What sharding strategies exist?

The main sharding strategies determine what principle data gets distributed between servers (shards) by. This affects load balance, query simplicity, and scalability.

#1. Hash sharding

• Each record's hash is computed on a specific key (e.g. user_id), and the hash result determines which shard it lands in.
• Provides even data distribution.
• Downside: aggregate queries that need data from every shard become harder (e.g. aggregates).

Example: shard_id = hash(user_id) % N, where N is the number of shards.

#2. Range sharding

• Data is split by ranges of key values (e.g. user_id from 1 to 1,000,000 is one shard, the next range is another).
• Simple to implement, convenient for ordered data.
• Downside: possible uneven load if some ranges are more active than others.

#3. Geographic (or attribute-based) sharding

• Data is split by an attribute tied to region, country, company branch, and so on.
• Convenient when users are physically spread across regions, it reduces latency and improves local availability.
• Downside: shards can vary a lot in size.

#4. Composite (hybrid) sharding

• Combines several strategies, e.g. first by region, then within each region by hash.
• A flexible, scalable option for large systems.

Summary:

• Hash sharding, even, but harder to aggregate.
• Range sharding, simple, but can produce uneven load.
• Geographic sharding, reduces latency, but requires balancing the data.
• Hybrid sharding, universal, but harder to manage.