Asynchronous Coordination

Problem Statement

Edge agents can't always wait for cloud responses (too slow). Cloud agents can't always push to edge immediately (network full). How do we coordinate asynchronously?

Solution Approach

Message Queue Pattern:

Edge Agent:
  task = {user_input: "...", priority: "normal"}
  queue.send("cloud_tasks", task)
  // Don't wait for response; continue

Cloud Agent:
  task = queue.recv("cloud_tasks")
  result = {output: "...", status: "success"}
  queue.send("edge_results", result)

Edge Agent (later):
  result = queue.recv("edge_results", timeout=5s)
  return result to user

Async Patterns:

1. Fire-and-forget — edge doesn't wait for response

2. Use: non-critical operations (logging, analytics)

3. Request-response with timeout — edge waits with timeout

4. Use: normal operations

5. Batch aggregation — accumulate edge requests, send in batch

6. Use: high volume, batch-able work

Ordering Guarantees:

Without ordering:
  Edge sends: Request A, Request B
  Cloud receives: Request B, Request A (out of order)
  Result: B processed before A; wrong!

With ordering (partition key):
  Request A: partition_key = "user:alice"
  Request B: partition_key = "user:alice"
  Queue guarantees: A then B (same partition)

When to Use

• Use async for high-latency edge-cloud connections
• Use fire-and-forget for non-critical work
• Use request-response for interactive workflows
• Use batch for high-volume work

Trade-offs

Pattern	Latency	Ordering	Complexity
Fire-and-forget	Low (no wait)	None (async)	Low
Request-response	High (wait)	Per-key (partition)	Medium
Batch	Medium (batch delay)	Per-batch	High

Recommendation: Request-response with timeout for interactive; batch for analytics.

References

• Network Partition Handling
• Asynchronous Coordination