Skip to content

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

Open
Prajwal-banakar wants to merge 4 commits into
apache:mainfrom
Prajwal-banakar:RoaringBitmap-UDFs
Open

[FIP-37] Add bitmap infrastructure: BitmapUtils, RoaringBitmapSerializer, AbstractRbAggFunction#3319
Prajwal-banakar wants to merge 4 commits into
apache:mainfrom
Prajwal-banakar:RoaringBitmap-UDFs

Conversation

@Prajwal-banakar
Copy link
Copy Markdown
Contributor

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

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.
  • fluss-test-coverage/pom.xml: Added a JaCoCo exclusion for AbstractRbAggFunction, since it is abstract and cannot be directly covered by unit tests.
  • 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

}

@Override
public void serialize(RoaringBitmap record, DataOutputView target) throws IOException {
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.

record.runOptimize() is called here and then BitmapUtils.toBytes(record) also calls runOptimize().

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.

Fixed, removed the redundant runOptimize() call from serialize(); BitmapUtils.toBytes() handles it.

}

@Override
public boolean equals(Object obj) {
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.

equals doesn't delegate through canEqual, which breaks Flink's documented symmetry contract for TypeInformation subclasses.

Either: return obj instanceof RoaringBitmapTypeInfo && ((RoaringBitmapTypeInfo) obj).canEqual(this); or simplify the whole class to reference equality since the constructor is private and INSTANCE is the only instance. Also worth adding @threadsafe here and on RoaringBitmapSerializer

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.

Fixed equals to delegate through canEqual. Added @threadsafe to both classes.

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
Copy link
Copy Markdown
Contributor

polyzos commented May 18, 2026

@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!

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