_CORE
AI & Agentic Systems Core Information Systems Cloud & Platform Engineering Data Platform & Integration Security & Compliance QA, Testing & Observability IoT, Automation & Robotics Mobile & Digital Banking & Finance Insurance Public Administration Defense & Security Healthcare Energy & Utilities Telco & Media Manufacturing Logistics & E-commerce Retail & Loyalty
References Technologies Blog Know-how Tools
About Collaboration Careers
CS EN
Let's talk

Zero-downtime deployments

Deployment is a non-event.

Blue-green, canary releases, feature flags. Your team deploys daily without stress — automated, tested, with instant rollback.

Daily
Deployment frequency
<15 min
Lead time
<30s
Rollback time
<5%
Change failure rate

Why deployment must not be an event

If your team has a “deployment window” and everyone is on standby, the system is telling you something important: the deployment process is broken. Deployment should be routine — like a commit, like code review. Automated, tested, with a clear rollback plan.

Blue-Green Deployment

Two identical environments:

  1. Blue — current production, serving 100% of traffic
  2. Green — new version, deployed and tested
  3. Smoke tests on green ✅ → traffic switch (DNS or load balancer)
  4. Problem? Switch back to blue in seconds

For Kubernetes: Argo Rollouts with automatic health checks. Rollout stops if metrics exceed threshold.

Canary Releases

Progressive rollout for large systems:

  1. New version gets 5% traffic
  2. Automatic metrics: error rate, latency, business KPIs
  3. Metrics OK → 25%50%100%
  4. Bad metrics → automatic rollback

Entire process: 30-60 minutes. Human intervention only on escalation. Argo Rollouts or Flagger for Kubernetes.

Feature Flags

Separating deploy from release:

  • Code in production, feature disabled — deploy without risk
  • Progressive enablement — internal → beta → 10% → 100%
  • A/B testing — variant A vs. B, metrics decide
  • Kill switch — instant disable of problematic feature
  • Targeting — feature only for specific users/segments

Tools: LaunchDarkly, Unleash, custom solutions. CI/CD integration — feature flag automatically created when merging PR with feature flag annotation.

Database Migrations

The riskiest part of deployment. Rules:

Expand-Contract Pattern: 1. Expand — Add new column, keep the old one 2. Migrate — Copy/transform data 3. Switch — Application reads from new column 4. Contract — Remove old column

No ALTER TABLE with locks. No DROP COLUMN during deployment. Every migration is rollbackable.

CI/CD Pipeline

┌─────────┐   ┌──────────┐   ┌──────────┐   ┌─────────┐   ┌──────────┐
│  Commit  │──▶│  Build   │──▶│  Test    │──▶│  Stage  │──▶│  Prod    │
│          │   │  + Lint  │   │  Unit    │   │  E2E    │   │  Canary  │
│          │   │  + SAST  │   │  Integ   │   │  Smoke  │   │  → Full  │
└─────────┘   └──────────┘   └──────────┘   └─────────┘   └──────────┘
                                                                │
                                                          Auto rollback
                                                          on threshold

Commit → production in < 15 minutes. Automated. With quality gates at every step.

Časté otázky

Expand-contract pattern. We add a new column (expand), migrate data, remove the old one (contract). No breaking changes, no table locks.

Yes. Blue-green deployment works for both monoliths and microservices. Feature flags separate deploy from release regardless of architecture.

Basic pipeline (build → test → deploy): 2-4 weeks. Full progressive delivery stack (canary, feature flags, automated rollback): 4-8 weeks.

Máte projekt?

Pojďme si o něm promluvit.

Domluvit schůzku