High Availability

Yuptime is designed as a single-instance application with built-in resilience.

Architecture

Yuptime uses a single scheduler with isolated workers pattern:

┌─────────────────────┐
│   Yuptime Pod       │  ← Single instance (uses Kubernetes Lease)
│   (Controller)      │
└─────────────────────┘
         │
         │ Creates Jobs
         ▼
┌─────────────────────────────────────────┐
│          Checker Job Pods               │  ← Multiple isolated workers
├─────────┬─────────┬─────────┬──────────┤
│ Check 1 │ Check 2 │ Check 3 │ Check N  │
└─────────┴─────────┴─────────┴──────────┘

Single Instance Guarantee

Yuptime uses a Kubernetes Lease to ensure only one scheduler runs:

# YuptimeSettings
spec:
  mode:
    singleInstanceRequired: true

If the pod fails, Kubernetes restarts it. The Lease prevents duplicate scheduling.

Resilience Features

Isolated Execution

Each check runs in its own pod:

• Failures don't cascade
• Resources are isolated
• Easy to debug

Stateless Design

No database means:

• Fast recovery after restart
• No data corruption risk
• State is in Kubernetes API (etcd)

Deterministic Jitter

Checks resume at consistent times after restart thanks to hash-based scheduling.

Recommendations

Resource Limits

resources:
  requests:
    cpu: 100m
    memory: 256Mi
  limits:
    cpu: 1000m
    memory: 1Gi

Pod Disruption Budget

podDisruptionBudget:
  enabled: true
  minAvailable: 1

Node Selection

Run on stable nodes:

nodeSelector:
  kubernetes.io/os: linux
  node-role.kubernetes.io/worker: ""

tolerations:
  - key: "CriticalAddonsOnly"
    operator: "Exists"

Monitoring Yuptime

Monitor Yuptime itself:

apiVersion: monitoring.yuptime.io/v1
kind: Monitor
metadata:
  name: yuptime-health
  namespace: yuptime
spec:
  type: http
  schedule:
    intervalSeconds: 30
    timeoutSeconds: 10
  target:
    http:
      url: "http://yuptime-api.yuptime:3000/health"
      method: GET