Failover is currently not available for Transformer pipelines in Control Hub and offsets are not migratable to another Transformer instance. Since the pipeline is running directly on the Spark Cluster, YARN/Spark is managing any failures/retries on the Spark Cluster.
Two primary scenarios exist:
- Transformer Instance fails But Spark Driver Continues to Run: If the backend Spark Driver is still up and the Transformer instance recovers, the Transformer instance will connect to the Spark Driver to continue to obtain metrics which it will recommence to send to Control Hub
- Transformer Instances fails and Spark Driver fails or is not started : If the backend Spark Driver fails or was never started manual retry can be performed by the end-user. The transformer does not support automatic retry as of this writing.
