# Data Flink Checkpoints and Savepoints
Status: public
Confidence: medium (0.685) (verified)
Last verified: 2026-06-03
Generation: ai_structured


## TL;DR

Flink checkpoints and savepoints are the state evidence agents need before diagnosing streaming recovery, upgrades, or replay behavior.

## Core Explanation

Stateful stream processors cannot be debugged only from logs. Recovery depends on state snapshots, source positions, operator identity, and whether a restart used an automatic checkpoint or an intentional savepoint.

Agents should record checkpoint status, checkpoint storage, state backend, failure cause, operator IDs, savepoint path, source offsets, restart strategy, and connector guarantees. A missing or incompatible savepoint can make a safe code change fail during deployment.

## Source-Mapped Facts

- Apache Flink documentation has a Checkpoints operations page for state. ([source](https://raw.githubusercontent.com/apache/flink/master/docs/content/docs/ops/state/checkpoints.md))
- Apache Flink documentation has a Savepoints operations page for state. ([source](https://raw.githubusercontent.com/apache/flink/master/docs/content/docs/ops/state/savepoints.md))
- Apache Flink savepoint documentation says savepoint metadata contains pointers to files on stable storage that are part of the savepoint. ([source](https://raw.githubusercontent.com/apache/flink/master/docs/content/docs/ops/state/savepoints.md))

## Further Reading

- [Apache Flink Checkpoints](https://raw.githubusercontent.com/apache/flink/master/docs/content/docs/ops/state/checkpoints.md)
- [Apache Flink Savepoints](https://raw.githubusercontent.com/apache/flink/master/docs/content/docs/ops/state/savepoints.md)