perf: combine_hashes for byte-view multi-column rehash#22352
Closed
Dandandan wants to merge 1 commit into
Closed
Conversation
Aligns the byte-view paths (Utf8View / BinaryView) in hash_string_view_array_inner and hash_generic_byte_view_array's no-buffers rehash branch with the pattern already used by the generic variable-width path (hash_array) and the dictionary path (hash_dictionary_array): hash the current value once with the query RandomState and combine it with the previous row hash via combine_hashes, instead of constructing a fresh foldhash::SeedableRandomState per row that folds the seed into a new hasher state. Measured on ClickBench q16 and q17 (`GROUP BY UserID, SearchPhrase`, partitioned dataset, dfbench, current main baselines 1106.69 / 1127.64 ms over 3 iterations): patched 3-iteration runs settled around 986.76 ms (q16, -10.8%) and 1002.29 ms (q17, -11.1%). Validation: - cargo test -p datafusion-common hash_utils - cargo clippy -p datafusion-common --all-targets --all-features -- -D warnings
Contributor
Author
|
run benchmarks |
|
🤖 Benchmark running (GKE) | trigger CPU Details (lscpu)Comparing perf/byte-view-rehash-multi-column (d48d1a4) to c8b784a (merge-base) diff using: clickbench_partitioned File an issue against this benchmark runner |
|
🤖 Benchmark running (GKE) | trigger CPU Details (lscpu)Comparing perf/byte-view-rehash-multi-column (d48d1a4) to c8b784a (merge-base) diff using: tpcds File an issue against this benchmark runner |
|
🤖 Benchmark running (GKE) | trigger CPU Details (lscpu)Comparing perf/byte-view-rehash-multi-column (d48d1a4) to c8b784a (merge-base) diff using: tpch File an issue against this benchmark runner |
|
🤖 Benchmark completed (GKE) | trigger Instance: CPU Details (lscpu)Details
Resource Usagetpch — base (merge-base)
tpch — branch
File an issue against this benchmark runner |
|
🤖 Benchmark completed (GKE) | trigger Instance: CPU Details (lscpu)Details
Resource Usagetpcds — base (merge-base)
tpcds — branch
File an issue against this benchmark runner |
|
🤖 Benchmark completed (GKE) | trigger Instance: CPU Details (lscpu)Details
Resource Usageclickbench_partitioned — base (merge-base)
clickbench_partitioned — branch
File an issue against this benchmark runner |
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
2 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
When multiple group keys participate in a grouped aggregate,
create_hasheswalks the columns left-to-right and rehashes each row by combining the current value's hash with the previously accumulated row hash. The generic variable-width path (hash_array) and the dictionary path (hash_dictionary_array) already do this withcombine_hashes(value.hash_one(random_state), *hash). The byte-view paths (hash_string_view_array_innerand the no-buffers rehash arm ofhash_generic_byte_view_array) instead construct a freshfoldhash::fast::SeedableRandomStateper row, seeded from the previous hash, and run the value through that. That approach is heavier on the hot loop and isn't necessary for correctness — the same hash distribution is achievable withcombine_hashes.The motivating workloads are ClickBench q16 and q17, which both group by
UserID, SearchPhrase.SearchPhraseis aUtf8View, so the second column rehashes throughhash_string_view_array_innerfor every row. On current-main baselines of 1106.69 ms / 1127.64 ms (dfbench clickbench --query 16 --iterations 3/--query 17 --iterations 3, partitioned dataset), 3-iteration runs with this change settled around:Guardrail single-query runs of other byte-view second-column rehash workloads (q14, q18, q39) did not regress.
This change is independent of, and stackable with, #22350 (the same idiom applied to
hash_array_primitive). After both land,seeded_statebecomes unused and can be removed in a small follow-up.What changes are included in this PR?
Replaces the three
seeded_state(...).build_hasher() / value.hash_write(...) / hasher.finish()blocks in the byte-view rehash paths:hash_string_view_array_inner: inlined-view rehash branch and out-of-line bytes rehash branchhash_generic_byte_view_array:(no nulls, no buffers, rehash)specializationEach is replaced with
combine_hashes(value.hash_one(random_state), *hash), matching the surrounding generic and dictionary paths.seeded_stateitself andhash_array_primitiveare unchanged.Are these changes tested?
Covered by existing tests in
datafusion-common::hash_utils, including the multi-column / string-view tests:cargo test -p datafusion-common hash_utilscargo clippy -p datafusion-common --all-targets --all-features -- -D warningsAre there any user-facing changes?
No. This is an internal performance change with no API or behavioral impact — hash distribution properties are unchanged because
combine_hashesis already the standard pattern in this file.