I have enabled the spark.streaming.backpressure.enabled setting and also set spark.streaming.backpressure.initialRate to 15000, but my spark job is not respecting these settings when reading from Kafka after a failure.
In my kafka topic around 500k records are waiting for being processed and they are all taken in 1 huge batch which ultimately takes a long time and fails with executor failure exception. We don't have more resources to give in our test cluster and we expect the backpressure to kick in and take smaller batches. What can I be doing wrong? Thanks & Regards Biplob Biswas