4.2.5 · Lesson 9 of 16
Streaming and Async Execution Paths
Use streaming for interactive UX and async/batch paths when work can finish later without blocking users.
WHAT
Choose streaming vs asynchronous execution based on user wait expectations and workload timing.
HOW
| Need | Path |
|---|---|
| Interactive drafting / chat | Streaming (ConverseStream + WebSocket/SSE) |
| Nightly document summaries | Bedrock Batch Inference / async |
| Identical FAQ repeats | Semantic cache (skip FM) |
EXAM TRAP
Adding SQS or Glacier between the user and Bedrock usually worsens perceived latency for interactive apps.
Key points
- Streaming for TTFT
- Async batch for offline jobs
- WebSockets / SSE
- Do not queue when users need live tokens
- Match path to SLA
Exam focus
Streaming improves perceived latency for interactive apps; Batch Inference / queues fit non-interactive overnight work.