda-core/
https://nvidia.github.io/cuda-python/pr-preview/pr-2703/cuda-bindings/
https://nvidia.github.io/cuda-python/pr-preview/pr-2703/cuda-pathfinder/
Preview will be ready when the GitHub Pages deployment is complete.

CI test_subpackage_exports_are_documented[typing] failed because the new typing export was missing from docs/source/*.rst. Signed-off-by: Omar Atie <atiaomar1978@gmail.com>

atiaomar1978-hub · 2026-08-27T22:27:17Z

Posted a docs fix for the CI failure on test_subpackage_exports_are_documented[typing].

ClusterSchedulingPolicyType was exported from cuda.core.typing but missing from docs/source/*.rst. Commit 39c9fc0843 adds it to api_private.rst (SSH-signed / Verified).

RunPod (RTX 2000 Ada, CC 8.9 — L4 community was out of stock; same compute capability as the failing L4 jobs):

test_subpackage_exports_are_documented[typing] passed
cluster launcher suite: 10 passed, 3 skipped (launch smokes require CC >= 9.0)

PR metadata is now set (cuda.core, feature, assignee, milestone).

atiaomar1978-hub · 2026-08-27T22:28:15Z

Hi @lijinf2 — could you please trigger CI again on 39c9fc0843?

The previous /ok to test covered c6c283bb5d3f. That matrix failed; this SHA has the docs fix and has not gotten a new GPU run yet.

Why CI failed

Every GPU job died on the same assertion — not on LaunchConfig or Hopper launch:

tests/test_api_docs_consistency.py::test_subpackage_exports_are_documented[typing] FAILED
E   AssertionError: public by cuda.core.typing.__all__ but missing from docs/source/*.rst:
E       ['ClusterSchedulingPolicyType']

ClusterSchedulingPolicyType was added to cuda.core.typing.__all__ but was not listed in cuda_core/docs/source/api_private.rst (the autosummary the consistency test reads). Example: linux-64 / py3.10 / CUDA 13.3 wheels / L4 — 1 failed, launcher cluster tests passed or skipped by CC.

Why it works again

39c9fc0843 adds the missing autosummary line:

typing.ClusterSchedulingPolicyType

That name is now present in docs/source/*.rst, so the export-vs-docs check passes. Runtime tests were already green.

RunPod replay (pod `7xp9vqqo44kk1s`, RTX 2000 Ada, CC 8.9)

L4 community was out of stock; Ada is the same CC 8.9 as the failing L4 jobs. HEAD on the pod was 39c9fc0.

############ CI FAILURE: docs export consistency (typing) ############
tests/test_api_docs_consistency.py::test_subpackage_exports_are_documented[typing] PASSED [100%]
======================= 1 passed, 5 deselected in 1.10s ========================

############ CLUSTER SCHEDULING POLICY TESTS (CI-like GPU) ############
compute_capability ComputeCapability(major=8, minor=9)
tests/test_launcher.py::test_to_native_launch_config_cluster_scheduling_policy[0] PASSED
tests/test_launcher.py::test_to_native_launch_config_cluster_scheduling_policy[1] PASSED
tests/test_launcher.py::test_to_native_launch_config_cluster_scheduling_policy[2] PASSED
tests/test_launcher.py::test_to_native_launch_config_cluster_scheduling_policy_accepts_driver_enum PASSED
tests/test_launcher.py::test_launch_config_cluster_scheduling_policy_invalid PASSED
tests/test_launcher.py::test_launch_config_cluster_scheduling_policy_rejects_pre_hopper_cc PASSED
tests/test_launcher.py::test_to_native_launch_config_cluster_and_policy PASSED
tests/test_launcher.py::test_launch_config_cluster_scheduling_policy_getter_setter[0] PASSED
tests/test_launcher.py::test_launch_config_cluster_scheduling_policy_getter_setter[1] PASSED
tests/test_launcher.py::test_launch_config_cluster_scheduling_policy_getter_setter[2] PASSED
tests/test_launcher.py::test_launch_cluster_scheduling_policy_smoke[0] SKIPPED
tests/test_launcher.py::test_launch_cluster_scheduling_policy_smoke[1] SKIPPED
tests/test_launcher.py::test_launch_cluster_scheduling_policy_smoke[2] SKIPPED
=========== 10 passed, 3 skipped, 61 deselected, 1 warning in 0.74s ============

Launch smokes skip on CC < 9.0, same as the L4 CI jobs. Pod has been deleted.

Full pod log (nvidia-smi + pytest; apt/pip noise omitted)

Thu Aug 27 21:56:33 2026
NVIDIA-SMI 570.195.03   Driver Version: 570.195.03   CUDA Version: 12.8
GPU: NVIDIA RTX 2000 Ada Generation

HEAD: 39c9fc0 docs(cuda.core): document ClusterSchedulingPolicyType in API RST
Fetching mini CUDA toolkit 13.3.0...
#define CUDA_VERSION 13030
cuda.core 0.0.0

tests/test_api_docs_consistency.py::test_subpackage_exports_are_documented[typing] PASSED
======================= 1 passed, 5 deselected in 1.10s ========================

compute_capability ComputeCapability(major=8, minor=9)
10 passed, 3 skipped, 61 deselected, 1 warning in 0.74s

Thank you.

lijinf2 · 2026-08-28T00:19:09Z

/ok to test 39c9fc0

lijinf2 · 2026-08-28T00:23:30Z

@atiaomar1978-hub Hey, there are 7 test cases, and it looks like some are redundant. Could we remove redundancy and merge them into at most 3 cases to reduce code maintenance overhead going forward?

atiaomar1978-hub · 2026-08-28T00:25:17Z

I will do now @lijinf2

Merge seven overlapping LaunchConfig cluster-scheduling tests into mapping, rejection, and Hopper launch smoke to cut maintenance. Signed-off-by: Omar Atie <atiaomar1978@gmail.com>

atiaomar1978-hub · 2026-08-28T00:28:19Z

@lijinf2 folded the seven cluster-policy tests into three in 1a78ad5cd1 (Verified):

test_launch_config_cluster_scheduling_policy — ctor / getter-setter / driver enum / native attrs for DEFAULT, SPREAD, LOAD_BALANCING (including cluster + policy together)
test_launch_config_cluster_scheduling_policy_rejected — invalid value and CC < 9.0
test_launch_cluster_scheduling_policy_smoke — launch() for all three policies on Hopper+

Could you /ok to test 1a78ad5cd1cf0294d91994b0214ec71bdf9bdd2f when you have a moment?

lijinf2 · 2026-08-28T00:39:54Z

/ok to test 1a78ad5

CUDA 13.0.2's CUlaunchAttributeValue setter reads .value, so a raw int breaks native LaunchConfig conversion in CI. Signed-off-by: Omar Atie <atiaomar1978@gmail.com>

atiaomar1978-hub · 2026-08-28T01:49:12Z

@lijinf2 CUDA 13.0.2 jobs failed on test_launch_config_cluster_scheduling_policy (AttributeError — that setter reads .value, so a raw int breaks). 3a90298eb1 passes CUclusterSchedulingPolicy instead.

RunPod replay (RTX 2000 Ada, bindings 13.0.3): mapping + reject PASSED; smoke SKIPPED on CC 8.9 (already passed on CI H100).

Could you please /ok to test 3a90298eb1e788047e702076b81dcdb28073e699?

…-2629-signed Signed-off-by: Omar Atie <atiaomar1978@gmail.com>

lijinf2 · 2026-08-28T18:57:41Z

/ok to test 3a90298

copy-pr-bot · 2026-08-28T18:57:45Z

/ok to test 3a90298

@lijinf2, there was an error processing your request: E2

See the following link for more information: https://docs.gha-runners.nvidia.com/cpr/e/2/

lijinf2 · 2026-09-03T20:16:05Z

            Whether to allow programmatic stream serialization / PDL (default: False)
+        cluster_scheduling_policy_preference : str, optional
+            Cluster scheduling policy for the launch: ``"DEFAULT"``,
+            ``"SPREAD"``, or ``"LOAD_BALANCING"`` (default: None)


Please fix doc to make sure None case is specific (not used or DEFAULT?)

Also, why do we allow 'None' here? Is it different from "DEFAULT" option?

Done in f80f638 — docstrings now spell this out:

None (default): omit CU_LAUNCH_ATTRIBUTE_CLUSTER_SCHEDULING_POLICY_PREFERENCE; the driver uses the kernel function's default policy.

"DEFAULT": emit the launch attribute with CU_CLUSTER_SCHEDULING_POLICY_DEFAULT.

They are intentionally different: omission leaves the attribute unset; "DEFAULT" sets it explicitly (useful when overriding a kernel-level non-default policy at launch time).

Yes — None and "DEFAULT" differ:

None: no launch attribute is sent; the kernel function's default applies.

"DEFAULT": the launch attribute is sent with CU_CLUSTER_SCHEDULING_POLICY_DEFAULT.

Both are documented in f80f638.

lijinf2 · 2026-09-09T20:40:12Z

+    "DEFAULT": driver.CUclusterSchedulingPolicy.CU_CLUSTER_SCHEDULING_POLICY_DEFAULT,
+    "SPREAD": driver.CUclusterSchedulingPolicy.CU_CLUSTER_SCHEDULING_POLICY_SPREAD,
+    "LOAD_BALANCING": driver.CUclusterSchedulingPolicy.CU_CLUSTER_SCHEDULING_POLICY_LOAD_BALANCING,
+}


Let's move this to be a static member variable under class LaunchConfig

Done in f80f638 — _CLUSTER_SCHED_POLICY_TO_DRIVER is now a class attribute on LaunchConfig; validation and driver lookup use LaunchConfig._CLUSTER_SCHED_POLICY_TO_DRIVER.

lijinf2 · 2026-09-09T20:42:59Z

+    if config.cluster_scheduling_policy_preference is not None:
+        attr = driver.CUlaunchAttribute()
+        attr.id = driver.CUlaunchAttributeID.CU_LAUNCH_ATTRIBUTE_CLUSTER_SCHEDULING_POLICY_PREFERENCE
+        # 13.0.2 setter reads .value; pass FastEnum, not a raw int.


Let's remove the comment. Code already explains itself.

Done in f80f638 — removed that comment from the cpdef conversion path.

lijinf2 · 2026-09-09T20:56:14Z

@atiaomar1978-hub I just sent out the suggested edits/comments. Other than that, the PR looks in good shape. Thanks!

Move _CLUSTER_SCHED_POLICY_TO_DRIVER onto LaunchConfig, clarify None versus "DEFAULT" in docs, restore cluster-only CC check in __init__, and drop the redundant cpdef conversion comment. Signed-off-by: Omar Atie <atiaomar1978@gmail.com>

…-2629-signed Integrate upstream LaunchConfig priority support (NVIDIA#2706) with cluster scheduling policy while preserving PR commit history. Signed-off-by: Omar Atie <atiaomar1978@gmail.com>

atiaomar1978-hub · 2026-09-09T21:28:19Z

@lijinf2 — addressed the latest review on f80f638991 (no force-push; merge commit preserves history):

_CLUSTER_SCHED_POLICY_TO_DRIVER — moved onto LaunchConfig as a class attribute.
Docstring — clarified that None omits the launch attribute (driver/kernel default), while "DEFAULT" sets the driver default explicitly via the attribute.
Comment removed — dropped the redundant cpdef conversion note.
Merged main — integrated upstream #2706 (LaunchConfig.priority) alongside cluster scheduling policy; resolved conflicts in _launch_config.* and test_object_protocols.py.

RunPod (H100 80GB HBM3, CC 9.0, driver 580, CTK 13.3 / bindings 13.3.1):

3 passed — test_launch_config_cluster_scheduling_policy
         test_launch_config_cluster_scheduling_policy_rejected
         test_launch_cluster_scheduling_policy_smoke

Could you please re-review and trigger CI with /ok to test f80f638991?

lijinf2

LGTM. thanks!

lijinf2 · 2026-09-09T21:37:19Z

/ok to test f80f638

Align _launch_config.pyi with stubgen-pyx 0.2.22 after moving the cluster policy map onto LaunchConfig. Signed-off-by: Omar Atie <atiaomar1978@gmail.com>

atiaomar1978-hub · 2026-09-09T22:11:02Z

/ok to test f80f638

@lijinf2 could you re-run the CI build ?

lijinf2 · 2026-09-09T23:28:29Z

/ok to test a44088c

lijinf2 · 2026-09-09T23:30:29Z

@atiaomar1978-hub pre-commit check is still failing. CI needs that passed to let pr merge.

Use the repo-relative stubgen source path so pre-commit stubgen-pyx does not rewrite _launch_config.pyi on every platform. Signed-off-by: Omar Atie <atiaomar1978@gmail.com>

atiaomar1978-hub · 2026-09-09T23:36:36Z

@lijinf2 pre-commit is green now

lijinf2 · 2026-09-10T17:15:53Z

/ok to test b33856a

Align test_launch_config_cluster_rejects_pre_hopper_cc with the cluster-only CUDAError text restored per review feedback. Signed-off-by: Omar Atie <atiaomar1978@gmail.com>

atiaomar1978-hub · 2026-09-10T18:11:38Z

@lijinf2 — latest fixes are on d8253206e9:

b33856a40b — fix stubgen header path for CI pre-commit
d8253206e9 — match cluster pre-Hopper error message on main

RunPod verification (H100 80GB HBM3, CC 9.0, driver 580, CTK 13.3): 4/4 passed

Test	Result
`test_launch_config_cluster_rejects_pre_hopper_cc`	PASSED
`test_launch_config_cluster_scheduling_policy`	PASSED
`test_launch_config_cluster_scheduling_policy_rejected`	PASSED
`test_launch_cluster_scheduling_policy_smoke`	PASSED

Full RunPod log (runpod-cluster-policy-20260910T174049Z.log)

Thu Sep 10 17:40:51 2026       
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 580.126.09             Driver Version: 580.126.09     CUDA Version: 13.0     |
+-----------------------------------------+------------------------+----------------------+
| GPU  Name                 Persistence-M | Bus-Id          Disp.A | Volatile Uncorr. ECC |
| Fan  Temp   Perf          Pwr:Usage/Cap |           Memory-Usage | GPU-Util  Compute M. |
|                                         |                        |               MIG M. |
|=========================================+========================+======================|
|   0  NVIDIA H100 80GB HBM3          On  |   00000000:AB:00.0 Off |                    0 |
| N/A   32C    P0             74W /  700W |       0MiB /  81559MiB |      0%      Default |
|                                         |                        |             Disabled |
+-----------------------------------------+------------------------+----------------------+

+-----------------------------------------------------------------------------------------+
| Processes:                                                                              |
|  GPU   GI   CI              PID   Type   Process name                        GPU Memory |
|        ID   ID                                                               Usage      |
|=========================================================================================|
|  No running processes found                                                             |
+-----------------------------------------------------------------------------------------+
HEAD: d825320 test(cuda.core): match cluster pre-Hopper error message on main
Fetching mini CUDA toolkit 13.3.0...
#define CUDA_VERSION 13030
cuda.bindings (build) 13.3.1
cuda.core 0.0.0
cuda.bindings (test) 13.3.1

############ CLUSTER SCHEDULING POLICY TESTS ############
compute_capability ComputeCapability(major=9, minor=0)
============================= test session starts ==============================
platform linux -- Python 3.11.10, pytest-9.1.1, pluggy-1.6.0 -- /usr/bin/python3
cachedir: .pytest_cache
rootdir: /workspace/cuda-python/cuda_core
configfile: pytest.ini
plugins: anyio-4.6.0
collecting ... collected 4 items

tests/test_launcher.py::test_launch_config_cluster_rejects_pre_hopper_cc PASSED [ 25%]
tests/test_launcher.py::test_launch_config_cluster_scheduling_policy PASSED [ 50%]
tests/test_launcher.py::test_launch_config_cluster_scheduling_policy_rejected PASSED [ 75%]
tests/test_launcher.py::test_launch_cluster_scheduling_policy_smoke PASSED [100%]

============================= slowest 20 durations =============================
0.72s setup    tests/test_launcher.py::test_launch_config_cluster_rejects_pre_hopper_cc
0.34s call     tests/test_launcher.py::test_launch_cluster_scheduling_policy_smoke
0.20s setup    tests/test_launcher.py::test_launch_cluster_scheduling_policy_smoke
0.02s teardown tests/test_launcher.py::test_launch_cluster_scheduling_policy_smoke

(8 durations < 0.005s hidden.  Use -vv to show these durations.)
============================== 4 passed in 4.96s ===============================

Could you trigger the CI build on this commit?

/ok to test d825320

atiaomar1978-hub · 2026-09-11T19:54:26Z

@lijinf2 could you please trigger the CI build and Tests when you have chance ?

lijinf2 · 2026-09-11T21:42:42Z

-        a device that does not support multiple stream priorities. A
-        nonzero value outside this range raises :class:`ValueError`.
-        When omitted (or 0), the launch uses the stream's priority.
-    """


Why the comment section got deleted?

lijinf2 · 2026-09-11T21:53:02Z

@atiaomar1978-hub I need to address the CI trigger overhead issue in this PR and going forward.

I've now triggered /ok to test 8 times on this PR, if not calculating wrongly, and you're requesting a 9th.

Each /ok to test generates a notification email for me, and the back-and-forth adds unnecessary latency to the review cycle for both of us. Going forward, please ensure:

(1) All local CI checks pass before asking me to trigger CI. That includes pre-commit, stubgen, and the docs-consistency test — the last several triggers were caused by issues that were catchable locally.

(2) Batch fixes together. If there are multiple outstanding issues, address them all in one commit and ask for a single CI run, rather than one trigger per fix.

I'm happy to trigger CI when the code is genuinely ready, but I'd like to keep this to under 5 runs per PR going forward.

lijinf2 · 2026-09-11T21:57:24Z

@atiaomar1978-hub One more thing: after I approved on Sep 9, additional commits came in that deleted an existing docstring section in _launch_config.pyi. This is a regression introduced post-approval.

Looking at the commit history of this PR more broadly, the pattern is clear — functions scoped at the wrong level, duplicated test logic, imprecise types, docstrings that don't match runtime behavior — these are all consistent with agent-generated code that wasn't reviewed by a human before being pushed.

Please do a manual diff review before every push. I shouldn't be the first human to read your commits. If you're using an AI assistant, treat its output as a draft, not a final submission.

newly pushed commits deleted the docstring section which should not be deleted.

Move the class docstring above _CLUSTER_SCHED_POLICY_TO_DRIVER so stubgen-pyx emits the full Note/Attributes section in _launch_config.pyi. Signed-off-by: Omar Atie <atiaomar1978@gmail.com>

atiaomar1978-hub · 2026-09-11T22:18:23Z

@lijinf2 let us run the CI build ? still in learning curve. I understand to check the code before submission.

lijinf2 · 2026-09-11T23:47:59Z

/ok to test d0926c8

atiaomar1978-hub · 2026-09-12T01:13:33Z

@lijinf2 — CI on d0926c8c2d (run 34659404263) is red, but the failures look unrelated to this PR. Could you take a look?

PR-related checks — green

pre-commit.ci — passed
All Build jobs — passed
Docs / Build docs — passed
API check (cuda.core) — passed
cuda.core GPU tests — passed, including cluster scheduling policy:

Test	H100 job
`test_launch_config_cluster_rejects_pre_hopper_cc`	PASSED
`test_launch_config_cluster_scheduling_policy`	PASSED
`test_launch_config_cluster_scheduling_policy_rejected`	PASSED
`test_launch_cluster_scheduling_policy_smoke`	PASSED

(L4 jobs: cuda.core 4149 passed; cluster smoke skipped on CC < 9.0 as expected.)

Failures — appear infra / other packages

Two failure modes across the GPU matrix (~49 jobs), neither in cuda.core launch-config code:

1. cuda.bindings — cufile abort (exit 134)

Some jobs die early, e.g. linux-64 / py3.10 / CUDA 13.3 / L4:

tests/test_cufile.py::test_get_stats_l1 PASSED
Fatal Python error: Aborted
  File ".../test_cufile.py", line 1689 in test_get_stats_l2
exit code 134

2. cuda_pathfinder — cutensorMg child-process flake

Jobs that finish cuda.core then fail on pathfinder, e.g. linux-64 / py3.13 / CUDA 13.3 / H100:

FAILED tests/test_load_nvidia_dynamic_lib.py::test_load_nvidia_dynamic_lib[cutensorMg]
RuntimeError: Child process failed for libname='cutensorMg' with exit code 0
1 failed, 1608 passed, 6 skipped

Same pattern on linux-64 / py3.11 / CUDA 13.3 / L4 after cuda.core passed fully.

Ask

Are these known CI flakes on main? If so, is a re-run sufficient, or is there anything you need from this PR side?

Local pre-commit (including stubgen) passed on d0926c8c2d; RunPod H100 verification on the prior commit was 4/4 for the cluster-policy suite.

github-actions Bot added the cuda.core Everything related to the cuda.core module label Aug 26, 2026

atiaomar1978-hub and others added 4 commits August 25, 2026 17:19

atiaomar1978-hub force-pushed the feat/launch-config-cluster-scheduling-policy-2629-signed branch from b00cd6f to c6c283b Compare August 26, 2026 00:19

This was referenced Aug 26, 2026

[FEA]: Support CU_LAUNCH_ATTRIBUTE_CLUSTER_SCHEDULING_POLICY_PREFERENCE in LaunchConfig #2629

Open

feat(cuda.core): add cluster scheduling policy to LaunchConfig #2702

Closed

lijinf2 assigned atiaomar1978-hub Aug 27, 2026

lijinf2 added the feature New feature or request label Aug 27, 2026

lijinf2 added this to the cuda.core next milestone Aug 27, 2026

docs(cuda.core): document ClusterSchedulingPolicyType in API RST

39c9fc0

CI test_subpackage_exports_are_documented[typing] failed because the new typing export was missing from docs/source/*.rst. Signed-off-by: Omar Atie <atiaomar1978@gmail.com>

test(cuda.core): collapse cluster policy tests to three cases

1a78ad5

Merge seven overlapping LaunchConfig cluster-scheduling tests into mapping, rejection, and Hopper launch smoke to cut maintenance. Signed-off-by: Omar Atie <atiaomar1978@gmail.com>

fix(cuda.core): pass cluster policy FastEnum to 13.0.2 bindings

3a90298

CUDA 13.0.2's CUlaunchAttributeValue setter reads .value, so a raw int breaks native LaunchConfig conversion in CI. Signed-off-by: Omar Atie <atiaomar1978@gmail.com>

Merge branch 'main' into feat/launch-config-cluster-scheduling-policy…

db871e5

…-2629-signed Signed-off-by: Omar Atie <atiaomar1978@gmail.com>

lijinf2 reviewed Aug 28, 2026

View reviewed changes

Comment thread cuda_core/cuda/core/_launch_config.pxd Outdated

Comment thread cuda_core/cuda/core/_launch_config.pyi Outdated

Comment thread cuda_core/cuda/core/_launch_config.pyi Outdated

Comment thread cuda_core/cuda/core/_launch_config.pyi Outdated

lijinf2 requested changes Sep 9, 2026

View reviewed changes

atiaomar1978-hub added 2 commits September 9, 2026 14:18

Merge branch 'main' into feat/launch-config-cluster-scheduling-policy…

f80f638

…-2629-signed Integrate upstream LaunchConfig priority support (NVIDIA#2706) with cluster scheduling policy while preserving PR commit history. Signed-off-by: Omar Atie <atiaomar1978@gmail.com>

lijinf2 previously approved these changes Sep 9, 2026

View reviewed changes

lijinf2 enabled auto-merge (squash) September 9, 2026 21:37

chore(cuda.core): regenerate LaunchConfig stubs for pre-commit

a44088c

Align _launch_config.pyi with stubgen-pyx 0.2.22 after moving the cluster policy map onto LaunchConfig. Signed-off-by: Omar Atie <atiaomar1978@gmail.com>

auto-merge was automatically disabled September 9, 2026 22:09
Head branch was pushed to by a user without write access

chore(cuda.core): fix LaunchConfig stub header path for CI

b33856a

Use the repo-relative stubgen source path so pre-commit stubgen-pyx does not rewrite _launch_config.pyi on every platform. Signed-off-by: Omar Atie <atiaomar1978@gmail.com>

lijinf2 enabled auto-merge (squash) September 10, 2026 17:16

test(cuda.core): match cluster pre-Hopper error message on main

d825320

Align test_launch_config_cluster_rejects_pre_hopper_cc with the cluster-only CUDAError text restored per review feedback. Signed-off-by: Omar Atie <atiaomar1978@gmail.com>

auto-merge was automatically disabled September 10, 2026 17:39
Head branch was pushed to by a user without write access

lijinf2 reviewed Sep 11, 2026

View reviewed changes

lijinf2 self-requested a review September 11, 2026 22:00

docs(cuda.core): restore LaunchConfig class docstring in stubs

d0926c8

Move the class docstring above _CLUSTER_SCHED_POLICY_TO_DRIVER so stubgen-pyx emits the full Note/Attributes section in _launch_config.pyi. Signed-off-by: Omar Atie <atiaomar1978@gmail.com>

Uh oh!

Conversation

atiaomar1978-hub commented Aug 26, 2026

Description

Checklist

Test plan

Uh oh!

copy-pr-bot Bot commented Aug 26, 2026

Uh oh!

atiaomar1978-hub commented Aug 26, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

lijinf2 commented Aug 27, 2026

Uh oh!

github-actions Bot commented Aug 27, 2026

Preview will be ready when the GitHub Pages deployment is complete.

Uh oh!

atiaomar1978-hub commented Aug 27, 2026

Uh oh!

atiaomar1978-hub commented Aug 27, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Why CI failed

Why it works again

RunPod replay (pod 7xp9vqqo44kk1s, RTX 2000 Ada, CC 8.9)

Uh oh!

lijinf2 commented Aug 28, 2026

Uh oh!

lijinf2 commented Aug 28, 2026

Uh oh!

atiaomar1978-hub commented Aug 28, 2026

Uh oh!

atiaomar1978-hub commented Aug 28, 2026

Uh oh!

lijinf2 commented Aug 28, 2026

Uh oh!

atiaomar1978-hub commented Aug 28, 2026

Uh oh!

lijinf2 commented Aug 28, 2026

Uh oh!

copy-pr-bot Bot commented Aug 28, 2026

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

lijinf2 commented Sep 9, 2026

Uh oh!

atiaomar1978-hub commented Sep 9, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

lijinf2 left a comment

Choose a reason for hiding this comment

Uh oh!

lijinf2 commented Sep 9, 2026

Uh oh!

atiaomar1978-hub commented Sep 9, 2026

Uh oh!

lijinf2 commented Sep 9, 2026

Uh oh!

lijinf2 commented Sep 9, 2026

Uh oh!

atiaomar1978-hub commented Sep 9, 2026

Uh oh!

lijinf2 commented Sep 10, 2026

atiaomar1978-hub commented Aug 26, 2026 •

edited

Loading

atiaomar1978-hub commented Aug 27, 2026 •

edited

Loading

RunPod replay (pod `7xp9vqqo44kk1s`, RTX 2000 Ada, CC 8.9)

atiaomar1978-hub commented Sep 9, 2026 •

edited

Loading

lijinf2 Sep 11, 2026 •

edited

Loading