Flink’s fault tolerance mechanism recovers programs in the presence of failures and continues to execute them. Such failures include machine hardware failures, network failures, transient program failures, etc.
Flink can guarantee exactly-once state updates to user-defined state only when the source participates in the snapshotting mechanism. The following table lists the state update guarantees of Flink coupled with the bundled connectors.
Please read the documentation of each connector to understand the details of the fault tolerance guarantees.
|Apache Kafka||exactly once||Use the appropriate Kafka connector for your version|
|AWS Kinesis Streams||exactly once|
|RabbitMQ||at most once (v 0.10) / exactly once (v 1.0)|
|Twitter Streaming API||at most once|
|Sockets||at most once|
To guarantee end-to-end exactly-once record delivery (in addition to exactly-once state semantics), the data sink needs to take part in the checkpointing mechanism. The following table lists the delivery guarantees (assuming exactly-once state updates) of Flink coupled with bundled sinks:
|HDFS rolling sink||exactly once||Implementation depends on Hadoop version|
|Elasticsearch||at least once|
|Kafka producer||at least once / exactly once||exactly once with transactional producers (v 0.11+)|
|Cassandra sink||at least once / exactly once||exactly once only for idempotent updates|
|AWS Kinesis Streams||at least once|
|File sinks||at least once|
|Socket sinks||at least once|
|Standard output||at least once|
|Redis sink||at least once|