Can we do blue‑green if we have long‑running WebSocket sessions?

Yes. Terminate at a gateway that supports connection draining (e.g., NLB/ALB, Envoy) and set low idle timeouts during cutover. Keep blue up for a TTL to drain while new connections go to green. Consider sticky routing for a short window if stateful clients can’t reconnect gracefully.

What about shared state like Redis?

Use backwards‑compatible schemas for cache payloads or versioned keys (e.g., suffix with :v2). Keep TTL short during the release so stale entries don’t cross versions. If schemas differ materially, run dual caches briefly and switch namespaces with the same flip mechanism.

Is Argo Rollouts mandatory?

No. It’s great when you need policy and analysis, but a simple Service selector flip plus Prometheus gates covers 80% of teams. Standardize on one, make it observable, and automate rollback before piling on new tools.

How does this interact with feature flags?

Flags decouple product behavior from deploys. Use flags for risky logic, not infrastructure. During expand/contract DB changes, dual‑write and read‑path selection should be flag‑guarded so you can revert behavior without redeploying.

Release-engineering · Oct 4, 2025 · 10 minute read

Blue‑Green Without the Drama: Zero‑Downtime Releases that Don’t Torch Your CFR

A practical, battle‑tested playbook for blue‑green that cuts change failure rate, shortens lead time, and makes recovery boring.

Back to all posts

Blue‑Green Without the Drama: Zero‑Downtime Releases that Don’t Torch Your CFR

Key takeaways

Implementation checklist