Summary of Flink Forward Asia 2022: Keynotes, Technical Innovations, and Industry Deployments of Apache Flink

Flink Forward Asia (FFA) 2022, jointly organized by the Apache Software Foundation and Alibaba Cloud, was held online on November 26‑27. The summit gathered thousands of participants, featured the fourth Flink Real‑Time Computing Challenge award ceremony, and summarized a year of rapid community growth: over 20,000 GitHub stars, 1,600+ contributors, and 14 million monthly downloads.

The conference emphasized that Apache Flink has become the global standard for real‑time stream computing, powering a wide range of scenarios such as real‑time dashboards, data integration, lakehouse analytics, personalized recommendation, and risk monitoring.

Keynote Highlights

Alibaba’s vice‑president Jia Yangqing discussed the symbiotic relationship between cloud and open‑source, describing how cloud provides elastic environments for open‑source software and how this drives the emergence of cloud‑native concepts.

Alibaba’s senior engineer Wang Feng presented the 2022 technical achievements of the Flink community, including two major releases (Flink 1.15 and 1.16) that introduced upgrades to checkpointing, state storage, Hybrid Shuffle, and broader ecosystem compatibility.

Technical Innovations

Distributed Consistent Checkpoint Upgrade : Introduction of Unaligned Checkpoint, Buffer Debloating, and Log‑based Checkpoint to reduce checkpoint latency and cost.

Cloud‑Native State Storage : Optimizations to RocksDB state backend in Flink 1.16 (2‑10× performance) and a roadmap toward a fully separated storage architecture called Tiered State Backend.

Hybrid Shuffle : A new shuffle mechanism that combines the low‑latency pipelined shuffle of streaming with the robustness of batch blocking shuffle, enabling better resource utilization.

Flink CDC 2.0 : A universal incremental snapshot framework offering high‑performance parallel reads, lock‑free source ingestion, and reliable breakpoint‑resume capabilities; supports MySQL, PolarDB, Oracle, MongoDB, PostgreSQL, TiDB, and more.

Flink ML 2.0 : Re‑built on the DataStream API to provide iterative computation for online training, checkpoint‑based recovery, and a growing library of algorithms.

Flink Table Store : A cloud‑native, storage‑compute‑separated table format (LakeStore + LogStore) that delivers superior update and query performance compared with Hudi.

Industry Deployments

Midea Group : Uses Flink for long‑cycle B‑side analytics, factory production monitoring, and real‑time promotional dashboards, integrating Kafka, Hive, Redis, and HBase.

miHoYo : Processes billions of game logs daily with Flink, builds near‑real‑time data warehouses on Iceberg, and implements real‑time risk control, leveraging Flink SQL, CDC, and Table Store.

Disney (Hulu/Disney+) : Deploys Flink for ad decision‑funnel reconstruction, exposure monitoring, and ad‑system dashboards, running on Kubernetes with Flink Operator, gang scheduling, and elastic scaling.

The summit concluded with optimism about Flink’s continued evolution: community activity remains high, innovations in state, fault‑tolerance, shuffle, data integration, and machine learning are accelerating, and more enterprises are contributing back to the ecosystem. The next steps include advancing the Streaming Data Warehouse vision and expanding cloud‑native capabilities.