Alerting & Health Checks

Configure Alertmanager, define SLOs, set up health checks, and implement monitoring best practices

Alerting with Alertmanager

Configure Alerts

prometheus-rules.yaml:

groups:
- name: langgraph_alerts
  rules:
  # High error rate
  - alert: HighErrorRate
    expr: rate(http_requests_total{status=~"5.."}[5m]) > 0.05
    for: 5m
    labels:
      severity: critical
    annotations:
      summary: "High error rate detected"
      description: "Error rate is {{ $value }} requests/second"

  # High latency
  - alert: HighLatency
    expr: histogram_quantile(0.95, rate(http_request_duration_seconds_bucket[5m])) > 2
    for: 10m
    labels:
      severity: warning
    annotations:
      summary: "High API latency"
      description: "P95 latency is {{ $value }}s"

  # LLM failures
  - alert: LLMFailures
    expr: rate(llm_requests_total{status="error"}[5m]) > 0.1
    for: 5m
    labels:
      severity: critical
    annotations:
      summary: "LLM requests failing"
      description: "LLM error rate: {{ $value }}"

  # Redis down
  - alert: RedisDown
    expr: up{job="redis"} == 0
    for: 1m
    labels:
      severity: critical
    annotations:
      summary: "Redis is down"
      description: "Redis instance {{ $labels.instance }} is unreachable"

  # High token usage
  - alert: HighTokenUsage
    expr: sum(rate(llm_tokens_total[1h])) > 1000000
    for: 10m
    labels:
      severity: warning
    annotations:
      summary: "Unusually high token usage"
      description: "Token usage: {{ $value }} tokens/hour"

  # Low session count (possible issue)
  - alert: NoActiveSessions
    expr: active_sessions == 0
    for: 30m
    labels:
      severity: warning
    annotations:
      summary: "No active sessions"
      description: "No users have active sessions for 30 minutes"

Alertmanager Configuration

global:
  slack_api_url: 'https://hooks.slack.com/services/YOUR/SLACK/WEBHOOK'

route:
  group_by: ['alertname']
  group_wait: 10s
  group_interval: 10s
  repeat_interval: 1h
  receiver: 'slack-notifications'

  routes:
  - match:
      severity: critical
    receiver: 'pagerduty'

receivers:
- name: 'slack-notifications'
  slack_configs:
  - channel: '#alerts'
    title: 'Alert: {{ .GroupLabels.alertname }}'
    text: '{{ range .Alerts }}{{ .Annotations.description }}{{ end }}'

- name: 'pagerduty'
  pagerduty_configs:
  - service_key: 'YOUR_PAGERDUTY_KEY'

Health Checks

Kubernetes Probes

apiVersion: apps/v1
kind: Deployment
spec:
  template:
    spec:
      containers:
      - name: agent
        livenessProbe:
          httpGet:
            path: /health/live
            port: 8000
          initialDelaySeconds: 30
          periodSeconds: 10
          failureThreshold: 3

        readinessProbe:
          httpGet:
            path: /health/ready
            port: 8000
          initialDelaySeconds: 10
          periodSeconds: 5
          failureThreshold: 3

        startupProbe:
          httpGet:
            path: /health/startup
            port: 8000
          initialDelaySeconds: 10
          periodSeconds: 5
          failureThreshold: 30

Health Check Implementation

from fastapi import FastAPI, status
from fastapi.responses import JSONResponse

@app.get("/health/live")
async def liveness():
    """Liveness probe - is the application running?"""
    return {"status": "alive"}

@app.get("/health/ready")
async def readiness():
    """Readiness probe - can the application serve traffic?"""

    checks = {}

    # Check Redis
    try:
        await redis_client.ping()
        checks["redis"] = "healthy"
    except Exception as e:
        checks["redis"] = f"unhealthy: {e}"

    # Check Keycloak
    try:
        response = await httpx.get(f"{settings.keycloak_url}/health")
        checks["keycloak"] = "healthy" if response.status_code == 200 else "unhealthy"
    except Exception as e:
        checks["keycloak"] = f"unhealthy: {e}"

    # Check OpenFGA
    try:
        response = await httpx.get(f"{settings.openfga_url}/healthz")
        checks["openfga"] = "healthy" if response.status_code == 200 else "unhealthy"
    except Exception as e:
        checks["openfga"] = f"unhealthy: {e}"

    # Determine overall status
    all_healthy = all(v == "healthy" for v in checks.values())

    return JSONResponse(
        content={"status": "ready" if all_healthy else "not_ready", "checks": checks},
        status_code=status.HTTP_200_OK if all_healthy else status.HTTP_503_SERVICE_UNAVAILABLE
    )

@app.get("/health/startup")
async def startup():
    """Startup probe - has the application finished starting?"""
    # Check if initialization is complete
    if not app_initialized:
        return JSONResponse(
            content={"status": "starting"},
            status_code=status.HTTP_503_SERVICE_UNAVAILABLE
        )

    return {"status": "started"}

Best Practices

Use Structured Logging

Always use structured JSON logging for easy parsing:

# Good
logger.info(
    "User authenticated",
    user_id=user_id,
    provider="keycloak",
    duration_ms=duration
)

# Bad
logger.info(f"User {user_id} authenticated via keycloak in {duration}ms")

Track Business Metrics

Monitor business KPIs, not just technical metrics:

• Conversations per user
• Average conversation length
• Token cost per user
• Tool usage patterns
• User retention

Set SLOs and Alerts

Define Service Level Objectives:

• Availability: 99.9% uptime
• Latency: P95 < 2s
• Error rate: < 0.1%
• Token budget: < $1000/day

Use Correlation IDs

Track requests across all services:

# Generate once per request
request_id = str(uuid.uuid4())

# Pass to all downstream services
headers = {"X-Request-ID": request_id}

# Log with correlation ID
logger.info("Processing request", request_id=request_id)

Next Steps

Scaling

Auto-scaling configuration

Disaster Recovery

Backup and recovery

Dashboards

Create visualization dashboards

Back to Overview

Return to monitoring overview

---

Monitoring Complete: Comprehensive observability with metrics, traces, logs, and alerts!