Partitioning and Sharding

Split data and traffic while preserving locality, fairness, and operability.

35 min

3 sections

sharding

partitioning

scale

01. Partition for the dominant access path

Section 1 of 3

A shard key should spread load and keep common queries local. User ID works for user-owned data. Tenant ID works for SaaS isolation but can create hot tenants. Time works for logs but can create hot current partitions.

text

Shard key comparison:

user_id:
  Good for user timelines and profiles.
  Bad for global analytics.

tenant_id:
  Good for enterprise isolation.
  Risky when one tenant is huge.

hash(order_id):
  Good distribution.
  Bad for listing all orders for a user.

created_month:
  Good for retention and archival.
  Risky for hot current-month writes.

Rate Limiting and Backpressure

Search and Read Models

Back to Course