In a sharding interview strategy, after proving a single database cannot handle the load, what should you propose next?

In a sharding interview strategy, after proving a single database cannot handle the load, what should you propose next?

• A. Propose 2PC as the default solution
• B. Propose a shard key aligned with access patterns and hash-based distribution
• C. Immediately implement range-based sharding
• D. Avoid discussing trade-offs

Answer: (B)

The main idea being tested is choosing a sharding approach that actually distributes load in line with how the data is accessed. After proving a single database can’t handle the load, you should propose a shard key strategy that matches access patterns and uses hash-based distribution. Why this is the best choice: selecting a shard key that reflects how queries access data helps route most requests to the appropriate shard (or a small set of shards), which reduces bottlenecks and balances load across many machines. Hash-based distribution further ensures an even spread of data, preventing hot shards that can become performance bottlenecks even when the data is partitioned. This combination enables scalable reads and writes while keeping routing simple and minimizing the need for cross-shard operations. Be mindful of trade-offs: hashing aids load balance but can complicate range queries or scans and may require rebalancing as data grows. It’s important to discuss how you’ll handle cross-shard operations, transactions, and future re-sharding, rather than skipping trade-offs altogether. Proposing default 2PC would introduce heavy coordination overhead, and jumping to range-based sharding without considering access patterns can create hotspots or migration challenges.

The main idea being tested is choosing a sharding approach that actually distributes load in line with how the data is accessed. After proving a single database can’t handle the load, you should propose a shard key strategy that matches access patterns and uses hash-based distribution.

Why this is the best choice: selecting a shard key that reflects how queries access data helps route most requests to the appropriate shard (or a small set of shards), which reduces bottlenecks and balances load across many machines. Hash-based distribution further ensures an even spread of data, preventing hot shards that can become performance bottlenecks even when the data is partitioned. This combination enables scalable reads and writes while keeping routing simple and minimizing the need for cross-shard operations.

Be mindful of trade-offs: hashing aids load balance but can complicate range queries or scans and may require rebalancing as data grows. It’s important to discuss how you’ll handle cross-shard operations, transactions, and future re-sharding, rather than skipping trade-offs altogether. Proposing default 2PC would introduce heavy coordination overhead, and jumping to range-based sharding without considering access patterns can create hotspots or migration challenges.