How can I ensure exactly-once delivery when reading data from Kafka with Spark?
To ensure exactly once delivery, you can use the idempotent producer setting in Kafka and the checkpointing mechanism in Spark Structured Streaming. This is possible to achieve if your sink is a delta table, but not if the sink is another Kafka topic.