[DistPart] Fix corner case in dist partition which always led to an assertion error being triggered. #7395
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Outdated suggestions cannot be applied.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.
Description
In some edge cases where workers end up with no rows of features to send over the network (e.g. in range partition) the existing code was creating tensors of shape (0, 0) and trying to communicate/aggregate those over the network, whereas workers that had non-zero rows assigned would communicate (num_rows, feature_dimension), leading to an assertion error being triggered in
dgl/tools/distpartitioning/gloo_wrapper.py
Lines 144 to 146 in 6f2ccbf
This PR handles the 2D tensor case as a special case, ensuring the correct shape for the tensors being sent over the network, (0, feature_dimension).
This PR also removes an array that was being created but never used, reducing the memory footprint of the function.
Ping @Rhett-Ying for review
Checklist
Please feel free to remove inapplicable items for your PR.
Changes