Which statement best describes how shard keys influence cross-shard lookups?

Which statement best describes how shard keys influence cross-shard lookups?

• A. Shard keys should be chosen to minimize cross-shard queries based on common access patterns.
• B. Shard keys should be chosen randomly without regard to access patterns.
• C. Shard keys should always be timestamps.
• D. Shard keys are irrelevant to cross-shard performance.

Answer: (A)

Shard keys determine where data lands across the set of shards, so they shape how many shards a given query must touch. The best choice is to align the shard key with the application's common access patterns, so most queries can be satisfied by a single shard. When the shard key maps related data to the same shard, cross-shard lookups are minimized and latency stays lower because fewer network hops and coordination are needed. For example, if most reads retrieve all data for a single user, partitioning by user_id means that user’s data stays on one shard, avoiding cross-shard retrievals. Conversely, if you choose a shard key at random, data distribution becomes unpredictable and queries often need to span multiple shards, increasing latency and complexity. Relying on timestamps as the shard key can push related data across many shards as time ranges grow, similarly hurting cross-shard performance. Saying shard keys don’t matter for cross-shard behavior isn’t correct, since they control data placement and thus how many shards a query touches.

Shard keys determine where data lands across the set of shards, so they shape how many shards a given query must touch. The best choice is to align the shard key with the application's common access patterns, so most queries can be satisfied by a single shard. When the shard key maps related data to the same shard, cross-shard lookups are minimized and latency stays lower because fewer network hops and coordination are needed.

For example, if most reads retrieve all data for a single user, partitioning by user_id means that user’s data stays on one shard, avoiding cross-shard retrievals. Conversely, if you choose a shard key at random, data distribution becomes unpredictable and queries often need to span multiple shards, increasing latency and complexity. Relying on timestamps as the shard key can push related data across many shards as time ranges grow, similarly hurting cross-shard performance. Saying shard keys don’t matter for cross-shard behavior isn’t correct, since they control data placement and thus how many shards a query touches.