Skip to content

[FIP-37] Add bitmap infrastructure: BitmapUtils, RoaringBitmapSerializer, AbstractRbAggFunction#3319

Merged
polyzos merged 5 commits into
apache:mainfrom
Prajwal-banakar:RoaringBitmap-UDFs
May 26, 2026
Merged

[FIP-37] Add bitmap infrastructure: BitmapUtils, RoaringBitmapSerializer, AbstractRbAggFunction#3319
polyzos merged 5 commits into
apache:mainfrom
Prajwal-banakar:RoaringBitmap-UDFs

Conversation

@Prajwal-banakar

@Prajwal-banakar Prajwal-banakar commented May 14, 2026

Copy link
Copy Markdown
Contributor

Purpose

Linked issue: Part of #3289

This PR adds the foundational infrastructure for FIP-37 RoaringBitmap SQL function implementation. It provides the serialization utilities, custom Flink type serializer, and base aggregate function class that will be used by the bitmap SQL functions (rb_build_agg, rb_or_agg, rb_and_agg, etc.) in subsequent PRs.

Brief change log

Added the following infrastructure files in fluss-flink/fluss-flink-common:

  • BitmapUtils.java: Utility methods for serializing/deserializing RoaringBitmap using the ByteBuffer-based approach, which matches the server-side RoaringBitmapUtils.serializeRoaringBitmap32 format used by FieldRoaringBitmap32Agg for wire compatibility.
  • RoaringBitmapSerializer.java: Custom Flink TypeSerializer for RoaringBitmap accumulators to ensure correct checkpoint/savepoint behavior. Without this, Flink falls back to Kryo which is sensitive to internal class layout changes across RoaringBitmap library versions.
  • RoaringBitmapTypeInfo.java: TypeInformation wrapper that provides the custom serializer to Flink's type system.
  • AbstractRbAggFunction.java: Base class for bitmap aggregate UDFs with @FunctionHint(accumulator = @DataTypeHint(value = "RAW", bridgedTo = RoaringBitmap.class)) annotation. This tells Flink's Table planner to skip reflection-based POJO field extraction on RoaringBitmap and use the custom TypeInformation instead.
  • BitmapUtilsTest.java: Unit tests covering null handling, empty bitmap, known values round-trip, large cardinality (100K elements), and server serialization compatibility.
  • RoaringBitmapSerializerTest.java: Unit tests covering RoaringBitmapSerializer and RoaringBitmapTypeInfo behavior, including serialization round-trip, copy methods, snapshot configuration, and type information checks.
  • pom.xml: Added RoaringBitmap dependency (version 1.3.0 from root pom).

The aggregate functions (rb_build_agg, rb_or_agg, rb_and_agg) and catalog registration will follow in subsequent PRs linked to this issue.

Tests

Unit tests added and passing:

  • BitmapUtilsTest.testNullInputToBytes() - null handling.
  • BitmapUtilsTest.testNullInputFromBytes() - null handling.
  • BitmapUtilsTest.testEmptyBitmapRoundTrip() - empty bitmap serialization.
  • BitmapUtilsTest.testKnownValuesRoundTrip() - correctness with known values.
  • BitmapUtilsTest.testLargeCardinality() - performance with 100K elements.
  • BitmapUtilsTest.testFormatCompatibleWithServerSerialization() - wire compatibility.

Tests are still running and this PR description will be updated with the final verification results once the full validation completes.

Verified with:

  • ./mvnw spotless:apply -pl fluss-flink/fluss-flink-common - BUILD SUCCESS.
  • ./mvnw test -pl fluss-flink/fluss-flink-common -Dtest=BitmapUtilsTest - BUILD SUCCESS.
  • ./mvnw clean install -pl fluss-flink/fluss-flink-common -DskipTests - BUILD SUCCESS.
  • ./mvnw clean package -DskipTests (full project build) - BUILD SUCCESS.
  • Checkstyle: 0 violations.

API and Format

This change does not affect any public API or storage format. It adds internal infrastructure utilities that will be used by future bitmap SQL functions.

Documentation

This change does not introduce new user-facing features yet. The bitmap SQL functions (rb_build_agg, rb_or_agg, rb_and_agg) and their documentation will be added in follow-up PRs.

@Prajwal-banakar

Copy link
Copy Markdown
Contributor Author

Hi @polyzos @wuchong @platinumhamburg could you please help review here?

try {
return BitmapUtils.toBytes(acc);
} catch (IOException e) {
throw new RuntimeException("Failed to serialize bitmap accumulator.", e);

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

use FlussRuntimeException

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Done switched to FlussRuntimeException

return new RoaringBitmap();
}

/** Merges multiple accumulators — required for session window aggregation. */

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

"required for session window aggregation" is misleading.
The merge operation is required for any two-phase / batch / merge-capable aggregation in the Flink Table API, not specifically session windows

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Updated the Javadoc

Comment thread fluss-test-coverage/pom.xml Outdated
<exclude>org.apache.fluss.flink.tiering.FlussLakeTieringEntrypoint</exclude>
<exclude>org.apache.fluss.flink.tiering.FlussLakeTiering</exclude>
<!-- end exclude for flink tiering service -->
<exclude>

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

instead of excluding can we add some test, for example for
merge / getValue / resetAccumulator? maybe ship with just only one concrete implementation?

without a consumer, there's no way to validate that the planner actually accepts the RAW(... bridgedTo = RoaringBitmap.class) hint together with the custom TypeInformation.
WDYT?

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Added a minimal concrete TestRbAggFunction in the test class and covered merge, getValue, resetAccumulator, and getAccumulatorType. Removed the jacoco exclude.

@polyzos

polyzos commented May 18, 2026

Copy link
Copy Markdown
Contributor

@Prajwal-banakar great work so far .. I just added a few comments, as small impovements, but it's almost there

@Prajwal-banakar

Copy link
Copy Markdown
Contributor Author

Hi @polyzos Thanks for the review, addresses all the comments, Please take another look when you get a time!

@Prajwal-banakar Prajwal-banakar requested a review from polyzos May 20, 2026 14:24
@polyzos

polyzos commented May 26, 2026

Copy link
Copy Markdown
Contributor

@Prajwal-banakar thank you for all the great work.. I pushed one small commit with some small changes.
Let me know if it looks to you and i can merge once the CI passes

@Prajwal-banakar

Copy link
Copy Markdown
Contributor Author

Hi @polyzos Thank you for the improvements! the ITCase is a much cleaner solution for validating the planner contract, and the serialize fix is cleaner too. LGTM!
the failing CI check is in fluss-server (JVM crash: "pure virtual method called" from native RocksDB code), which is unrelated to this PR's changes in fluss-flink-common. This looks like a pre-existing runner flake

@polyzos polyzos merged commit 68ad16c into apache:main May 26, 2026
17 of 19 checks passed
@Prajwal-banakar Prajwal-banakar deleted the RoaringBitmap-UDFs branch May 28, 2026 06:15
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants