Apache Flink Release History and Key Features from 1.7 to 1.12

Flink 1.7

Introduced full Scala 2.12 support, Exactly‑once semantics for S3 StreamingFileSink, state schema evolution, MATCH_RECOGNIZE in Streaming SQL, temporal tables and joins, a versioned REST API, Kafka 2.0 connector, and local recovery to speed up failover.

Flink 1.8

Added the final Schema Evolution story, TTL‑based continuous cleanup of old state, user‑defined functions and aggregates in SQL, RFC‑compliant CSV format, a new KafkaDeserializationSchema exposing ConsumerRecord, Kinesis watermark options, DynamoDB stream support, and global aggregation for sub‑task coordination.

Flink 1.9

Released in August 2019 with batch‑oriented fine‑grained recovery, a Blink‑based query engine for Table API/SQL (preview), State Processor API for flexible savepoint manipulation, stop‑with‑savepoint semantics, a redesigned Web UI, and preview Hive integration.

Flink 1.10

Marked as the largest release to date, bringing extensive performance and stability optimizations, native Kubernetes integration (beta), PyFlink enhancements, full Blink integration, production‑ready Hive support, expanded SQL DDL (watermarks, temporal tables), and a pluggable module system for built‑in functions.

Flink 1.11

Introduced unaligned checkpoints, a new unified Source API, CDC support in Table & SQL, JDBC catalog for relational databases, enhanced Hive real‑time warehousing, application‑mode deployment, and major Python improvements including vectorized Pandas UDFs and Cython‑accelerated UDFs.

Flink 1.12

Unified DataStream API with batch execution mode, Kubernetes‑based high‑availability without ZooKeeper, upsert‑Kafka connector, metadata columns in SQL, temporal table joins via FOR SYSTEM_TIME AS OF, new Kinesis connector, sort‑merge shuffle (experimental), and extensive Table API/SQL type‑system and optimizer upgrades.

Command‑line examples

./bin/flink run -d -e kubernetes-session -Dkubernetes.cluster-id=examples/streaming/WindowJoin.jar

bin/flink stop -p [:targetSavepointDirectory] :jobId

StreamExecutionEnvironment env = StreamExecutionEnvironment.getExecutionEnvironment();
env.setRuntimeMode(RuntimeMode.BATCH);

from pyflink.datastream import StreamExecutionEnvironment, MapFunction
class MyMapFunction(MapFunction):
    def map(self, value):
        return value + 1
env = StreamExecutionEnvironment.get_execution_environment()
env.from_collection([1,2,3,4,5]).map(MyMapFunction()).print()
env.execute("datastream job")