Misc hash / hash aggregation performance improvements #19910
Merged
+26
−21
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
Contributor
Author
|
run benchmarks |
github-actions
bot
added
physical-expr
Dandandan
changed the title
Dandandan
changed the title
|
🤖 |
Dandandan
commented
Jan 20, 2026
| let entry = self.map.find_mut(hash, |header| { | ||
| // compare value if hashes match | ||
| if header.len != value_len { | ||
| if header.hash != hash { |
Contributor
Author
There was a problem hiding this comment.
Comment says "compare value if hashes match" but actual implementation is comparing lengths 🤔
Contributor
There was a problem hiding this comment.
I am double checking the reasoning, if the hash values aren't the same we know the values can not be the same either due to the requirements of the Hash trait
However (as codex points out to me) there is some small subtle chance that if two different values collide on hash, and their inline encodings match, they will be treated as equal even though the byte sequences differ which is what the old len check avoided
So I think we still need the old length checks too 🤔
Dandandan
commented
Jan 20, 2026
| let entry = self.map.find_mut(hash, |header| { | ||
| // compare value if hashes match | ||
| if header.len != value_len { | ||
| if header.hash != hash { |
Contributor
Author
There was a problem hiding this comment.
This should save some random access if a lot of values share the same length
Contributor
There was a problem hiding this comment.
I think we also need to update this place to check header.len
Dandandan
commented
Jan 20, 2026
| let v = self.builder.get_value(header.view_idx); | ||
|
|
||
| if v.len() != value.len() { | ||
| if header.hash != hash { |
Contributor
Author
There was a problem hiding this comment.
This should save some random access if a lot of values share the same length
Contributor
Author
|
run benchmark sql_aggregation |
|
🤖 Hi @Dandandan, thanks for the request (#19910 (comment)).
Please choose one or more of these with |
Contributor
Author
|
run benchmark aggregate_query_sql aggregate_vectorized |
|
🤖: Benchmark completed Details
|
Dandandan
changed the title
Contributor
Author
|
run benchmarks |
Dandandan
changed the title
|
🤖 |
|
🤖: Benchmark completed Details
|
|
🤖 |
|
Benchmark script failed with exit code 101. Last 10 lines of output: Click to expand |
|
🤖 |
|
🤖: Benchmark completed Details
|
Contributor
Author
|
run benchmarks |
|
🤖 |
|
🤖: Benchmark completed Details
|
Contributor
Author
|
Unfortunately the benchmark runner is very noisy lately 😢 |
Contributor
Author
|
run benchmark tpch tpcds |
|
🤖 |
|
🤖: Benchmark completed Details
|
|
🤖 |
|
🤖: Benchmark completed Details
|
Contributor
Author
|
run benchmark tpch_mem |
|
🤖 |
|
🤖: Benchmark completed Details
|
alamb
reviewed
Jan 23, 2026
| hash, | ||
| |&(g, _)| unsafe { self.values.get_unchecked(g).is_eq(key) }, | ||
| |&(g, h)| unsafe { | ||
| hash == h && self.values.get_unchecked(g).is_eq(key) |
Contributor
There was a problem hiding this comment.
I double checked that Hashbrown says
https://docs.rs/hashbrown/latest/hashbrown/struct.HashTable.html#method.entry
This method will call eq for all entries with the given hash, but may also call it for entries with a different hash. eq should only return true for the desired entry, at which point the search is stopped.
So checking the hash first makes senes to me as it will avoid a memory access
alamb
approved these changes
Jan 23, 2026
datafusion/common/src/utils/proxy.rs
Outdated
| self.reserve(bump_elements, &hasher); | ||
| } | ||
|
|
||
| // still need to insert the element since first try failed |
Contributor
There was a problem hiding this comment.
I don't see a first attempt to insert -- maybe this comment needs to be updated
Contributor
|
Thank you @Dandandan |
alamb
changed the title
Dandandan
commented
Jan 23, 2026
| if header.hash != hash { | ||
| return false; | ||
| } | ||
| let v = self.builder.get_value(header.view_idx); |
Contributor
Author
There was a problem hiding this comment.
I saw in profiles this is pretty expensive (also self.builder.append_value and always comparing by bytes - I think we are not really using the potential of binary views here, I opened #19961
Contributor
Author
|
run benchmarks |
|
🤖 |
|
🤖: Benchmark completed Details
|
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
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?
Rationale for this change
Just a couple of optimizations for hash table lookups usage in hash aggregate.
What changes are included in this PR?
Are these changes tested?
Are there any user-facing changes?