Database Internals for App Engineers

Storage Engines: Row vs Column

Understand how row-oriented and columnar storage engines work under the hood so you can pick the right one for your workload—before you hit production.

B-trees and LSM

B-trees and LSM-trees make opposite bets on read vs write amplification—knowing which bet matches your workload saves you from the wrong index structure at scale.

WAL and Durability

The Write-Ahead Log is what separates a durable database from an expensive key-value store—here is what actually happens between your COMMIT and the disk.

MVCC and Snapshot Isolation

MVCC lets readers never block writers—but the version heap it creates has maintenance costs that surprise every team running PostgreSQL at scale.

Replication Mechanics

Physical and logical replication make different promises about consistency and flexibility—understanding the difference determines whether your failover works or loses data.

Sharding Strategies

Sharding moves complexity from the database to the application layer—choosing the wrong strategy means hot spots, unbalanced data, or cross-shard transactions you cannot avoid.

Distributed Consensus

Raft and Paxos give your distributed database a single authoritative truth—but the guarantees come with latency and availability tradeoffs that your application must be designed around.

Query Planning

The query planner turns your SQL into a physical execution plan—knowing how it estimates costs lets you write queries that stay fast as your data grows.

Indexes, End to End

Indexes are not free—every index you add is a write tax on every INSERT, UPDATE, and DELETE, and the wrong index type silently fails to help your queries.

The Cost of Consistency

Every step you take toward stricter consistency—from eventual to linearizable—has a measurable latency and availability price; this post puts numbers on that price.