Apache Flink 1.11.0 Release: New Features and Optimizations

The Apache Flink community announced the official release of Flink 1.11.0, with contributions from over 200 developers and more than 1,300 fixes and optimizations.

Key engine improvements include the introduction of unaligned checkpoints (FLIP‑76), which reduce checkpoint latency under back‑pressure by allowing barriers to bypass buffered data, and a new Source interface (FLIP‑27) that unifies stream and batch source implementations, simplifying event‑time handling, watermark generation, and idle detection.

Table API/SQL enhancements add native support for Change Data Capture (CDC) via Debezium and Canal formats, a unified WatermarkGenerator interface, and a richer file‑system connector supporting Avro, ORC, and Parquet formats. New dynamic table source/sink interfaces (FLIP‑95) and support for Hive DDL/DML (FLIP‑123) further extend SQL capabilities.

PyFlink upgrades bring vectorized Python UDFs with Pandas support (FLIP‑97), enabling efficient data exchange via Apache Arrow, and introduce Pandas UDF syntax such as:

@udf(input_types=[DataTypes.BIGINT(),DataTypes.BIGINT()],result_type=DataTypes.BIGINT(),udf_type="pandas")
def add(i, j):
    return i + j

Additional PyFlink improvements include fromPandas() / toPandas() converters, custom Table Functions, Cython‑based UDF performance boosts, and user‑defined metrics.

Deployment mode changes add an Application mode (FLIP‑85) where the job’s main method runs on the cluster, simplifying submission and enabling native Kubernetes support.

Other notable updates cover unified JobManager memory configuration (FLIP‑116), Web UI enhancements, standardized Docker images, and various planner and metric refinements.

Binary releases and source code are available on the Flink download page, with PyFlink packages on PyPI. Users are encouraged to test the new version and provide feedback via the Flink mailing list and JIRA.