Yash Mayya created KAFKA-14732: ---------------------------------- Summary: Use an exponential backoff retry mechanism while reconfiguring connector tasks Key: KAFKA-14732 URL: https://issues.apache.org/jira/browse/KAFKA-14732 Project: Kafka Issue Type: Improvement Components: KafkaConnect Reporter: Yash Mayya Assignee: Yash Mayya
Kafka Connect in distributed mode retries infinitely with a fixed retry backoff (250 ms) in case of errors arising during connector task reconfiguration. Tasks can be "reconfigured" during connector startup (to get the initial task configs from the connector), a connector resume or if a connector explicitly requests it via its context. Task reconfiguration essentially entails requesting a connector instance for its task configs and writing them to the Connect cluster's config storage (in case a change in task configs is detected). A fixed retry backoff of 250 ms leads to very aggressive retries - consider a Debezium connector which attempts to initiate a database connection in its [taskConfigs method|https://github.com/debezium/debezium/blob/bf347da71ad9b0819998a3bc9754b3cc96cc1563/debezium-connector-sqlserver/src/main/java/io/debezium/connector/sqlserver/SqlServerConnector.java#L63]. If the connection fails due to something like an invalid login, the Connect worker will essentially spam connection attempts frequently and indefinitely (until the connector config / database side configs are fixed). An exponential backoff retry mechanism seems more well suited for the [DistributedHerder::reconfigureConnectorTasksWithRetry|https://github.com/apache/kafka/blob/a54a34a11c1c867ff62a7234334cad5139547fd7/connect/runtime/src/main/java/org/apache/kafka/connect/runtime/distributed/DistributedHerder.java#L1873-L1898] method. -- This message was sent by Atlassian Jira (v8.20.10#820010)