2026 Apache Flink Community Checkpoint

Join this fast-paced session through major themes in the Flink community: What are the most important FLIPs around Flink SQL, core engine improvements or LLMs?

Flink is a big project, it is difficult to keep track of the main developments. In this talk, we will give a high level overview over the most important developments in the community, providing "tl;dr" summaries on important topics so that you can decide what to dive deeper on.

Attendees will get a good overview over the recent and upcoming developments in Flink, and how they relate each other. We won’t be able to go very deep into each FLIP, but attendees will get an overview of what is going on in Flink and which FLIPs are worth exploring further.

Transcript

1. 2026 Flink Community Checkpoint Robert Metzger Confluent, Principal Engineer, Flink PMC Chair
2. Flink SQL Grow user base beyond Java developers and distributed system experts
3. Flink SQL: Materialized Tables • • • Declarative, long running Flink SQL statements with lifecycle management Schema evolution via ALTER MATERIALIZED TABLE Streaming or periodic batch exec INSERT INTO orders_total SELECT SUM(total_gbp) FROM orders; Old approach: need for creating the table and job management CREATE MATERIALIZED TABLE mt_orders_total FRESHNESS = INTERVAL '5' MINUTE REFRESH_MODE = CONTINUOUS AS SELECT SUM(total_gbp) FROM orders; Flink-managed table Source: https://rmoff.net/2026/04/28/materialized-tables-in-apache-flink/
5. Flink SQL: PTFs – Process Table Functions • • • • SQL UDFs on steroids Direct access to the Flink state, timers, events within SQL Transform a single or multiple input streams Status: PTFs are ready for production ◦ The community is now adding more built-in PTFs public static class GreetingWithFollowUp extends ProcessTableFunction<String> { public void eval(Context ctx, @StateHint StayState state, @ArgumentHint(ArgumentTrait.SET_SEMANTIC_TABLE) Row input) { state.name = input.getFieldAs("name"); state.counter++; collect("Hello " + state.name + ", your " + state.counter + " time?"); TimeContext<Instant> timeCtx = ctx.timeContext(Instant.class); timeCtx.registerOnTime(timeCtx.time().plus(Duration.ofMinutes(1))); } public void onTimer(StayState state) { collect("Hello " + state.name + ", I hope you enjoyed your stay!"); }}
7. Other Flink SQL improvements • FLIP-599 State Catalog (--> Room 2002, 10:30am, Gyula Fora, Apple) ◦ • FLIP-579 LATERAL SNAPSHOT Join ◦ • Query checkpoints and savepoints from Flink SQL Processing time temporal join / Flink state lookup join FLIP-516 Multi-way join (Flink 2.1)
8. Flink Operations Run Flink at scale
9. Flink Operations: SDK-native File Systems • Motivation ◦ Hadoop-based FS: painful to maintain: dependency and CVE footprint • • Performance: faster checkpoints (native bulk copy) Better o11y, client-side encryption
10. Flink Kubernetes Operator • • Today’s standard for (declaratively) deploying autoscaled, stateful Flink clusters FLIP-503: Blue/Green Deployments for Flink on Kubernetes: Phase 1 ◦ • Managed transitions between blue/green deployments FLIP-504: Blue/Green Deployments for Flink on Kubernetes: Phase 2 (proposed) ◦ Transition between jobs with no record duplication using watermarks (GateProcessFunction ) • FLIP-611: Running Flink jobs in MiniCluster (proposed) ◦ Run a Flink cluster in a small single pod
11. AI / LLMs Make Flink play nice with machine brains
12. AI in Flink: Landscape • Back in the day: FlinkML era ◦ The repo is not maintained anymore • Flink-agents.git subproject ◦ Run agents in Flink operators, with “exactly once action consistency” • AI in Flink SQL ◦ ◦ CREATE MODEL / ML_PREDICT – call external models VECTOR_SEARCH – like a lookup join for similar rows
13. AI in Flink: Ongoing discussions • FLIP-577: “AI-native Flink”: Visionary FLIP, 11 sub-FLIPs ◦ ◦ ◦ ◦ RpcOperator for delegating expensive inference VECTOR, TENSOR, IMAGE and FILE types New operators for images, video, audio, text Python DataFrame API • FLIP-608: Flink on Ray. Ray: a framework for running Python code on a cluster"
15. Flink versions overview
16. Q&A Robert Metzger Confluent Principal Engineer, Flink PMC
17. Flink Community Metrics How active is Flink compared to peer projects
18. Flink commits alone → solid growth in 2026
19. Flink vs Kafka → Flink is ahead again
20. Flink vs Spark → Same trajectory in recent years
21. LLMs and Slack have killed user mailing lists
22. Backup slides