Tech.
45 posts — Posts about Tech
AgentMux: A Control Plane for AI Coding Agents on Remote Servers
Open-sourced AgentMux, a desktop app for managing AI coding agents on remote servers: every server, project and agent in one tree, sessions running inside tmux on the server, so closing the app only detaches. Go + Wails 3 + React, MIT.
Earth: A 3D Globe in the Browser, All the Way Out to the Observable Universe
Open-sourced a new project, Earth: a pure-frontend 3D globe in the spirit of Google Earth. Click a country for its flag, population, GDP, and government; GDP bars, trade routes, a time-travel mode, and a cosmic scale ladder from Earth to the observable universe.
PolarDB's Storage-Compute Separation
An architectural look at PolarDB's storage-compute separation: how decoupling compute from storage addresses scalability, storage cost, and high availability, compared with AWS Aurora and traditional MySQL architecture.
Lessons from Managing an Engineering Team
Some plain-spoken lessons from a few years of leading engineering teams: think things through first, make ownership clear, let the team grow by letting go, use process to reduce problems—and why team atmosphere and a long-term view matter just as much.
Microservice Architecture Design
Starting from the bottlenecks of monolithic architecture, this post walks through the core design principles of microservices, the overall architecture, and key design questions — plus their scalability, adoption challenges, and where they actually fit.
Designing L1 and L2 Caches
A look at the two-tier caching architecture combining a local (L1) cache with a Redis (L2) cache: the access flow, what each tier solves, consistency and capacity control, and common cache update strategies.
Exploring etcd
Starting from real distributed coordination needs, this post covers what etcd is for and how Raft delivers strong consistency, walks through a cluster setup verified with etcdctl, and compares etcd with Redis for coordination workloads.
Understanding Flink
Starting from the shift from batch to stream processing, this post walks through the problems Flink solves — real-time computation, state management, fault tolerance, and event time — and its core design ideas: Stream First, State Driven, and Exactly Once.
A Deep Dive into Netty
Starting from the limitations of BIO, this post works through Java NIO's three core components — Channel, Buffer, and Selector — and the Reactor threading model, to understand the design ideas behind Netty's high-performance networking.
Algorithms: The Basics
Starting from 'what is an algorithm': how algorithms relate to data structures, why time complexity matters, and why learning algorithms is about modeling problems rather than grinding through problem sets.