Fix test_cuda_device_order on some multi-GPU systems#1590
Merged
mdboom merged 3 commits intoNVIDIA:mainfrom Feb 9, 2026
Merged
Fix test_cuda_device_order on some multi-GPU systems#1590mdboom merged 3 commits intoNVIDIA:mainfrom
mdboom merged 3 commits intoNVIDIA:mainfrom
Conversation
Contributor
|
Auto-sync is disabled for ready for review pull requests in this repository. Workflows must be run manually. Contributors can view more details about this message here. |
mdboom
added
test
Contributor
Author
|
/ok to test |
Contributor
There was a problem hiding this comment.
Pull request overview
Adjusts the NVML/CUDA device-order test to avoid failures on multi-GPU systems where device visibility differs depending on CUDA_VISIBLE_DEVICES.
Changes:
- Updates
test_cuda_device_orderto accept themonkeypatchfixture. - Deletes
CUDA_VISIBLE_DEVICESduring the test run before querying CUDA/NVML devices.
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
This comment has been minimized.
This comment has been minimized.
Contributor
Author
|
/ok to test |
rwgk
reviewed
Feb 9, 2026
| # and each of them should still be found in NVML devices. | ||
| assert len(cuda_devices) <= len(nvml_devices) | ||
| for cuda_device in cuda_devices: | ||
| assert cuda_device in nvml_devices, f"CUDA device {cuda_device} not found in NVML device list" |
Contributor
There was a problem hiding this comment.
Give me a sec to experiment: does the f-string here suppress the helpful pytest default behavior or not?
Contributor
There was a problem hiding this comment.
This is great as-is.
I hacked the test so that it fails on my workstation. This is the diff with/OUT the f-string:
(TestVenv) smc120-0009.ipp2a2.colossus.nvidia.com:/wrk/forked/cuda-python/cuda_bindings $ diff -u $Z/withOUT_fstring $Z/with_fstring
--- /wrk/z/withOUT_fstring 2026-02-09 08:59:27.153944722 -0800
+++ /wrk/z/with_fstring 2026-02-09 08:59:11.085012520 -0800
@@ -2,7 +2,7 @@
platform linux -- Python 3.12.3, pytest-9.0.2, pluggy-1.6.0 -- /wrk/forked/cuda-python/TestVenv/bin/python
cachedir: .pytest_cache
benchmark: 5.2.3 (defaults: timer=time.perf_counter disable_gc=False min_rounds=5 min_time=0.000005 max_time=1.0 calibration_precision=10 warmup=False warmup_iterations=100000)
-Using --randomly-seed=1514160743
+Using --randomly-seed=3206929400
rootdir: /wrk/forked/cuda-python/cuda_bindings
configfile: pyproject.toml
plugins: repeat-0.9.4, benchmark-5.2.3, mock-3.15.1, randomly-4.0.1
@@ -25,8 +25,9 @@
# and each of them should still be found in NVML devices.
assert len(cuda_devices) <= len(nvml_devices)
for cuda_device in cuda_devices:
-> assert cuda_device not in nvml_devices
-E AssertionError: assert {'id': 75, 'name': 'NVIDIA A10G'} not in [{'id': 75, 'name': 'NVIDIA A10G'}]
+> assert cuda_device not in nvml_devices, f"CUDA device {cuda_device} not found in NVML device list"
+E AssertionError: CUDA device {'name': 'NVIDIA A10G', 'id': 75} not found in NVML device list
+E assert {'id': 75, 'name': 'NVIDIA A10G'} not in [{'id': 75, 'name': 'NVIDIA A10G'}]
cuda_device = {'id': 75, 'name': 'NVIDIA A10G'}
cuda_devices = [{'id': 75, 'name': 'NVIDIA A10G'}]
@@ -34,5 +35,5 @@
tests/nvml/test_cuda.py:67: AssertionError
=========================== short test summary info ============================
-FAILED tests/nvml/test_cuda.py::test_cuda_device_order - AssertionError: asse...
-============================== 1 failed in 0.38s ===============================
+FAILED tests/nvml/test_cuda.py::test_cuda_device_order - AssertionError: CUDA...
+============================== 1 failed in 0.41s ===============================
rwgk
approved these changes
Feb 9, 2026
Contributor
rwgk
left a comment
rwgk
left a comment
There was a problem hiding this comment.
LGTM except for the suspected possible/possibly typo.
| # and each of them should still be found in NVML devices. | ||
| assert len(cuda_devices) <= len(nvml_devices) | ||
| for cuda_device in cuda_devices: | ||
| assert cuda_device in nvml_devices, f"CUDA device {cuda_device} not found in NVML device list" |
Contributor
There was a problem hiding this comment.
This is great as-is.
I hacked the test so that it fails on my workstation. This is the diff with/OUT the f-string:
(TestVenv) smc120-0009.ipp2a2.colossus.nvidia.com:/wrk/forked/cuda-python/cuda_bindings $ diff -u $Z/withOUT_fstring $Z/with_fstring
--- /wrk/z/withOUT_fstring 2026-02-09 08:59:27.153944722 -0800
+++ /wrk/z/with_fstring 2026-02-09 08:59:11.085012520 -0800
@@ -2,7 +2,7 @@
platform linux -- Python 3.12.3, pytest-9.0.2, pluggy-1.6.0 -- /wrk/forked/cuda-python/TestVenv/bin/python
cachedir: .pytest_cache
benchmark: 5.2.3 (defaults: timer=time.perf_counter disable_gc=False min_rounds=5 min_time=0.000005 max_time=1.0 calibration_precision=10 warmup=False warmup_iterations=100000)
-Using --randomly-seed=1514160743
+Using --randomly-seed=3206929400
rootdir: /wrk/forked/cuda-python/cuda_bindings
configfile: pyproject.toml
plugins: repeat-0.9.4, benchmark-5.2.3, mock-3.15.1, randomly-4.0.1
@@ -25,8 +25,9 @@
# and each of them should still be found in NVML devices.
assert len(cuda_devices) <= len(nvml_devices)
for cuda_device in cuda_devices:
-> assert cuda_device not in nvml_devices
-E AssertionError: assert {'id': 75, 'name': 'NVIDIA A10G'} not in [{'id': 75, 'name': 'NVIDIA A10G'}]
+> assert cuda_device not in nvml_devices, f"CUDA device {cuda_device} not found in NVML device list"
+E AssertionError: CUDA device {'name': 'NVIDIA A10G', 'id': 75} not found in NVML device list
+E assert {'id': 75, 'name': 'NVIDIA A10G'} not in [{'id': 75, 'name': 'NVIDIA A10G'}]
cuda_device = {'id': 75, 'name': 'NVIDIA A10G'}
cuda_devices = [{'id': 75, 'name': 'NVIDIA A10G'}]
@@ -34,5 +35,5 @@
tests/nvml/test_cuda.py:67: AssertionError
=========================== short test summary info ============================
-FAILED tests/nvml/test_cuda.py::test_cuda_device_order - AssertionError: asse...
-============================== 1 failed in 0.38s ===============================
+FAILED tests/nvml/test_cuda.py::test_cuda_device_order - AssertionError: CUDA...
+============================== 1 failed in 0.41s ===============================
Co-authored-by: Ralf W. Grosse-Kunstleve <rwgkio@gmail.com>
Contributor
Author
|
/ok to test |
This comment has been minimized.
This comment has been minimized.
1 similar comment
|
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
The
CUDA_VISIBLE_DEVICESenvironment variable controls whether all devices are included incuDeviceGetCount, or just those that are available for CUDA compute.We should make the test adaptable to either case.