Perf: remove clone on uninitiated_partitions in SortPreservingMergeStream
#15562
+8
−8
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
… custom fixed size queue this is done to avoid having to clone or change the actual values
|
I noticed something similar when profiling some click bench queries locally. I am going to take a look at this one |
alamb
reviewed
Apr 3, 2025
| /// ```plain | ||
| /// index: | 0 | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | | ||
| /// | -------------------------------------------------------- | | ||
| /// value: | 0 | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | |
There was a problem hiding this comment.
I would find these examples easier to follow if index and value didn't have the same values. maybe something like
Suggested change
| /// value: | 0 | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | | |
| /// value: | 9 | 8 | 7 | 5 | 5 | 4 | 3 | 2 | 1 | 0 | |
|
FYI @jayzhan211 |
|
BTW I filed a ticket to track this as I have seen it too |
|
I removed the custom fixed size queue and replaced with matching vec deque functions that are 0-cost as it's just index manipulation like my prev impl |
|
🤖 |
|
🤖: Benchmark completed Details
|
|
🤖 |
|
🤖: Benchmark completed Details
|
|
Interesting - I wonder if this is something related to how sampling is done with |
|
My current intuition about it is twofold:
|
Dandandan
approved these changes
Apr 4, 2025
I agree with this assesment -- so we save CPU work with this change but the total query time doesn't really decrease because we are not fully using all of the cores I also have found Samply super awesome -- thank you @comphead for showing that Something else I have observed when looking at Samply is that on my laptop at least, it appears there are several times where processing stalls due to parsing parquet metadata:
I was running this It would be an interesting experiment to see if we can improve performance by caching the metadata. I will file a ticket to investigate this |
|
Dandandan
changed the title
merge uninitiated_partitions VecDeque<usize> with custom fixed size queuemerge uninitiated_partitions VecDeque<usize>
Dandandan
changed the title
merge uninitiated_partitions VecDeque<usize>clone on uninitiated_partitions in SortPreservingMergeStream
|
Thanks @rluvaton |
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.
Which issue does this PR close?
Related to:
Rationale for this change
I saw in my code that clone and pop front and push takes some of the time from
poll_next_innerThis is from my sampling
Before:
After:
What changes are included in this PR?
FixedSizeQueuewhich avoid cloning the vector (could be avoided even without introducing this)Are these changes tested?
Existing tests
Are there any user-facing changes?
No