System DesignPatterns

20 parts · 169 chapters · English + 日本語

Mechanism-level reasoning for distributed, data, ML, and AI systems. Every chapter is written as a production contract: the invariant, the arithmetic behind it, and the evidence that defends it.

Start with your question

Is this data still correct under concurrent writes?Transactions, isolation levels, replication and conflict resolution, traced through the interleavings that break each guarantee.Part 1 · ACID Transactions →What gives way first when the load grows?Cache economics, invalidation races, partitioning and edge delivery, with the capacity arithmetic behind every choice.Part 4 · Semantics & Economics →How does work survive past the request?Queues, worker pools, durable execution, commit protocols for retries and idempotency, leases and recovery.Part 18 · Workflow Fundamentals →How do you find the right thing quickly?Inverted indexes, lexical query execution, vector retrieval, ranking, and typeahead under real traffic.Part 14 · Index Architecture & Internals →How do models and agents reach production?Feature stores, GPU inference, serving, offline and online evaluation, retrieval, and platforms for coding agents.Part 16 · ML System Fundamentals →Is this design ready to be signed off?Design documents and ADRs, review governance, supply-chain risk, verification strategy, and lawful retention.Part 20 · Design Documents & ADRs →

The whole edition

Parts are grouped by role rather than difficulty. Every chapter opens with a TL;DR and closes with three to eight curated neighbours, so any page can be the first one you open.