Scatter-gather is a fundamental distributed query execution pattern. The coordinator "scatters" a query to all shards, each shard processes its local data, and the coordinator "gathers" partial results into a final response.

Query execution flow:

1. Client sends a query to the coordinator
2. Coordinator forwards the query to all known shards
3. Each shard executes the query on its local data
4. Each shard returns partial results to the coordinator
5. Coordinator merges all partial results into a complete response
6. Coordinator returns the merged response to the client

Handling partial failures:

• Set a timeout for each shard response (e.g., 1000ms)
• If a shard times out, exclude its results but continue with other shards
• Track which shards responded successfully
• Return a "shards_responded" count so the client knows if results are complete

Example query:

Request:  {"type": "scatter_query", "msg_id": 1, "query": "SELECT * FROM users WHERE age > 25"}
Response: {"type": "scatter_query_ok", "in_reply_to": 1, "results": [...], "shards_total": 3, "shards_responded": 3}

If shard 2 is down:

Response: {"type": "scatter_query_ok", "in_reply_to": 1, "results": [...], "shards_total": 3, "shards_responded": 2}