[REVIEW] Add scheduler_file argument to support MNMG setup #1593
Conversation
This file contains 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
cjnolet
added
improvement
VibhuJawa
commented
Jun 13, 2023
Comment on lines
-27
to
-40
| initialize.initialize( | ||
| create_cuda_context=True, | ||
| enable_tcp_over_ucx=enable_tcp_over_ucx, | ||
| enable_nvlink=enable_nvlink, | ||
| enable_infiniband=enable_infiniband, | ||
| ) | ||
| cluster = LocalCUDACluster( | ||
| protocol="ucx", | ||
| enable_tcp_over_ucx=enable_tcp_over_ucx, | ||
| enable_nvlink=enable_nvlink, | ||
| enable_infiniband=enable_infiniband, | ||
| ) | ||
| yield cluster | ||
| cluster.close() |
There was a problem hiding this comment.
This is no longer needed: https://docs.rapids.ai/api/dask-cuda/nightly/examples/ucx/#localcudacluster-with-automatic-configuration
VibhuJawa
changed the title
VibhuJawa
commented
Jun 13, 2023
| yield client | ||
| client.close() | ||
|
|
||
|
|
||
| @pytest.fixture() | ||
| def ucx_client(ucx_cluster): | ||
| client = Client(cluster) | ||
| client = create_client(ucx_cluster) |
There was a problem hiding this comment.
I think this was a bug previously where we passed cluster instead of ucx_cluster .
Will revert if this was intentional.
VibhuJawa
commented
Jun 13, 2023
There was a problem hiding this comment.
Needed to add this as we run tests on workers and scheduler and this module needs to be importable for that to happen in dask when running on a non local cluster.
VibhuJawa
changed the title
VibhuJawa
changed the title
|
/merge |
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Outdated suggestions cannot be applied.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.
Add scheduler_file argument to support MNMG setup
Overview:
The primary goal is to provide more flexibility and adaptability in how the Dask cluster for testing is configured.
Changes:
LocalCUDAClusterinstances is now contingent on the presence of aSCHEDULER_FILEenvironment variable. If this variable exists, the path to the Dask scheduler file is returned instead of creating a new cluster. This change allows the use of pre-existing clusters specified via theSCHEDULER_FILEenvironment variable.enable_tcp_over_ucx,enable_nvlink,enable_infiniband) previously used to initialize theLocalCUDACluster. This is because sinceDask-CUDA 22.02andUCX >= 1.11.1we dont need those.See docs: https://docs.rapids.ai/api/dask-cuda/nightly/examples/ucx/#localcudacluster-with-automatic-configuration
This could help in situations where test scenarios need to be conducted on a specific pre-existing cluster (especially for MNMG setups) .
Testing:
I tested using the following setup:
Start Cluster:
Run Tests: