Inside Modern Storage Engines

What actually happens after a database accepts a write?

Inside Modern Storage Engines opens the database black box and follows data through memory, write-ahead logs, pages, trees, indexes, files, compaction, snapshots, checksums, caches, and recovery. Written for backend, database, and systems engineers, it connects application workloads to the physical mechanisms that determine durability, latency, amplification, and operational risk.

The book compares B-tree and LSM-based designs without reducing them to simplistic read-versus-write slogans. It explains how acknowledged data becomes durable, why checkpoints and compaction can create latency spikes, how Bloom filters and indexes trade memory for avoided I/O, and why long-lived snapshots can prevent safe reclamation. Local storage is then connected to placement, object storage, erasure coding, integrity checks, and repair.

Inside, you will learn how to:

- Trace write-ahead logging, fsync, checkpoints, and crash recovery
- Understand page layouts, B-trees, buffer pools, and access paths
- Follow memtables, SSTables, Bloom filters, and compaction scheduling
- Measure read, write, space, and recovery amplification
- Reason about MVCC visibility, vacuum, garbage collection, and snapshots
- Evaluate batching, caching, compression, and hardware-aware I/O
- Design format changes, corruption drills, checksums, and repair processes
- Connect local engine behavior to distributed storage responsibilities

Production case studies, inspected open-source engines, runnable teaching implementations, failure analysis, and review checklists turn internal mechanisms into architectural judgment. The book also makes the limitations of simplified models explicit: a toy WAL or tree can reveal an invariant without pretending to be a production database.

Inside Modern Storage Engines is for readers who want to understand not only which database they selected, but how that system transforms logical operations into durable bytes - and where the deferred costs eventually appear.