Basic Information
| Field | Details |
|---|---|
| English name | Amazon SQS |
| Full name | Amazon Simple Queue Service |
| Chinese description | 消息队列服务 |
| Japanese description | マネージドメッセージキューサービス |
| Exam frequency | ⭐⭐⭐⭐⭐ |
| Often confused with | SNS / EventBridge / Kinesis |
In one sentence
A managed message queue that decouples producers and consumers and buffers work for reliable asynchronous processing.
Key points
- Standard queues provide very high throughput with at-least-once delivery and possible duplicates or reordering.
- FIFO queues preserve order within message groups and support deduplication with lower throughput limits.
- Visibility Timeout hides a received message while it is processed; successful consumers delete it.
Exam takeaway
Use DLQs, retries, idempotent consumers, and suitable retention and visibility settings.
Update: polling, ordering, and idempotency
- Long polling waits up to 20 seconds and reduces empty responses and API cost, but it does not extend visibility timeout.
- In FIFO queues, the message group ID defines the ordering boundary: strict order within a group and parallelism across groups.
- Deduplication IDs address producer duplicates; group IDs address order and concurrency. Consumers still need idempotency for at-least-once delivery.
- Size visibility timeout for normal processing and combine retries with a DLQ and redrive policy.
Update: retries, DLQs, and partial failures
- If processing exceeds the visibility timeout, a message becomes visible again; coordinate processing time, heartbeats, and retry count.
- FIFO message group IDs define ordering and parallelism boundaries; exactly-once processing still depends on consumer idempotency and external writes.
- Lambda batch consumers can return partial batch responses so only failed records are retried.