Key Updates and New Features in Apache Flink 1.14.2 Release

Apache Flink 1.14.2 was released on December 16, addressing a critical Log4j vulnerability that allowed arbitrary code execution and delivering several important feature updates.

Flink‑Pulsar Connector OOM Issue

The Pulsar client uses Netty, which allocates memory differently on Java 11 versus Java 8. On Java 11 it draws from the direct memory pool and is limited by MaxDirectMemory, potentially causing OOM because the current client lacks a memory‑limit configuration. Users are advised to run the connector on JDK 8 or allocate sufficient memory for Flink.

Table API & SQL

Pipeline name alignment

In batch mode the default job name for DataStream programs changes from "Flink Streaming Job" to "Flink Batch Job"; a custom name can be set via the pipeline.name option.

fromChangelogStream changes

The StreamTableEnvironment.fromChangelogStream method may now produce different streams because primary‑key handling has been corrected compared with version 1.13.2.

Table#flatMap type inference

Table.flatMap()

now supports the new type system, enabling upgraded functionality.

Scala implicit conversions

New implicit conversions have been added between the DataStream API and Table API for Scala users.

YAML environment file removal

The sql-client-defaults.yaml file, deprecated since 1.13, is fully removed. Users should use the -i startup option to execute an SQL initialization file that defines catalogs, table sources/sinks, UDFs, and other session properties.

Deprecated/Removed APIs

Maven modules renamed:

flink-table-planner-blink -> flink-table-planner
flink-table-runtime-blink -> flink-table-runtime
flink-table-uber-blink -> flink-table-uber

The removal of BatchTableEnvironment also eliminates BatchTableSource and BatchTableSink; users should migrate to DynamicTableSource and DynamicTableSink. TableEnvironment#connect method removed. toAppendStream and toRetractStream deprecated.

Old versions of the SQL Kafka and Elasticsearch connectors, along with their legacy format options ( connector.type / format.type), have been removed; use the unified connector option instead.

BatchTableSource/Sink, HBaseTableSource/Sink, ParquetTableSource, OrcTableSource, and related classes deleted.

BatchTableEnvironment and the old planner have been removed from PyFlink.

DataStream API

Idle processing fix for multiple input operators

Classes such as AbstractStreamOperator and Input now include processWatermarkStatusX methods, allowing proper handling of WatermarkStatus when combining watermarks across two or more inputs.

@TypeInfo annotation on POJO fields

Users can now apply the @TypeInfo annotation directly to POJO fields.

Connectors

Standardized metrics exposure

Connectors that implement the unified Source and Sink interfaces automatically expose a set of standardized metrics.

KafkaSink replaces FlinkKafkaProducer

The new KafkaSink supersedes the older FlinkKafkaProducer.

FlinkKafkaConsumer deprecated

The FlinkKafkaConsumer API is now deprecated.

Checkpoints

alignmentTimeout semantics change

The meaning of the alignmentTimeout configuration has shifted to represent the time between checkpoint initiation and the receipt of checkpoint barriers by tasks.

BROADCAST disables unaligned checkpoints

Broadcast partitions cannot be used with unaligned checkpoints because consistent consumption rates across all channels cannot be guaranteed, potentially leading to state inconsistencies during recovery.

Apache Mesos no longer supported

Mesos is being phased out in favor of Kubernetes; users are encouraged to transition accordingly.

Note: The article also promotes a PDF collection titled “The Road to Becoming a Big‑Data God” and provides instructions to obtain the PDF via a WeChat public account, along with numerous links to related big‑data articles and resources.