일괄 소비와 승인 모드

By default a @KafkaListener processes one record at a time. For high-throughput pipelines, processing records in batches reduces overhead and lets you bulk-write to databases.

Set batch-listener to true on the container factory, or configure it in properties.

When batch mode is on, your listener accepts a List of payloads instead of a single value.

You can also receive the full list of ConsumerRecord objects to read keys, partitions, and offsets per record.

The batch size is bounded by max.poll.records. Tune it to control how many records each poll fetches.

Spring Kafka offers several AckMode values:

• BATCH — commit offsets after the whole batch.
• RECORD — commit after each record.
• MANUAL / MANUAL_IMMEDIATE — you call ack explicitly.

Configure the ack mode on the listener container factory.

With MANUAL mode, inject an Acknowledgment and call acknowledge() only after successful processing of the batch.

Acknowledging after processing gives at-least-once delivery: if the app crashes mid-batch, the batch is redelivered.

Make your batch processing idempotent so reprocessing is safe.

Larger batches improve throughput but increase per-message latency and memory use. Start with a moderate max.poll.records and measure.

Batch consumption plus the right ack mode lets you build efficient, reliable consumers. Combine type: batch, a tuned poll size, manual ack, and idempotent writes.

You learned batch consumption and acknowledgment modes.

• Set type: batch to receive a List of records.
• Tune max.poll.records for batch size.
• Choose an AckMode to control offset commits.
• Acknowledge after processing and keep writes idempotent.