fix: ensure cache affinity in Mod partition shuffle #19309
Draft
+29
−16
Conversation
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
The previous implementation of PartitionsShuffleKind::Mod broke cache affinity even when cluster membership remained unchanged. It first sorted partitions by hash % num_executors, then redistributed them evenly by partition count. This meant that when table data changed (segments added or removed), the redistribution cut points would shift, causing the same segment to be assigned to different executors. For example, with 7 segments (indices 0-6) and 2 executors, the sorted partitions would be split as indices 0,1,2 to executor0 and indices 3,4,5,6 to executor1. When a new segment is added (now 8 segments), the split becomes indices 0,1,2,3 to executor0 and indices 4,5,6,7 to executor1, moving the segment originally at index 3 from executor1 to executor0 and invalidating its cache. This fix removes the redundant redistribution step and directly assigns partitions to executors based on hash % num_executors. Now the same partition always goes to the same executor as long as cluster membership is unchanged, regardless of other partitions being added or removed.
Member
Author
|
@codex review |
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: e0bf4c74fa
ℹ️ About Codex in GitHub
Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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.
I hereby agree to the terms of the CLA available at: https://docs.databend.com/dev/policies/cla/
Summary
The previous implementation of PartitionsShuffleKind::Mod broke cache affinity even when cluster membership remained unchanged. It first sorted partitions by hash % num_executors, then redistributed them evenly by partition count. This meant that when table data changed (segments added or removed), the redistribution cut points would shift, causing the same segment to be assigned to different executors.
For example, with 7 segments (indices 0-6) and 2 executors, the sorted partitions would be split as indices 0,1,2 to executor0 and indices 3,4,5,6 to executor1. When a new segment is added (now 8 segments), the split becomes indices 0,1,2,3 to executor0 and indices 4,5,6,7 to executor1, moving the segment originally at index 3 from executor1 to executor0 and invalidating its cache.
This fix removes the redundant redistribution step and directly assigns partitions to executors based on hash % num_executors. Now the same partition always goes to the same executor as long as cluster membership is unchanged, regardless of other partitions being added or removed.
Tests
Type of change
This change is