Apache Flink Release History

2.2.0 ? expected H2 2026 Forecast

Drawn from active FLIPs and public community discussion. Not a commitment - scope can shift, slip, or be cut between now and the actual release.

Materialized Tables GA hardening - production-grade refresh scheduler, observability, schema-evolution flows
ForSt state backend evolution - more checkpointing modes, faster restore on rescale
Further disaggregated state work (FLIP-423 follow-ups) for cloud-native deployments
Adaptive batch scheduler + speculative execution refinements
Java 21 support; Java 11 minimum to be removed
More streaming-SQL operator improvements: better changelog reasoning, EXPLAIN diagnostics
PyFlink: closer feature parity with Java APIs, performance work
Removal of remaining deprecated APIs (legacy SinkFunction, older state APIs)

2.1.0 2026-04 Latest

Delta join - planner-selected join that replaces large state with bidirectional source lookups for CDC-style joins
Improved adaptive batch scheduler with dynamic parallelism inference
SQL Gateway materialised-table scheduler refinements
ForSt state backend hardening for object-storage-backed jobs

2.0.0 2025-03 Major

API cleanup - removal of long-deprecated DataSet API, Scala APIs, and legacy SourceFunction
ForSt distributed state backend - state stored on remote object storage instead of local RocksDB
Disaggregated state architecture (FLIP-423) for cloud-native deployments
Materialized Tables become first-class with managed refresh lifecycle
Java 8 support dropped (Java 11 remains the minimum, Java 17 fully supported)

1.20.0 2024-08

Materialized Tables (preview) - declarative SQL tables with managed refresh
Full job restart from a savepoint without restoring state (clean re-bootstrap)
Adaptive batch scheduler improvements; dynamic graph optimisation
Unified state API consolidation across DataStream / Table / SQL

1.19.0 2024-03

Dynamic source parallelism inference for batch jobs
Configurable source-sink table options at runtime
Trace and metric reporting standardisation
Deprecation of legacy SinkFunction in favour of the Sink V2 API

1.18.0 2023-10

Generic incremental file compaction in the streaming file sink
SQL Gateway REST endpoint stabilised; JDBC driver for Flink SQL
Java 17 support (preview)
Watermark alignment GA for sources

1.17.0 2023-03

Speculative execution for batch jobs to mitigate stragglers
Watermark alignment across split sources
STATEMENT SET multi-sink statements in SQL
Generic incremental checkpoints for non-RocksDB backends

1.16.0 2022-10

Adaptive batch scheduler GA - automatic parallelism per stage
Hybrid shuffle - blocking shuffle that can spill to disk
Async lookup-join cache, retryable lookup mechanism
PyFlink performance and feature parity improvements

1.15.0 2022-05

Pulsar source and unified Kafka Source / Sink (FLIP-27 / FLIP-191)
JSON / Avro format improvements; CSV support enhanced
Operator finished state for upgrades
Scala-free runtime (Scala isolated to user jars)

1.14.0 2021-09

Hybrid source - read bounded historical data then switch to unbounded streaming
Buffer debloating for backpressure relief
Fine-grained resource management for SlotSharingGroups
Pluggable scheduler interface

1.13.0 2021-05

Reactive scaling mode - cluster auto-scales by adding/removing TaskManagers
Passive deployment mode
SQL improvements: window TVFs, batch in SQL CLI
Bottleneck detection in the web UI

1.12.0 2020-12

DataStream API unification for batch and streaming (BATCH execution mode)
PyFlink DataStream API
Native Kubernetes HA without ZooKeeper
Upsert Kafka connector for retract/upsert streams

1.11.0 2020-07

New source/sink interface (FLIP-27) - unified bounded + unbounded sources
Flink SQL DDL improvements (CREATE TABLE WITH connector options)
Hive integration GA, Hive catalog stable
Application mode for cluster-per-job deployment

1.10.0 2020-02

Blink SQL planner merged - more capable optimiser, becomes default in 1.11
Native Kubernetes integration
Fully managed memory model for TaskManager
Python UDFs in Flink SQL

1.9.0 2019-08

Redesigned web UI
Blink SQL planner contributed by Alibaba (preview)
State Processor API for batch reads/writes of savepoints
Batch shuffle service improvements

1.8.0 2019-04

State schema evolution for POJOs and Avro
Continuous cleanup of expired keyed state in RocksDB
Universal Kafka connector for 1.x and 2.x brokers

1.7.0 2018-11

Streaming SQL with temporal joins, CEP via MATCH_RECOGNIZE
S3 StreamingFileSink with exactly-once
Versioned state schema serializers

1.6.0 2018-08

State TTL - per-state expiry with configurable cleanup
Jar management REST API
FlinkCEP improvements; PatternProcessFunction

1.5.0 2018-05

FLIP-6 deployment - per-job and session cluster modes redesign
New REST API for cluster management
Network stack rewrite with credit-based flow control
Broadcast state pattern

1.4.0 2017-12

End-to-end exactly-once with Kafka 0.11 transactional sink
TwoPhaseCommitSinkFunction for transactional sinks
Hadoop-free Flink distribution

1.3.0 2017-06

Side outputs for late data and multi-way splitting
Rescaling history in savepoints
Async I/O operator for non-blocking external lookups

1.2.0 2017-02

Dynamic scaling through savepoints with rescaled parallelism
Queryable state (preview) for external state lookups
ProcessFunction with timers

1.1.0 2016-08

Rescalable state - state can be redistributed across parallel instances
Library improvements: Gelly graph, FlinkML, FlinkCEP
Connectors for Kinesis, Cassandra, ElasticSearch

1.0.0 2016-03 Major

API stability guarantee - DataStream / DataSet APIs frozen
At-least-once Kafka 0.9 connector with offset checkpoints (end-to-end exactly-once arrives later in 1.4 with Kafka 0.11 transactions)
Out-of-core state via RocksDB backend
Web monitoring dashboard maturity

0.10.0 2015-11

Event time and watermarks in the DataStream API
Kafka 0.9 connector with offset checkpoints
Side outputs and timer service foundations

0.9.0 2015-06

Off-heap managed memory for the runtime
Table API (preview) - relational queries on DataStream/DataSet
Streaming connectors for Kafka, RabbitMQ, Twitter, Flume

0.8.0 2015-01

Improved type system with TypeInformation extractors
Hadoop 2.x integration
Iteration improvements (bulk and delta)

0.7.0 2014-11

Streaming API introduced (continuous DataStream programs)
Scala API matures alongside Java
FlinkML and Gelly libraries seeded

0.6.0 2014-08 First

First release under the Apache Flink name (formerly Stratosphere)
Batch DataSet API with optimiser and cost-based execution
Initial cluster manager support (Standalone, YARN)