

Partition is considered “idle” and will not hold back the progress of watermarks in downstream operators.ĭescribes details about how to define a WatermarkStrategy#withIdleness. If no records flow in a partition of a stream for that amount of time, then that You will either need to lower the parallelism or add an idle timeout to the The Kafka Source does not go automatically in an idle state if the parallelism is higher than the

fromSource ( kafkaSource, new CustomWatermarkStrategy (), "Kafka Source With Custom Watermark Strategy" ) ĭetails about how to define a WatermarkStrategy. To enable partition discovery, set a non-negative value for Job, Kafka source can be configured to periodically discover new partitions under provided In order to handle scenarios like topic scaling-out or topic creation without restarting the Flink SetBounded(OffsetsInitializer) has been invoked

committedOffsets ()) // Start from committed offset, also use EARLIEST as reset strategy if committed offset doesn't exist builder () // Start from committed offset of the consuming group, without reset strategy Modern Kafka clients are backwards compatible with broker versions 0.10.0 or later.įor details on Kafka compatibility, please refer to the official Kafka documentation. The version of the client it uses may change between Flink releases. We recommend you use the latest stable version.įlink provides an Apache Kafka connector for reading data from and writing data to Kafka topics with exactly-once guarantees.Īpache Flink ships with a universal Kafka connector which attempts to track the latest version of the Kafka client. This documentation is for an out-of-date version of Apache Flink. Upgrading to the Latest Connector Version.Hadoop MapReduce compatibility with Flink.Conversions between Table and DataStream.Conversions between PyFlink Table and Pandas DataFrame.
