Scatter–gather

Problem

You need to answer a query that requires data from multiple shards/services, and you need the aggregate result.

Context

Search across N sharded indexes; get user's items from 5 different services; run analytics across shards.

Solution

Fan out the query in parallel to all N sources. Wait for responses (with timeouts). Aggregate the results into a single response.

Trade-offs

• Latency = max(slowest response) — tail latency dominates
• Bandwidth cost: 1 query becomes N queries
• Partial results may be returned if some sources time out

Failure modes

• One slow shard makes every query slow — mitigate with per-shard timeout + partial results
• One shard returns bad data — aggregator can't validate
• Fanout multiplied by N shards → massive load on all shards

When to use

• Sharded search (find X across all shards)
• Federated queries across services
• Multi-region aggregation

When NOT to use

• You can pre-aggregate the data (denormalize)
• Query hits only one shard by key — no scatter needed