adobe-load-scale

What this skill does

# Adobe Load & Scale

## Overview

Load testing and scaling strategies for Adobe API integrations. Adobe APIs are async and relatively slow (5-30s per operation), requiring different load testing approaches than typical REST APIs.

## Prerequisites

- k6 load testing tool installed (`npm install -g k6` or `brew install k6`)
- Adobe Developer Console credentials for testing (separate from production)
- Kubernetes cluster with HPA configured (for auto-scaling)
- Understanding of your Adobe API rate limits

## Instructions

### Step 1: k6 Load Test for Firefly API

```javascript
// adobe-firefly-load.js
import http from 'k6/http';
import { check, sleep } from 'k6';
import { Rate, Trend } from 'k6/metrics';

const errorRate = new Rate('adobe_errors');
const fireflyDuration = new Trend('firefly_duration');

export const options = {
  stages: [
    { duration: '1m', target: 2 },    // Warm up (Adobe APIs are slow)
    { duration: '3m', target: 5 },    // Steady state
    { duration: '2m', target: 10 },   // Stress (watch for 429s)
    { duration: '1m', target: 0 },    // Ramp down
  ],
  thresholds: {
    http_req_duration: ['p(95)<30000'],  // 30s — Firefly is async/slow
    adobe_errors: ['rate<0.05'],         // < 5% error rate
  },
};

// Pre-generate token (shared across VUs)
const TOKEN = __ENV.ADOBE_ACCESS_TOKEN;
const CLIENT_ID = __ENV.ADOBE_CLIENT_ID;

export default function () {
  const response = http.post(
    'https://firefly-api.adobe.io/v3/images/generate',
    JSON.stringify({
      prompt: `Load test image ${Date.now()}`,
      n: 1,
      size: { width: 512, height: 512 },  // Smallest size for speed
    }),
    {
      headers: {
        'Authorization': `Bearer ${TOKEN}`,
        'x-api-key': CLIENT_ID,
        'Content-Type': 'application/json',
      },
      timeout: '60s',
    }
  );

  const success = check(response, {
    'status is 200': (r) => r.status === 200,
    'status is not 429': (r) => r.status !== 429,
  });

  errorRate.add(!success);
  fireflyDuration.add(response.timings.duration);

  if (response.status === 429) {
    const retryAfter = parseInt(response.headers['Retry-After'] || '30');
    console.log(`Rate limited, waiting ${retryAfter}s`);
    sleep(retryAfter);
  } else {
    sleep(3); // Respect rate limits between requests
  }
}
```

### Step 2: k6 Load Test for PDF Services

```javascript
// adobe-pdf-load.js
import http from 'k6/http';
import { check, sleep } from 'k6';

export const options = {
  stages: [
    { duration: '1m', target: 3 },
    { duration: '5m', target: 10 },
    { duration: '1m', target: 0 },
  ],
  thresholds: {
    http_req_duration: ['p(95)<15000'],  // PDF ops are faster than Firefly
    http_req_failed: ['rate<0.02'],
  },
};

export default function () {
  // Test your app's PDF endpoint (which calls Adobe PDF Services internally)
  const response = http.post(
    `${__ENV.APP_URL}/api/extract-pdf`,
    http.file(open('./test-fixtures/sample-5page.pdf', 'b'), 'test.pdf'),
    { timeout: '30s' }
  );

  check(response, {
    'extraction successful': (r) => r.status === 200,
    'has text content': (r) => JSON.parse(r.body).text?.length > 0,
  });

  sleep(2);
}
```

### Step 3: Run Load Tests

```bash
# 1. Generate access token for load test
export ADOBE_ACCESS_TOKEN=$(curl -s -X POST \
  'https://ims-na1.adobelogin.com/ims/token/v3' \
  -d "client_id=${ADOBE_CLIENT_ID}&client_secret=${ADOBE_CLIENT_SECRET}&grant_type=client_credentials&scope=${ADOBE_SCOPES}" | jq -r '.access_token')

# 2. Run Firefly load test
k6 run --env ADOBE_ACCESS_TOKEN=${ADOBE_ACCESS_TOKEN} \
  --env ADOBE_CLIENT_ID=${ADOBE_CLIENT_ID} \
  adobe-firefly-load.js

# 3. Run PDF Services load test
k6 run --env APP_URL=https://staging.yourapp.com \
  adobe-pdf-load.js

# 4. Export results to InfluxDB for Grafana dashboards
k6 run --out influxdb=http://localhost:8086/k6 adobe-firefly-load.js
```

### Step 4: Kubernetes Auto-Scaling

```yaml
# k8s/adobe-hpa.yaml
apiVersion: autoscaling/v2
kind: HorizontalPodAutoscaler
metadata:
  name: adobe-service-hpa
spec:
  scaleTargetRef:
    apiVersion: apps/v1
    kind: Deployment
    name: adobe-service
  minReplicas: 2
  maxReplicas: 20
  metrics:
    - type: Resource
      resource:
        name: cpu
        target:
          type: Utilization
          averageUtilization: 60
    # Scale on pending Adobe jobs (custom metric from Prometheus)
    - type: Pods
      pods:
        metric:
          name: adobe_pending_jobs
        target:
          type: AverageValue
          averageValue: 5
  behavior:
    scaleUp:
      stabilizationWindowSeconds: 60
      policies:
        - type: Pods
          value: 2
          periodSeconds: 60
    scaleDown:
      stabilizationWindowSeconds: 300  # Slow scale-down (Adobe jobs are long)
```

### Step 5: Capacity Planning

```typescript
// Adobe-specific capacity considerations:
// 1. Each Firefly request blocks for 5-30s — need many concurrent workers
// 2. PDF Services is faster (2-10s) but has monthly transaction limits
// 3. Photoshop API jobs are async — workers poll and block on I/O
// 4. IMS token generation is shared — cache aggressively

interface AdobeCapacityPlan {
  api: string;
  peakRps: number;
  avgLatencyMs: number;
  concurrencyNeeded: number;  // peakRps * avgLatencyMs / 1000
  podsNeeded: number;         // concurrencyNeeded / connectionsPerPod
  monthlyTransactions: number;
  tierNeeded: string;
}

function planCapacity(metrics: {
  peakRps: number;
  avgLatencyMs: number;
  connectionsPerPod: number;
}): AdobeCapacityPlan {
  const concurrency = metrics.peakRps * metrics.avgLatencyMs / 1000;
  const pods = Math.ceil(concurrency / metrics.connectionsPerPod);

  return {
    api: 'firefly',
    peakRps: metrics.peakRps,
    avgLatencyMs: metrics.avgLatencyMs,
    concurrencyNeeded: Math.ceil(concurrency),
    podsNeeded: pods,
    monthlyTransactions: metrics.peakRps * 3600 * 8 * 22, // 8h/day, 22 days
    tierNeeded: pods > 5 ? 'Enterprise' : 'Pro',
  };
}

// Example: 2 RPS peak, 10s avg latency, 5 connections per pod
// concurrency = 2 * 10 = 20 concurrent requests
// pods = 20 / 5 = 4 pods minimum
```

## Output

- k6 load test scripts for Firefly and PDF Services
- Kubernetes HPA with Adobe-aware scaling metrics
- Capacity planning model accounting for async API latency
- Benchmark results template for documentation

## Error Handling

| Issue | Cause | Solution |
|-------|-------|----------|
| All requests 429 in k6 | Rate limit exceeded | Reduce VU count; add sleep |
| k6 timeout | Adobe API > 60s | Increase k6 request timeout |
| HPA not scaling | Custom metric not exposed | Verify Prometheus metric exists |
| Token expires mid-test | Long test duration | Token valid 24h; pre-generate |

## Resources

- [k6 Documentation](https://k6.io/docs/)
- [Kubernetes HPA](https://kubernetes.io/docs/tasks/run-application/horizontal-pod-autoscale/)
- [Firefly Async API](https://developer.adobe.com/firefly-services/docs/firefly-api/guides/how-tos/using-async-apis)

## Next Steps

For reliability patterns, see `adobe-reliability-patterns`.