Fleet Upgrades

In Development (Q4 2026)

This feature is currently in development. Fleet upgrade orchestration described here is subject to change.

Purpose: For platform engineers and operators, explains how to coordinate Kubernetes and platform service upgrades across a fleet of clusters safely.

Upgrade Strategy

Fleet upgrades use a wave-based promotion model:

Wave 1: Canary cluster (dev/staging)
  ↓ validate (automated health checks)
Wave 2: Non-critical production clusters
  ↓ validate (SLO checks, 24h soak)
Wave 3: Critical production clusters
  ↓ validate (full regression)
Wave 4: Regulated/air-gapped clusters

Fleet Upgrade Plan

apiVersion: fleet.opencenter.cloud/v1alpha1
kind: FleetUpgradePlan
metadata:
  name: k8s-1-34-upgrade
spec:
  targetVersion: "1.34"
  waves:
  - name: canary
    clusterSelector:
      matchLabels:
        upgrade-wave: canary
    validation:
      healthCheckDuration: 1h
      automated: true
  - name: production-standard
    clusterSelector:
      matchLabels:
        upgrade-wave: standard
    validation:
      healthCheckDuration: 24h
      automated: false  # Manual gate
  - name: production-critical
    clusterSelector:
      matchLabels:
        upgrade-wave: critical
    validation:
      healthCheckDuration: 48h
      automated: false

Validation Gates

Between waves, the system validates:

Check	Automated	Blocks Next Wave
Node health (Ready status)	Yes	Yes
Pod restart rate	Yes	Yes (if > threshold)
FluxCD reconciliation success	Yes	Yes
Prometheus alert firing	Yes	Yes (critical severity)
Custom health endpoint	Yes	Configurable
Manual approval	No	Yes (if configured)

Rollback

If validation fails:

• Wave halts automatically
• Affected clusters remain at current version
• Alert fires to fleet operators
• Manual rollback available via opencenter fleet upgrade rollback

Service Upgrades

Platform service upgrades (gitops-base tag bumps) follow the same wave model:

• Hub updates fleet GitOps repo with new tag
• FleetKustomizations propagate to clusters per wave schedule
• Each cluster's FluxCD reconciles the new service versions