[FLINK-37213][table-runtime] Improve performance of unbounded OVER aggregations #26075
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
davidradl
reviewed
Jan 24, 2025
| .withDescription( | ||
| "Which version of the unbounded over aggregation to use: " | ||
| + " 1 - legacy version" | ||
| + " 2 - new version with improved performance"); |
There was a problem hiding this comment.
nit: remove the word new so it does not need to be changed if we add a 3rd value at some stage
There was a problem hiding this comment.
If this goes into the first version 2 - should we default to 2?
There was a problem hiding this comment.
I think we can default to version 2 indeed and keep this as an opt-out feature.
|
CI failing for spotless |
|
Reviewed by Chi on 23/01/2025 Go back to the submitter with review comments. |
…gregations Instead of sorting all of the records based on the row time explicilty use timers to achieve the same thing. This version vs the previous one register a timer for each record, as opposed to just one timer per key. However since we are using RocksDB for timers, this is a minor problem. In exchange, we: - don't have to iterate over all of the state for each timer - we are firing timers only when needed, vs for each watermark for each key. For example if watermarks are fire every 200ms and for a given key, we have only one record that should be fired 20s into the future, the previous version would be firing a timer for that key for each watermark unnecessarily without doing any work.
|
CI is green. Thank you for the reviews. @davidradl I have addressed your minor comments, so I'm going to merge this. In case you have some further comments please let me know, we can discuss them and address in a separate PR if needed. |
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.
Instead of sorting all of the records based on the row time explicilty, use timers to achieve the same thing.
This version vs the previous one register a timer for each record, as opposed to just one timer per key. However since we are using RocksDB for timers, this is a minor problem. In exchange, we:
Verifying this change
This is covered by various existing tests, that were parametrised to check for both old and new code paths.
Does this pull request potentially affect one of the following parts:
@Public(Evolving): (yes / no)Documentation