Operating Kubernetes in Production
From “it runs” to “it survives”
Today
- 10–15 min recap
- Finish Lab 85
- Clarify grading expectations
- Introduce rollout & controlled change
- Think like production engineers
Quick Recap: Controllers
Key idea:
Kubernetes continuously reconciles desired state with actual state.
Interactive Question
If I delete a Pod manually:
- What happens?
- Who recreates it?
- Why?
Readiness vs Liveness
Readiness:
- Should this Pod receive traffic?
Liveness:
- Should this container be restarted?
Production impact:
- One controls traffic
- One controls restarts
Layered Architecture View
When debugging, always ask:
Which layer am I in?
Failure in Production
In real systems:
- Pods crash
- Images fail to pull
- Nodes run out of memory
- Deployments misconfigure secrets
Good engineers expect failure. They design for it.
Controlled Change: Rolling Updates
When you change an image:
kubectl set image deployment/quote-app app=quote-app:v2Kubernetes does not destroy everything.
It rolls forward gradually.
Observe a Rollout
kubectl rollout status deployment quote-appkubectl rollout history deployment quote-appYou can:
- Monitor progress
- Inspect revision history
Rollback
If something breaks:
kubectl rollout undo deployment quote-appProduction mindset:
Change must be reversible.
Think Like an Engineer
Ask yourself:
- What is my blast radius?
- What layer isolates failure?
- What is my rollback plan?
- What happens if the node dies?
- Where is persistence guaranteed?
This is what differentiates operators from YAML writers.
Lab 85 Reminder
Make sure your repository contains:
architecture-notes.md- Embedded architecture diagram
- Secret-based configuration
- Resource limits + probes
- Controlled failure analysis
If it is not in GitHub, it cannot be graded.
Next
- Finish required Lab 85 tasks
- Perform one controlled failure
- Be ready to explain your architecture
- Then we move into final production challenge next session