Prompt
Yes, actually I used this approach to avoid blocking the Kafka polling thread for a long time. Earlier, I was using CallerRunsPolicy. When all 16 threads were busy and the queue of 50 was full, the Kafka listener thread could start processing the message itself. The problem was that if the SFMC call took a long time due to retries, it could block the Kafka polling thread and cause consumer rebalancing. So I changed it to a bounded waiting approach. Now, when the queue is full, it waits for up to 10 seconds for an available slot. If a slot becomes available, the task is accepted for processing. If the queue is still full after 10 seconds, it throws a RejectedExecutionException instead of processing the SFMC call on the polling thread. For failed message processing, we log the error and do not explicitly acknowledge the failed record. But this 10-second waiting is not a two-attempt retry policy. I will verify the existing Kafka retry and error-handler behavior to confirm what happens to ...