[SPARK-56801][SQL] Add bulk read+widen path for INT32 to Double Parquet vector updater#55795
Closed
LuciferYang wants to merge 8 commits into
Closed
[SPARK-56801][SQL] Add bulk read+widen path for INT32 to Double Parquet vector updater#55795LuciferYang wants to merge 8 commits into
LuciferYang wants to merge 8 commits into
Conversation
…et vector updater
…uet.ParquetVectorUpdaterBenchmark (JDK 17, Scala 2.13, split 1 of 1)
LuciferYang
commented
May 11, 2026
| DateToTimestampNTZUpdater 36 36 1 29.1 34.4 0.0X | ||
| DowncastLongUpdater (INT64 -> Decimal(9,2)) 2 2 0 487.0 2.1 0.4X | ||
| IntegerToLongUpdater 1 1 0 1277.6 0.8 1.0X | ||
| IntegerToDoubleUpdater 1 1 0 1475.2 0.7 1.2X |
Contributor
Author
There was a problem hiding this comment.
This is the embodiment of this optimization. will update others results later
…uet.ParquetVectorUpdaterBenchmark (JDK 21, Scala 2.13, split 1 of 1)
…uet.ParquetVectorUpdaterBenchmark (JDK 25, Scala 2.13, split 1 of 1)
…uet.ParquetVectorUpdaterBenchmark (JDK 21, Scala 2.13, split 1 of 1)
…uet.ParquetVectorUpdaterBenchmark (JDK 21, Scala 2.13, split 1 of 1)
…uet.ParquetVectorUpdaterBenchmark (JDK 21, Scala 2.13, split 1 of 1)
Contributor
Author
yaooqinn
approved these changes
May 12, 2026
LuciferYang
added a commit
that referenced
this pull request
May 12, 2026
…et vector updater ### What changes were proposed in this pull request? Extend the bulk read+widen pattern introduced in SPARK-56791 to `IntegerToDoubleUpdater` (parquet INT32 read into Spark `DoubleType`). A new `readIntegersAsDoubles` default method on `VectorizedValuesReader` does the per-row fallback. `VectorizedPlainValuesReader` overrides it to fetch source bytes once via `getBuffer(total * 4)` and run a tight in-method conversion loop. `IntegerToDoubleUpdater.readValues` becomes a one-line delegation. The widen is Java's primitive int-to-double conversion, lossless for all INT32 values. ### Why are the changes needed? `IntegerToDoubleUpdater.readValues` allocates a fresh `ByteBuffer` slice inside `getBuffer(4)` for every element on the legacy path, and that allocation dominates the loop. Collapsing N allocations into one is the same win SPARK-56791 delivered for the INT32 -> Long sibling, with the gain again amplifying on newer JDKs where escape analysis better optimizes the tight loop: | JDK | Baseline | After | Speedup | |----:|---------:|----------:|--------:| | 17 | 479.1 M/s | 1475.2 M/s | 3.08x | | 21 | 531.8 M/s | 6215.0 M/s | 11.68x | | 25 | 454.6 M/s | 6423.7 M/s | 14.13x | Peer Updaters in the same benchmark group hold within run-to-run noise, confirming the change is local to `IntegerToDoubleUpdater`. ### Does this PR introduce _any_ user-facing change? No. ### How was this patch tested? New unit tests in `ParquetVectorUpdaterSuite` cover boundary batch lengths (0, 1, 7, 8, 9, 17, 1024, 4097) and the singular `readValue` path. A new end-to-end test in `ParquetIOSuite` round-trips INT32 written to parquet and read back as `DoubleType`, exercising both REQUIRED columns (no def-levels) and OPTIONAL columns with interleaved nulls so that `readValue` and `readValues` are both invoked. Benchmark results on JDK 17, 21, and 25 are committed on the branch. ### Was this patch authored or co-authored using generative AI tooling? Generated-by: Claude Code Closes #55795 from LuciferYang/SPARK-56801-int32-to-double. Authored-by: YangJie <yangjie01@baidu.com> Signed-off-by: yangjie01 <yangjie01@baidu.com> (cherry picked from commit 1360e5c) Signed-off-by: yangjie01 <yangjie01@baidu.com>
Contributor
Author
|
Merged into master. Thanks @yaooqinn |
Contributor
Author
|
Thank you @dongjoon-hyun |
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
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.
What changes were proposed in this pull request?
Extend the bulk read+widen pattern introduced in SPARK-56791 to
IntegerToDoubleUpdater(parquet INT32 read into SparkDoubleType).A new
readIntegersAsDoublesdefault method onVectorizedValuesReaderdoes the per-row fallback.VectorizedPlainValuesReaderoverrides it to fetch source bytes once viagetBuffer(total * 4)and run a tight in-method conversion loop.IntegerToDoubleUpdater.readValuesbecomes a one-line delegation. The widen is Java's primitive int-to-double conversion, lossless for all INT32 values.Why are the changes needed?
IntegerToDoubleUpdater.readValuesallocates a freshByteBufferslice insidegetBuffer(4)for every element on the legacy path, and that allocation dominates the loop. Collapsing N allocations into one is the same win SPARK-56791 delivered for the INT32 -> Long sibling, with the gain again amplifying on newer JDKs where escape analysis better optimizes the tight loop:Peer Updaters in the same benchmark group hold within run-to-run noise, confirming the change is local to
IntegerToDoubleUpdater.Does this PR introduce any user-facing change?
No.
How was this patch tested?
New unit tests in
ParquetVectorUpdaterSuitecover boundary batch lengths (0, 1, 7, 8, 9, 17, 1024, 4097) and the singularreadValuepath. A new end-to-end test inParquetIOSuiteround-trips INT32 written to parquet and read back asDoubleType, exercising both REQUIRED columns (no def-levels) and OPTIONAL columns with interleaved nulls so thatreadValueandreadValuesare both invoked.Benchmark results on JDK 17, 21, and 25 are committed on the branch.
Was this patch authored or co-authored using generative AI tooling?
Generated-by: Claude Code