I'm running a JDBC source connector and try to monitor its status somehow via the exposed JMX metrics and a prometheus exporter. However the status of the connector and all its tasks are still in the running state when the query fails or db can't be reached.
In earlier versions it seems that no value for source-record-poll-total in the source-task-metrics was exported when the query failed, in the versions I use (connect-runtime-6.2.0-ccs, confluentinc-kafka-connect-jdbc-10.2.0, jmx_prometheus_javaagent-0.14.0) even when failing the metric is exported with value 0.0.
Any ideas how I could detect such a failing query or db-connection?
This is resolved in version 10.2.4 of the jdbc connector. Tasks now fail when a SQLNonTransientException occurs and this can be detected using the exported metrics. See https://github.com/confluentinc/kafka-connect-jdbc/pull/1096