Skip to content
Half Stack Python

Kubernetes: Deployment Strategies Reference

Table of Contents

Section titled “Table of Contents”

Kubernetes Deployment Strategies

Section titled “Kubernetes Deployment Strategies”

Kubernetes Deployment Strategies — Summary & Deeper Guidance

Section titled “Kubernetes Deployment Strategies — Summary & Deeper Guidance”

This document summarizes common Kubernetes deployment strategies and adds practical depth around when, why, and how to use each safely in real systems.

Recreate

Section titled “Recreate”

Traffic shift model: Cut-then-start (all old Pods terminated before new ones start) Downtime risk: High (full outage) Rollback effort: Very easy K8s primitives / tools: Deployment with strategy.type: Recreate

Ideal use cases

  • Local/dev or lab environments
  • One-off jobs or batch workloads
  • Breaking state resets or destructive migrations

Key caveats

  • Guaranteed outage
  • Fails readiness/liveness checks
  • Unsuitable for any SLA/SLO-backed service

Rolling Update (Default)

Section titled “Rolling Update (Default)”

Traffic shift model: Gradual Pod replacement Downtime risk: Low (tunable) Rollback effort: Easy (rollout undo) K8s primitives / tools: Deployment with RollingUpdate, maxSurge, maxUnavailable

Ideal use cases

  • Standard production services
  • Stateless or lightly stateful apps
  • Teams without service mesh

Key caveats

  • Partial failures may hide until most traffic is shifted
  • Slow for large fleets
  • Requires good readiness probes

Ramped / Slow Rollout

Section titled “Ramped / Slow Rollout”

Traffic shift model: Throttled rolling update with pauses Downtime risk: Very low Rollback effort: Easy K8s primitives / tools:

  • kubectl rollout pause/resume
  • Argo Rollouts manual steps
  • GitOps promotion gates

Ideal use cases

  • Mission-critical services
  • Regulated or compliance-heavy systems
  • Human-in-the-loop validation

Key caveats

  • Operational overhead
  • Requires on-call release ownership
  • Slower time-to-market

Blue-Green

Section titled “Blue-Green”

Traffic shift model: Instant switch between two identical environments Downtime risk: Zero (if readiness is correct) Rollback effort: Instant K8s primitives / tools:

  • Two Deployments
  • Service selector swap or Ingress routing

Ideal use cases

  • Major version upgrades
  • DB schema changes
  • Compliance or audit-driven releases

Key caveats

  • Doubles infrastructure cost
  • Risk of stale “green” environment
  • Requires strict cleanup discipline

Best-Effort Controlled Rollout

Section titled “Best-Effort Controlled Rollout”

Traffic shift model: Aggressive rolling update Downtime risk: Moderate Rollback effort: Easy K8s primitives / tools:

  • Deployment with high maxSurge and maxUnavailable (e.g., 100%/100%)

Ideal use cases

  • Large stateless fleets
  • Internal tools
  • Speed prioritized over strict uptime

Key caveats

  • Brief traffic drops possible
  • Can violate SLOs under peak load
  • Needs capacity headroom

Shadow Deployment

Section titled “Shadow Deployment”

Traffic shift model: Mirrored traffic (responses discarded) Downtime risk: Zero Rollback effort: None needed K8s primitives / tools:

  • Istio traffic mirroring
  • Envoy filters
  • Telepresence

Ideal use cases

  • Performance testing with real traffic
  • ML model validation
  • Regression detection before release

Key caveats

  • Double compute cost
  • Must scrub PII
  • No user-visible correctness signal

Canary

Section titled “Canary”

Traffic shift model: Progressive traffic percentages (1% → 100%) Downtime risk: Very low Rollback effort: Very easy K8s primitives / tools:

  • Argo Rollouts
  • Flagger
  • Istio VirtualService weights

Ideal use cases

  • Feature launches
  • Risk-sensitive changes
  • Error-budget-aware organizations

Key caveats

  • Requires strong metrics and alerts
  • Automation complexity
  • Slower than pure rolling updates

A/B Testing

Section titled “A/B Testing”

Traffic shift model: Attribute-based routing (user, cookie, header) Downtime risk: Very low Rollback effort: Moderate K8s primitives / tools:

  • Service mesh or API gateway rules
  • Feature-flag platforms

Ideal use cases

  • UX and pricing experiments
  • ML model comparison
  • Personalization

Key caveats

  • Requires analytics pipeline
  • Risk of state leakage between variants
  • Cohort cleanup needed post-test

Rule of Thumb

Section titled “Rule of Thumb”
  • Speed > uptime: Aggressive rolling
  • Uptime > speed: Canary or slow rollout
  • Zero-risk validation: Shadow
  • Hard cutovers: Blue-green
  • Experiments: A/B testing
  • Never production: Recreate