perf(parquet): Optimize DeltaByteArrayEncoder and friends - #10412
Open
MassivePizza wants to merge 17 commits into
Open
perf(parquet): Optimize DeltaByteArrayEncoder and friends#10412MassivePizza wants to merge 17 commits into
MassivePizza wants to merge 17 commits into
Conversation
MassivePizza
marked this pull request as ready for review
July 23, 2026 12:18
Contributor
|
run benchmark arrow_writer |
This comment was marked as duplicate.
This comment was marked as duplicate.
|
🤖 Arrow criterion benchmark completed (GKE) | trigger Instance: CPU Details (lscpu)Details
Resource Usagebase (merge-base)
branch
File an issue against this benchmark runner |
Contributor
Author
|
That bench sure looks noisy. I'm not sure which benches actually use DeltaByteArray encoding, but parquet_2 should generally enable it for byte arrays (i.e. decimals and strings) AFAIK. I'll test again locally to make sure nothing is too broken. |
Member
|
@MassivePizza I'm having issues with flaky benches on #10432 too. Maybe we should focus on pushing #9975 forward? |
Contributor
|
run benchmark parquet_round_trip env:
BENCH_FILTER: delta |
This comment was marked as duplicate.
This comment was marked as duplicate.
|
🤖 Arrow criterion benchmark completed (GKE) | trigger Instance: CPU Details (lscpu)Details
Resource Usagebase (merge-base)
branch
File an issue against this benchmark runner |
Contributor
|
run benchmark parquet_round_trip env:
BENCH_FILTER: byte_stream |
This comment was marked as duplicate.
This comment was marked as duplicate.
|
🤖 Arrow criterion benchmark completed (GKE) | trigger Instance: CPU Details (lscpu)Details
Resource Usagebase (merge-base)
branch
File an issue against this benchmark runner |
MassivePizza
marked this pull request as draft
July 30, 2026 12:52
Contributor
|
@MassivePizza I'm trying to run benchmarks on some local hardware. A first pass looked promising, but there was too much background stuff running. I'll try again today on a beefier workstation. In general the microbenches are good at identifying large changes, but smaller signals are hard to distinguish from noise :( |
Contributor
Author
|
I also found some new changes in ByteStreamSplit coders, but will do a second PR for that. |
MassivePizza
marked this pull request as ready for review
July 30, 2026 16:16
Contributor
Author
|
Diff ended up smaller than expected, so I put it here in the end. |
MassivePizza
force-pushed
the
optimize-delta-byte-array-encoder
branch
from
August 5, 2026 14:27
2aa786d to
9168a04
Compare
MassivePizza
force-pushed
the
optimize-delta-byte-array-encoder
branch
from
August 5, 2026 14:33
9168a04 to
738f4e4
Compare
Jefffrey
reviewed
Aug 19, 2026
Jefffrey
left a comment
Jefffrey
left a comment
Contributor
There was a problem hiding this comment.
do we have some local benchmark results to demonstrate the gains?
Contributor
There was a problem hiding this comment.
does this doc need to be updated now?
Contributor
Author
There was a problem hiding this comment.
I moved part of it into decoder, should be more sensible now.
MassivePizza
force-pushed
the
optimize-delta-byte-array-encoder
branch
from
August 19, 2026 16:30
2b79b65 to
10eb9b4
Compare
Contributor
Author
|
Sorry about the label spam, accidentally merged main into the PR branch :) |
github-actions
Bot
removed
arrow
Contributor
Author
|
Biggest win can probably be attributed to removing the copy in DeltaByteArrayEncoder. This gives up to 10% wins on decimals locally. |
Contributor
Author
|
Would be nice to modify |
Contributor
|
would you be able to post the numbers & benchmarks you ran locally for this? itll be easier to review if we can see those, as currently there arent any concrete numbers for this performance PR |
Contributor
Author
|
Ah right, should be doable. I'll look into new benches that attempt to replicate the usage we have internally. |
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
6 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?
N/A
Rationale for this change
In our case, improve performance for delta-encoded decimals.
What changes are included in this PR?
Reduce allocs and copies when delta-encoding byte-arrays.
Also removed some suspicious
#[cold]attributes onEncoder::encoding.Are these changes tested?
Encoder correctness should be covered by existing tests.
I can add benchmarks if desired, since the removal of byte array copies is a nice win (especially combined with #10364).
Are there any user-facing changes?
parquet::encodings::encoding::Encoderhas a newflush_tofunction, but the module is gated behind the experimental feature. I can rewrite it to makeflush_toa default method if needed.