Deployment, DNS, and HTTPS

What happens when someone types your URL

1. DNS — the browser asks the resolver for app.example.com; the resolver walks root → TLD → authoritative nameserver → returns an IP (cached along the way, per TTL).
2. TCP + TLS — connect to port 443; the TLS handshake verifies the server's certificate (issued by a CA the browser trusts) and establishes encryption. This is HTTPS: privacy and proof you're talking to the right server.
3. HTTP request — hits a load balancer → your server (or CDN edge, which answers before reaching you).

Certificates renew automatically these days (Let's Encrypt, platform-managed); the residual skill is diagnosing expiry mistakes and mixed-content warnings (https page loading http assets — blocked).

Release strategies

• Rolling — replace instances a few at a time; old and new serve simultaneously. Default in most platforms.
• Blue-green — two identical environments; flip the router from blue to green. Instant rollback = flip back.
• Canary — route 1% of traffic to the new version, watch metrics, ramp up. Catches what tests missed, on 1% of users instead of all of them.
• Feature flags (previous lesson) decouple the code deploy from the feature's release.

Choose by blast radius: flags for features, canary for risky refactors, rolling for the everyday.

Rollback: the skill that buys calm

Deploys fail. The professional move is a revert deploy — redeploy the previous artifact — not a hotfix under pressure. Requirements: artifacts are immutable and retrievable (previous lesson), migrations are forward-compatible with the previous code version (expand/contract!), and rollback is rehearsed. "We can't roll back because the migration broke compatibility" is the sentence that turns an outage into a long outage.

Databases in the deploy story

Schema migrations ride the same pipeline with extra care: migrate before the new code starts (or with expand/contract, any time), migrations must be idempotent-safe with the tooling's lock, and destructive changes wait until no running version needs the old column. Most deploy incidents are migration incidents.