From Scratch

54 boards, 92 lessons, 655 review questions, in 13 chapters. Read them in order. Each chapter builds on the one before.

Your progress0 of 92 read

Start here → Enough Go to read these builds

Contents

1Enough Go1 lesson
The handful of Go every build in this track leans on: errors as values, defer, maps, goroutines, channels and a mutex, each shown working and then found in a real build file.
- 01Enough Go to read these builds8 Q
2Redis9 lessons
Build Redis yourself, one problem at a time: a program that answers on a port grows a wire format, an in-memory map, expiry, eviction and crash-safe persistence, until the real client can't tell the difference.
The Map The Socket That Answers RESP The Wire Protocol The Map SET GET DEL One Thread Many Clients Keys That Die (Expiry)Eviction Memory Is Full Surviving A Crash (Persistence)Everything
3HTTP Server9 lessons
A raw socket becomes an HTTP server: parsing the request, framing the body, routing, keep-alive, concurrency and chunked encoding, one layer at a time.
The Map The Raw Socket Parse The Request Frame The Body Routing and Response Keep-Alive Concurrency Chunked Encoding Everything
4Kafka9 lessons
A naive queue becomes a log, then segments with an offset index, partitions, consumer groups, replication and exactly-once delivery.
The Map The Naive Queue The Log Segments and Offset Index Partitions Consumer Groups Replication and Durability Exactly Once Everything
5Database9 lessons
An append-only file becomes a database: an index, pages, a B-tree, a write-ahead log and MVCC transactions, built rather than only diagrammed.
after One Database, to the Bolt
The Map The Append-Only File The Index Pages The B-Tree Write-Ahead Log Transactions (MVCC)The Query Layer Everything
6LSM Tree6 lessons
A memtable and a WAL grow into SSTables, a sparse index, bloom filters and leveled compaction, the write-optimised engine behind LevelDB and Bigtable.
after One Database, to the Bolt
7Distributed Scheduler9 lessons
A tick loop grows a durable job store, a timing wheel, leader election and a dispatch queue, and runs straight into the hard part every scheduler eventually hits: time.
The Map The Tick Loop Durable Job Store Heap and Timing Wheel Leader Election Dispatch Queue Idempotency and Fencing The Hard Part, Time Everything
8Sharded KV store5 lessons
A key-value store spread over many machines: a hash ring, quorum reads and writes, hinted handoff when a node dies, and Merkle trees that find what drifted.
after The stall gets famous
9Raft6 lessons
A cluster that agrees: leader election, log replication, crash-safe persistence and a replicated KV on top, run in a deterministic simulation with chaos tests.
after Data on one machine
10Kubernetes9 lessons
Reimplementing what the Kubernetes module describes from the outside: the reconcile loop, store and watch, the scheduler and the kubelet, to see the control plane instead of only using it.
after Kubernetes
The Map Imperative to Declarative The Reconcile Loop Store and Watch Objects and Controllers The Scheduler The Kubelet Services and Networking Everything
11Git6 lessons
Content-addressed blobs, trees, commits, refs and checkout, written in Go until the real git reads and clones what the build wrote.
12DNS Server6 lessons
A UDP socket becomes a name server: header, question, answer, compressed names and a forwarding resolver, checked against the real dig.
13Container Runtime8 lessons
Namespaces, pivot_root, /proc, cgroups and a veth cable, written in Go until a box has its own hostname, files, processes, memory ceiling and network.

Borrowed from other tracks (22)

You don't need to take those tracks. These are the only lessons this track uses from them, and each is linked where it is needed.