What Is Exactly-Once Processing.

Exactly-once processing in Flink is a guarantee that each event in a stream is processed one time only—with no duplicates and no data loss—even if failures or restarts occur.

When to Use

Use exactly-once semantics when accuracy is critical. For example, bank transfers, payment pipelines, sensor alerts, or inventory updates require this guarantee to prevent double counting or missed events.

Example

Think of a Flink job tracking online purchases: exactly-once ensures that each purchase is recorded a single time, even if the system crashes and restarts.

Want to level up your prep?

Explore Grokking System Design Fundamentals, Grokking the System Design Interview, Grokking Database Fundamentals for Tech Interviews, or practice with Mock Interviews with ex-FAANG engineers.

Why Is It Important

It guarantees correctness in mission-critical systems. Analytics, billing, and fraud detection pipelines rely on exactly-once to ensure reliable, trustworthy results.

Interview Tips

Explain that Flink achieves exactly-once using checkpointing, state snapshots, and two-phase commit sinks. Contrast it with at-least-once (duplicates possible) and at-most-once (data loss possible).

Trade-offs

You gain accuracy and consistency, but incur extra latency and resource costs due to checkpointing and transactional sinks.

Pitfalls

Many assume it means events are never replayed. In reality, Flink may reprocess events after failure, but guarantees the final state is as if each event ran once. Misconfigured sinks or missing checkpoints often break this guarantee.