Conversation
Signed-off-by: Niranjan Artal <nartal@nvidia.com>
Signed-off-by: Niranjan Artal <nartal@nvidia.com>
Signed-off-by: Niranjan Artal <nartal@nvidia.com>
Collaborator
Author
|
build |
Contributor
|
This branch has not been deployed
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Contributes to #15956.
Description
Enable GPU
DELETE,UPDATE, andMERGEfor Delta 4.2 OSS Unity Catalog managed tables with persistent deletion vectors, building on #16022 and #15869.This change:
Persistent-DV dynamic partition overwrite remains on CPU, consistent with the existing non-managed Delta limitation.
AI assistance disclosure: Codex assisted with issue implementation and validation. It was reviewed by the author before publication.
Validation
Performance
Single RTX A5000, local fake-S3, 10M rows: dynamic partition overwrite reached 1.23x; persistent-DV DELETE, UPDATE, and MERGE measured 0.68x, 0.43x, and 0.53x. These local runs are dominated by catalog and commit overhead and do not replace cluster-scale qualification. For context, #15869 reported GPU speedups for the underlying generic DML path on an 8-node A100 cluster. Will do another set of runs on 8-node A100 cluster.
Checklists
Documentation
Testing
Performance