MEPX
Chapter 5 of 10All chapters
  1. 1Start with numbers
  2. 2Tiers and statelessness
  3. 3Scaling
  4. 4Caching
  5. 5Data at scale
  6. 6Queues and async work
  7. 7Consistency
  8. 8Designing for failure
  9. 9Knowing what is happening
  10. 10Making the trade-off

Chapter 5 of 10

Data at scale

Replicas, sharding and their costs.

Replication

Read replicas copy the primary and serve reads, which spreads load. They lag behind by milliseconds to seconds, so a user can write and then not see their own change.

  • Route reads that must be current to the primary.
  • Failover promotes a replica, and doing it automatically is harder than it looks.

Sharding

Splitting data across machines by key removes the single-machine limit and makes cross-shard queries and transactions painful. Postpone it as long as honestly possible.