High Availability OpenClaw with Kubernetes

💡 Quick Answer: OpenClaw runs as a single-instance gateway (WhatsApp/Signal require exclusive connections), but you can achieve high availability through Kubernetes restart policies, health probes, PodDisruptionBudgets, PVC backups, and monitoring alerts. The key is fast recovery, not horizontal scaling.

Key concept: HA for OpenClaw = fast automatic recovery + state persistence + monitoring, not multiple replicas (messaging protocols don’t allow it).

Gotcha: Don’t set replicas: 2 — messaging channels like WhatsApp and Discord only allow one active connection per token/session.

The Problem

• If the OpenClaw pod dies, your AI assistant goes offline
• Pod rescheduling can take minutes if nodes are full
• PVC data loss means re-pairing all messaging channels
• No alerting means you won’t know the bot is down

The Solution

Use Kubernetes health probes, priority classes, PDB, and monitoring for fast automatic recovery.

Architecture Overview

flowchart TB
    subgraph cluster["☸️ KUBERNETES CLUSTER"]
        subgraph ha["HA Configuration"]
            DEP["Deployment<br/>replicas: 1<br/>Recreate"]
            HP["Health Probes<br/>Liveness + Readiness"]
            PC["PriorityClass<br/>High Priority"]
            PDB["PodDisruptionBudget<br/>minAvailable: 1"]
        end
        subgraph backup["Backup"]
            PVC["PVC<br/>State"]
            SNAP["VolumeSnapshot<br/>CronJob"]
            S3["S3 Backup"]
        end
        subgraph monitoring["Monitoring"]
            PROM["Prometheus"]
            ALERT["AlertManager"]
        end
    end
    
    DEP --> HP
    DEP --> PC & PDB
    PVC --> SNAP --> S3
    DEP --> PROM --> ALERT

Step 1: HA Deployment Configuration

# openclaw-ha.yaml
apiVersion: scheduling.k8s.io/v1
kind: PriorityClass
metadata:
  name: openclaw-critical
value: 1000000
globalDefault: false
description: "OpenClaw gateway — keep running at all costs"
---
apiVersion: policy/v1
kind: PodDisruptionBudget
metadata:
  name: openclaw-pdb
  namespace: openclaw
spec:
  minAvailable: 1
  selector:
    matchLabels:
      app: openclaw
---
apiVersion: apps/v1
kind: Deployment
metadata:
  name: openclaw-gateway
  namespace: openclaw
spec:
  replicas: 1
  strategy:
    type: Recreate
  selector:
    matchLabels:
      app: openclaw
  template:
    metadata:
      labels:
        app: openclaw
    spec:
      priorityClassName: openclaw-critical
      terminationGracePeriodSeconds: 30
      containers:
        - name: openclaw
          image: node:22-slim
          command: ["sh", "-c", "npm i -g openclaw@latest && openclaw gateway"]
          ports:
            - containerPort: 18789
          livenessProbe:
            httpGet:
              path: /
              port: 18789
            initialDelaySeconds: 60
            periodSeconds: 30
            failureThreshold: 3
            timeoutSeconds: 5
          readinessProbe:
            httpGet:
              path: /
              port: 18789
            initialDelaySeconds: 30
            periodSeconds: 10
          startupProbe:
            httpGet:
              path: /
              port: 18789
            initialDelaySeconds: 10
            periodSeconds: 5
            failureThreshold: 30    # 2.5 min for npm install + startup
          resources:
            requests:
              cpu: 250m
              memory: 512Mi
            limits:
              cpu: "1"
              memory: 1Gi

Step 2: Automated PVC Backup

# openclaw-backup-cronjob.yaml
apiVersion: batch/v1
kind: CronJob
metadata:
  name: openclaw-backup
  namespace: openclaw
spec:
  schedule: "0 */6 * * *"    # Every 6 hours
  jobTemplate:
    spec:
      template:
        spec:
          containers:
            - name: backup
              image: amazon/aws-cli:latest
              command: ["sh", "-c"]
              args:
                - |
                  TIMESTAMP=$(date +%Y%m%d-%H%M%S)
                  cd /backup-source
                  tar czf /tmp/openclaw-backup-${TIMESTAMP}.tar.gz .
                  aws s3 cp /tmp/openclaw-backup-${TIMESTAMP}.tar.gz \
                    s3://my-backups/openclaw/openclaw-backup-${TIMESTAMP}.tar.gz
                  # Retain last 7 days
                  aws s3 ls s3://my-backups/openclaw/ | \
                    sort | head -n -28 | awk '{print $4}' | \
                    xargs -I{} aws s3 rm s3://my-backups/openclaw/{}
              envFrom:
                - secretRef:
                    name: aws-backup-credentials
              volumeMounts:
                - name: state
                  mountPath: /backup-source
                  readOnly: true
          volumes:
            - name: state
              persistentVolumeClaim:
                claimName: openclaw-state
          restartPolicy: OnFailure

Step 3: Monitoring and Alerts

# openclaw-monitoring.yaml
apiVersion: monitoring.coreos.com/v1
kind: PrometheusRule
metadata:
  name: openclaw-alerts
  namespace: openclaw
spec:
  groups:
    - name: openclaw
      rules:
        - alert: OpenClawDown
          expr: |
            absent(up{job="openclaw"}) or up{job="openclaw"} == 0
          for: 2m
          labels:
            severity: critical
          annotations:
            summary: "OpenClaw gateway is down"
        
        - alert: OpenClawHighRestarts
          expr: |
            increase(kube_pod_container_status_restarts_total{container="openclaw"}[1h]) > 3
          labels:
            severity: warning
          annotations:
            summary: "OpenClaw pod restarting frequently"
        
        - alert: OpenClawPVCNearFull
          expr: |
            kubelet_volume_stats_used_bytes{persistentvolumeclaim="openclaw-state"} /
            kubelet_volume_stats_capacity_bytes{persistentvolumeclaim="openclaw-state"} > 0.85
          labels:
            severity: warning
          annotations:
            summary: "OpenClaw PVC is 85% full"

Common Issues

Issue 1: Pod stuck in Pending after node failure

# High PriorityClass ensures OpenClaw is scheduled first
# If still pending, check node resources:
kubectl describe pod -n openclaw -l app=openclaw
kubectl get nodes -o wide

Issue 2: Health probe fails during model loading

# Use startupProbe with generous timeout
startupProbe:
  failureThreshold: 60    # 5 minutes
  periodSeconds: 5

Best Practices

1. Use PriorityClass — Ensure OpenClaw is never evicted for lower-priority workloads
2. PodDisruptionBudget — Prevent voluntary disruptions (drains, upgrades) from killing the pod
3. Backup PVC every 6 hours — WhatsApp sessions and memory are hard to recreate
4. Monitor pod health — Alert immediately when the gateway goes down
5. Pre-build a custom image — Eliminate npm install time from cold starts

Key Takeaways

• Single replica with fast recovery is the HA pattern for OpenClaw
• PriorityClass + PDB keep the pod running through cluster disruptions
• PVC backups protect against data loss and enable disaster recovery
• Health probes enable automatic restart on failures
• Monitoring alerts ensure you know when the bot is down