Start here
Engineering logs about production systems. Each one takes a failure mode, names the mechanism underneath it, and puts a number on the cost.
The topics are databases, caches, distributed systems, observability, and AI infrastructure. No news, no frameworks of the week.
How it publishes: I write in research-sprint batches and the site releases the queue on a steady schedule, currently a new log every two days. The subscription sends each log on the day it goes live, by email or RSS.
Pick your entry point
Your database is slow
- Why UUID Primary Keys Quietly Destroy Database Performance · why UUID primary keys shred a B+ tree
- Your ORM issued 400 queries and the p99 looked fine until it didn't · the query count nobody graphs
You are designing a distributed system
- System Design: Principles for Maintainability, Scalability, and Reliability · faults versus failures, the vocabulary for everything else
- Consistency Models in Azure Cosmos DB: From Strong to Eventual · the consistency spectrum with real semantics
- Clock synchronization is a nightmare: Why Spanner uses TrueTime and the rest of us suffer · why you cannot trust timestamps across machines
You want to know what the hardware is doing
- CPU Caches and Spatial Locality: Why an Array is 3x Faster Than a Linked List for the Exact Same Big-O Complexity · cache lines decide your loop performance
- Virtual Memory and Lazy Allocation: Why RSS Matters More Than malloc() · what RSS actually measures before the OOM killer visits
- Thread-per-Core Architecture: Why Extra Threads Eventually Destroy Throughput · why fewer threads can be faster
You are building with LLMs
- AI agents break connection pooling by holding the slot while they think · agents break the oldest contract in backend infra
- The Expensive Cosplay of Local Models: True 3 AM Operational Cost of Hosting Llama-3 · the real total cost of local inference
- The AI code review bottleneck: When writing code is 5x faster, but reviewing is 2x slower · where AI review actually helps
You would rather read code than prose
- Decoding isolation levels: I built a toy DB to force dirty reads and phantom reads · a toy database that demonstrates dirty and phantom reads
- Implementing LFU Cache in O(1) Time: A Hands-on Breakdown · an O(1) LFU cache, built up from the paper
- I Rebuilt a Minimal BitTorrent Client in Go to Understand Peer-to-Peer Choking Algorithms · a minimal BitTorrent client and its choking algorithm
Or follow a series
Ordered reading paths. Start at instalment one; the later ones assume the earlier ones.
- Database internals in production [5 published]
- Caching in anger [1 published]
- Distributed systems failure modes [5 published]
- AI infrastructure without the hype [3 published]
- Mechanical sympathy [6 published]