CNCK

Hello, CNCK

2026-06-01T00:00:00+00:00

This is the first post on CNCK — Cloud Native Community Korea — a blog built with Zola</a> and a minimal monochrome theme of the same name.

The goal is simple: write things down. Most posts here start as notes someone wished existed while solving a problem, polished just enough to be useful later.

The stack</h2>

Zola 0.22 for the static site, a single Rust binary with Sass, search, feeds, and syntax highlighting built in.</li>
A custom CNCK theme: serif body text, monospaced metadata, a lot of whitespace, and a light/dark toggle.</li>
GitHub Pages for hosting, deployed automatically on every push to `main</code>.</li> </ul> What to expect</h2> Short, practical posts about infrastructure and the systems around it. If a post saves one person an afternoon of debugging, it earned its place.`

Draining Kubernetes Nodes Without Waking Anyone 2026-05-18T00:00:00+00:00 Rotating nodes should be a non-event. In practice it pages people, because the defaults optimize for “it works on an empty cluster,” not “it works while serving traffic.” Here are the knobs that matter. PodDisruptionBudgets</h2> A PodDisruptionBudget</code> tells the eviction API how much disruption a workload can tolerate. Without one, kubectl drain</code> will happily evict every replica at once. apiVersion: policy/v1 kind: PodDisruptionBudget metadata: name: api spec: minAvailable: 80% selector: matchLabels: app: api</code></pre> Prefer minAvailable</code> as a percentage over an absolute count, so it scales with the deployment. Termination grace period</h2> The drain sends SIGTERM</code>, then waits terminationGracePeriodSeconds</code> before SIGKILL</code>. Your app must: Stop accepting new connections on SIGTERM</code>.</li> Finish in-flight requests.</li> Exit before the deadline.</li> </ol> If your readiness probe keeps the pod in rotation during shutdown, you’ll drop requests no matter how long the grace period is. Fail readiness first. </blockquote> Putting it together</h2> With a PDB, a sane grace period, and a preStop</code> sleep to let the load balancer deregister, a full node roll becomes something you can run at 2pm instead of 2am. A Small Rust CLI That Earns Its Keep 2026-04-27T00:00:00+00:00 Not every script needs to become a program. But once a shell script grows past a screen, sprouts flags, and starts parsing JSON, the maintenance cost flips. A tiny Rust CLI is often the cheaper long-term option. use std::process::ExitCode; fn main() -> ExitCode { let args: Vec<String> = std::env::args().skip(1).collect(); match args.first().map(String::as_str) { Some("version") => { println!("{}", env!("CARGO_PKG_VERSION")); ExitCode::SUCCESS } _ => { eprintln!("usage: tool <command>"); ExitCode::FAILURE } } }</code></pre> The win isn’t speed, it’s the type system catching the edge cases you’d otherwise discover in production: the empty input, the missing field, the timezone. Ship a single static binary and the operational story gets simpler too.

CNCK

Hello, CNCK

What to expect</h2>
Short, practical posts about infrastructure and the systems around it. If a post saves one person an afternoon of debugging, it earned its place.</p>

Draining Kubernetes Nodes Without Waking Anyone

Putting it together</h2>
With a PDB, a sane grace period, and a `preStop</code> sleep to let the load balancer deregister, a full node roll becomes something you can run at 2pm instead of 2am.</p>`

A Small Rust CLI That Earns Its Keep

CNCK

Hello, CNCK

What to expect</h2> Short, practical posts about infrastructure and the systems around it. If a post saves one person an afternoon of debugging, it earned its place.</p>

Draining Kubernetes Nodes Without Waking Anyone

Putting it together</h2> With a PDB, a sane grace period, and a preStop</code> sleep to let the load balancer deregister, a full node roll becomes something you can run at 2pm instead of 2am.</p>

A Small Rust CLI That Earns Its Keep

What to expect</h2>
Short, practical posts about infrastructure and the systems around it. If a post saves one person an afternoon of debugging, it earned its place.</p>

Putting it together</h2>
With a PDB, a sane grace period, and a `preStop</code> sleep to let the load balancer deregister, a full node roll becomes something you can run at 2pm instead of 2am.</p>`