Field notes / Published here
I write about how systems work.
Long-form notes on Kubernetes, AI infrastructure, open-source systems, and the engineering details that make them work.
Reading path
Every field note lives here, in one focused reading library.
14 notes in this path
From kubectl apply to a Linux Process: Kubernetes Internals, One State Transition at a Time
September 20, 2026Trace what really happens after kubectl apply, from API Server and etcd to controllers, scheduling, kubelet, networking, and a running, reachable Pod.
How Volcano Adds Safe Per Queue Admission Limits for AI Workloads
August 22, 2026A code-first look at Volcano's queue-scoped overcommit controls, which add safe per-queue admission limits while preserving the cluster-wide safety boundary for AI workloads.
Rolling Updates for LLM Serving on Kubernetes: How Kthena Protects Availability
August 5, 2026A source-level look at Kthena’s ServingGroup rolling updates, revision tracking, readiness gates, partitions, and availability budgets.
Inside Koordinator: Resource Policy, Scheduling, and Node-Level Enforcement
July 26, 2026A code-first walkthrough of how Koordinator connects API policy, scheduler plugins, node-local observation, and Linux resource enforcement.
Prefill-Decode Disaggregation on Kubernetes: How Kthena Actually Implements It
July 18, 2026A code-first look at how Kthena coordinates independent prefill and decode roles, compatible routing, and KV-cache transfer on Kubernetes.
Inside Volcano Webhooks: Admission, Validation, and the Gate That Rejects Bad Jobs Before They Run
July 12, 2026A code-first breakdown of Volcano’s AdmissionReview flow, shared webhook transport, validation, and workload mutation before storage.
Inside Volcano Scheduler Plugins: DRF and How Queue Fairness Actually Works (Part 2 of 2)
July 4, 2026Explore Volcano's plugin framework, Dominant Resource Fairness (DRF), queue fairness, and how scheduler plugins influence resource allocation decisions in Kubernetes.
Inside Volcano Scheduler: Sessions, Actions, and the Scheduling Cycle That Prevents Deadlocks (Part 1 of 2)
June 29, 2026A code-first walkthrough of SchedulerCache, Sessions, ordered Actions, and atomic gang planning in Volcano.
Inside Volcano Controllers: Gang Scheduling, State Machines, and Real Kubernetes Logs
June 19, 2026A deep dive into Volcano's source code — how Kubernetes batch scheduling works under the hood, from gang scheduling to state machines.
React Performance Engineering
June 16, 2026Most React apps are slow for the same three reasons: too many re-renders, too much in the DOM, and work that blocks the main thread. This article addresses all three.
Inside a Kubernetes Dashboard: Architecture Lessons from Volcano Dashboard
May 13, 2026Volcano is a batch-scheduling system for Kubernetes. Here's what I learned reading its dashboard source while preparing to contribute and apply for LFX mentorship.
Before ChatGPT, AI Couldn't Remember Your Last Sentence
April 19, 2026What the memory problem in early AI models was, why it mattered, and how the Transformer changed everything.
How I Added JavaScript Code Generation to Apache Fory's IDL Compiler
March 30, 2026Apache Fory lets you define data types once in a schema file. Java, Python, Go, Rust, C++, C# all had code generation — JavaScript didn't. Here's how I added it.
Apache Fory Row Format Explained
March 3, 2026Apache Fory supports a cache-friendly row format for analytics workloads. A walkthrough of the format design, memory layout, and how it compares to standard serialization.