Latest in Builds and Pipelines

Public TLS certificates now last at most 200 days, will last 100 days from March 2027 and 47 days from 2029, and big services like Sentry are moving from DigiCert to Let's Encrypt and Google Trust Services. Here's the schedule from the CA/Browser Forum rules, what breaks (manual renewals, certificate pinning, mTLS with public certificates, old Node.js trust stores), and four tested Node.js scripts: an expiry and chain checker, an ACME Renewal Information (ARI) lookup, an mTLS test, and the NODE_EXTRA_CA_CERTS and --use-system-ca trust options.

Vite+ 1.0 replaces Vite, Vitest, ESLint, Prettier and your Node version manager with one vp command. A React + TypeScript project migrated with vp migrate, tested step by step: what the migration changes, the lint error it introduces in its own output, a no-undef rule that produced 10,840 false errors on Excalidraw, a task cache that can replay a stale production build, and the missing library that stops managed Node 26 in slim Docker images.

A test suite upgraded from Vitest 4.1 to 5.0.3, hitting each breaking change on purpose: mocks cleared before every test, un-awaited .resolves failing, nested vi.mock, the new bench API, strict browser locators, and a -t filter that quietly runs zero tests and still passes CI. Here are the real error messages, the tested fixes, and a grep checklist to find affected code before you upgrade.

Apple shipped container 1.0 — an open-source, native CLI that runs Linux containers on macOS by giving each one its own lightweight micro-VM. It speaks OCI images and has a Docker-like CLI. Here's a real Node.js app built and run on it, the networking model, x86 images via Rosetta, and the gotchas — all verified on container 1.0.0 on an Apple-silicon Mac.

Which base image should a Node.js app use? The choice swings your image from ~1.8 GB down to ~210 MB and changes how you debug and secure it. Here are the real sizes for the same app on node:26, node:26-slim, node:26-alpine, and distroless — all measured with Docker — plus the tradeoffs that actually matter.

Node.js can bundle your script and the runtime into one executable file — a Single Executable Application — so users run your CLI without installing Node at all. Here's the full build, an embedded asset, and proof the binary runs on a machine with no Node installed. Verified on real Docker images.

You no longer need Jest, Mocha, or Vitest to test a Node.js project. node:test ships a complete runner — suites, hooks, mocking, filtering, watch mode, and coverage — that you run with node --test. No dependencies, no config. Every command and its output here is from a real node:26-slim Docker image.

Node.js runs .ts files directly — no ts-node, no tsx, no build step. But the details differ between Node 22, 24, and 26: what's flagged, what's stable, and one feature that was removed in Node 26. Every example here was run on real node:22-slim, node:24-slim, and node:26-slim Docker images.

Node.js has quietly absorbed the jobs that used to need dependencies: a test runner, a file watcher, a .env loader, native TypeScript, fetch, terminal colors, deep clone, even SQLite. Here's what you can drop on Node.js 26 — every example verified on a real node:26 Docker image.

ChunkLoadError: Failed to load chunk after a deploy isn't a build cache problem. It's version skew: a tab opened before the deploy asks for a JavaScript file that the new build deleted. I reproduced it in Next.js 16.3 and tested the fixes: keeping old static files, reloading once on ChunkLoadError, what deploymentId does and doesn't do, and the CDN cache that leaves pages silently dead.

docker stats shows memory per container but no total. The usual one-liner sums the MEM % column, and that's wrong as soon as any container has a memory limit. In a test it reported 13.53 GiB used, 174% of the machine, when the containers were really using about 1 GB. Here's a correct shell version, a Node.js version that groups by Compose project, and what the number actually measures.

Run several Node.js containers behind nginx with Docker Compose, spread requests evenly with least_conn, and add or remove replicas without touching nginx, using the resolve option that open-source nginx gained in 1.27.3. Measured: 73 → 279 req/s going from 1 to 4 replicas, and zero failed requests while scaling down, once one easy-to-miss timeout is set. Also: why the classic Docker-DNS nginx config silently ignores its own upstream block.