feat: Phase 7 - AI self healing with GPT-4 and Slack

Author: ch-Arunkumar

ai-healer/Dockerfile

@@ -0,0 +1,9 @@
+FROM python:3.11-slim
+WORKDIR /app
+RUN apt-get update && apt-get install -y wget curl && rm -rf /var/lib/apt/lists/*
+COPY requirements.txt .
+RUN pip install --no-cache-dir -r requirements.txt
+COPY src/ .
+RUN python -c "import flask; import openai; import kubernetes; print('All imports OK')"
+EXPOSE 5010
+CMD ["gunicorn", "--workers", "1", "--bind", "0.0.0.0:5010", "--timeout", "120", "healer:app"]

ai-healer/requirements.txt

@@ -0,0 +1,5 @@
+flask==3.0.3
+openai==0.28.0
+kubernetes==29.0.0
+requests==2.31.0
+gunicorn==22.0.0

ai-healer/src/healer.py

@@ -0,0 +1,113 @@
+from flask import Flask, request, jsonify
+import requests
+import os
+import json
+import logging
+import openai
+
+logging.basicConfig(level=logging.INFO)
+logger = logging.getLogger(__name__)
+
+app = Flask(__name__)
+
+openai.api_key = os.getenv("OPENAI_API_KEY")
+SLACK_WEBHOOK  = os.getenv("SLACK_WEBHOOK_URL")
+
+
+def ask_gpt4(pod_name, logs):
+    prompt = f"""
+    Kubernetes pod '{pod_name}' has crashed.
+    Last 50 lines of logs:
+    {logs}
+    Respond ONLY in valid JSON:
+    {{
+        "diagnosis": "what went wrong in 1-2 sentences",
+        "root_cause": "specific technical cause",
+        "fix": "restart_pod or manual_review",
+        "confidence": 0.95
+    }}
+    """
+    response = openai.ChatCompletion.create(
+        model="gpt-3.5-turbo",
+        messages=[
+            {"role": "system", "content": "You are a Kubernetes expert. Respond in valid JSON only."},
+            {"role": "user",   "content": prompt}
+        ],
+        temperature=0.1
+    )
+    raw = response.choices[0].message.content
+    raw = raw.replace("```json", "").replace("```", "").strip()
+    return json.loads(raw)
+
+
+def send_slack(pod_name, diagnosis, fix_applied, confidence):
+    if fix_applied:
+        color  = "#36a64f"
+        status = "✅ Auto-fixed by AI"
+    else:
+        color  = "#ff0000"
+        status = "🚨 Needs manual review"
+
+    message = {
+        "attachments": [{
+            "color": color,
+            "title": f"🤖 AI Healer — {pod_name}",
+            "fields": [
+                {"title": "Status",      "value": status,                              "short": True},
+                {"title": "Confidence",  "value": f"{int(confidence * 100)}%",         "short": True},
+                {"title": "Diagnosis",   "value": diagnosis.get("diagnosis", "Unknown"),"short": False},
+                {"title": "Root Cause",  "value": diagnosis.get("root_cause","Unknown"),"short": False},
+                {"title": "Action",      "value": diagnosis.get("fix", "None"),         "short": False}
+            ],
+            "footer": "AI DevOps Platform"
+        }]
+    }
+    try:
+        requests.post(SLACK_WEBHOOK, json=message)
+        logger.info("Slack notification sent")
+    except Exception as e:
+        logger.error(f"Slack failed: {e}")
+
+
+@app.route('/health')
+def health():
+    return jsonify({"status": "healthy", "service": "ai-healer"}), 200
+
+
+@app.route('/webhook', methods=['POST'])
+def handle_alert():
+    data = request.get_json()
+    if not data:
+        return jsonify({"error": "No data"}), 400
+
+    for alert in data.get('alerts', []):
+        labels    = alert.get('labels', {})
+        pod_name  = labels.get('pod', 'unknown-pod')
+        namespace = labels.get('namespace', 'default')
+
+        logger.info(f"Processing alert for pod: {pod_name}")
+
+        # Simulate logs since we are running locally
+        logs = f"Pod {pod_name} in namespace {namespace} crashed unexpectedly."
+
+        try:
+            diagnosis = ask_gpt4(pod_name, logs)
+        except Exception as e:
+            logger.error(f"GPT call failed: {e}")
+            diagnosis = {
+                "diagnosis": "AI analysis failed",
+                "root_cause": str(e),
+                "fix": "manual_review",
+                "confidence": 0.0
+            }
+
+        confidence  = diagnosis.get('confidence', 0.0)
+        fix_applied = False
+
+        send_slack(pod_name, diagnosis, fix_applied, confidence)
+
+    return jsonify({"status": "processed"}), 200
+
+
+if __name__ == '__main__':
+    app.run(host='0.0.0.0', port=5010)

k8s/ai-healer.yaml

@@ -0,0 +1,75 @@
+apiVersion: apps/v1
+kind: Deployment
+metadata:
+  name: ai-healer
+spec:
+  replicas: 1
+  selector:
+    matchLabels:
+      app: ai-healer
+  template:
+    metadata:
+      labels:
+        app: ai-healer
+    spec:
+      serviceAccountName: ai-healer-sa
+      containers:
+      - name: ai-healer
+        image: arunrox47/ai-healer:latest
+        ports:
+        - containerPort: 5010
+        env:
+        - name: OPENAI_API_KEY
+          valueFrom:
+            secretKeyRef:
+              name: ai-healer-secrets
+              key: OPENAI_API_KEY
+        - name: SLACK_WEBHOOK_URL
+          valueFrom:
+            secretKeyRef:
+              name: ai-healer-secrets
+              key: SLACK_WEBHOOK_URL
+        readinessProbe:
+          httpGet:
+            path: /health
+            port: 5010
+          initialDelaySeconds: 10
+          periodSeconds: 5
+---
+apiVersion: v1
+kind: Service
+metadata:
+  name: ai-healer
+spec:
+  selector:
+    app: ai-healer
+  ports:
+  - port: 5010
+    targetPort: 5010
+---
+apiVersion: v1
+kind: ServiceAccount
+metadata:
+  name: ai-healer-sa
+---
+apiVersion: rbac.authorization.k8s.io/v1
+kind: ClusterRole
+metadata:
+  name: ai-healer-role
+rules:
+- apiGroups: [""]
+  resources: ["pods", "pods/log"]
+  verbs: ["get", "list", "delete", "watch"]
+---
+apiVersion: rbac.authorization.k8s.io/v1
+kind: ClusterRoleBinding
+metadata:
+  name: ai-healer-binding
+roleRef:
+  apiGroup: rbac.authorization.k8s.io
+  kind: ClusterRole
+  name: ai-healer-role
+subjects:
+- kind: ServiceAccount
+  name: ai-healer-sa
+  namespace: default