Sharding

The shape

Sharding divides records among independent storage partitions. The partition key determines where a record lives and which machines must be consulted for a query.

The useful question

Does the key distribute the workload, not merely the number of records? One large tenant or time based write pattern can overload a single shard while the cluster average looks calm.

Stress test

Send most writes to one key range. Observe how tail latency and queue depth diverge from the cluster average.

Tradeoff

Random or hashed keys spread load but make ordered range scans harder. A key chosen for convenient queries may create uneven partitions.