N8N Queue Mode - Cost Optimization Guide¶

Cost Breakdown¶

Worker Costs (Azure Container Apps)¶

Each N8N worker (1 vCPU, 2GB RAM) costs approximately: - ~$75/month running 24/7 - ~$2.50/day running 24/7 - ~$0.10/hour when running

Configuration Options¶

Option 1: Min=0, Max=10 ⭐ RECOMMENDED FOR COST SAVINGS¶

# Deploy with zero minimum workers
./scripts/deploy-n8n-queue-mode.sh deploy-workers
# Or set in environment
export WORKER_MIN_REPLICAS=0
export WORKER_MAX_REPLICAS=10

Monthly Costs: - Main instance: $30/month (always running) - Workers when idle: $0/month - Workers during business hours (avg 2): ~$150/month - Total idle cost: $30/month - Total business hours: $180/month

Annual Savings: ~$1,440/year vs Min=2 (assuming 50% idle time)

Best for: - ✅ Development/staging environments - ✅ Intermittent workflow usage - ✅ Scheduled workflows (predictable timing) - ✅ Cost-sensitive deployments

Trade-off: - ⚠️ Cold start delay: 30-60 seconds for first workflow after idle period - First worker takes time to spin up, pull image, and start

Mitigation: - Use scheduled workflows to "warm up" workers before peak hours - Set higher min replicas during business hours only (see schedule-based scaling below)

Option 2: Min=1, Max=10 (Balanced)¶

export WORKER_MIN_REPLICAS=1
export WORKER_MAX_REPLICAS=10
./scripts/deploy-n8n-queue-mode.sh deploy-workers

Monthly Costs: - Main instance: $30/month - 1 worker (always): $75/month - Additional workers during peaks: $0-675/month - Total baseline: $105/month

Annual Savings: ~$540/year vs Min=2

Best for: - ✅ Production with time-sensitive workflows - ✅ Need immediate execution without cold start - ✅ Balance between cost and performance

Option 3: Min=2, Max=10 (High Availability)¶

export WORKER_MIN_REPLICAS=2
export WORKER_MAX_REPLICAS=10
./scripts/deploy-n8n-queue-mode.sh deploy-workers

Monthly Costs: - Main instance: $30/month - 2 workers (always): $150/month - Additional workers during peaks: $0-600/month - Total baseline: $180/month

Best for: - ✅ Production with high availability requirements - ✅ Parallel workflows needed immediately - ✅ Redundancy if one worker fails

Advanced Cost Optimization Strategies¶

1. Schedule-Based Scaling (Business Hours Only)¶

Use Azure CLI to change scaling based on time of day:

Morning Scale-Up (8 AM, weekdays):

# Run this via Azure Automation or GitHub Actions on schedule
az containerapp update \
  --name slidefactory-n8n-worker \
  --resource-group rg-slidefactory-prod-001 \
  --min-replicas 2 \
  --max-replicas 10

Evening Scale-Down (6 PM, weekdays):

az containerapp update \
  --name slidefactory-n8n-worker \
  --resource-group rg-slidefactory-prod-001 \
  --min-replicas 0 \
  --max-replicas 5

Weekend Scale-Down:

az containerapp update \
  --name slidefactory-n8n-worker \
  --resource-group rg-slidefactory-prod-001 \
  --min-replicas 0 \
  --max-replicas 2

Potential Savings: Up to 70% reduction in worker costs

2. Queue-Based Scaling (Future Enhancement)¶

Add custom scaling rule based on Redis queue depth:

az containerapp update \
  --name slidefactory-n8n-worker \
  --resource-group rg-slidefactory-prod-001 \
  --scale-rule-name queue-depth \
  --scale-rule-type azure-queue \
  --scale-rule-metadata "queueName=bull:n8n:waiting" "queueLength=10"

Scale up when queue has 10+ pending jobs.

3. Resource Right-Sizing¶

If workflows are lightweight, reduce worker resources:

export WORKER_CPU="0.5"
export WORKER_MEMORY="1.0Gi"
./scripts/deploy-n8n-queue-mode.sh deploy-workers

Cost per worker: ~$37.50/month (50% savings per worker)

Warning: Only if your workflows don't need full resources.

4. Azure Reserved Instances¶

For predictable baseline (e.g., Min=1): - Purchase 1-year or 3-year Azure Reserved Capacity - Savings: 30-50% discount on committed resources - Example: 1 worker reserved = ~$25-40/month instead of $75/month

5. Development/Staging Optimization¶

For non-production environments:

# Ultra-low cost config
export WORKER_MIN_REPLICAS=0
export WORKER_MAX_REPLICAS=2
export WORKER_CPU="0.5"
export WORKER_MEMORY="1.0Gi"

Cost: $30/month (main only) when idle, $105/month at peak (2 workers)

Cost Comparison: Annual Estimate¶

Note: Min=0 with queue mode is still more expensive than current regular mode, but provides 3-5x throughput during business hours.

Recommended Strategy by Environment¶

Production¶

WORKER_MIN_REPLICAS=1  # No cold start
WORKER_MAX_REPLICAS=10  # Handle bursts

Preview/Staging¶

WORKER_MIN_REPLICAS=0  # Save money when idle
WORKER_MAX_REPLICAS=5   # Limit max spend

Development¶

WORKER_MIN_REPLICAS=0
WORKER_MAX_REPLICAS=2
WORKER_CPU=0.5
WORKER_MEMORY=1.0Gi

Monitoring Costs¶

View Current Worker Count¶

az containerapp revision list \
  --name slidefactory-n8n-worker \
  --resource-group rg-slidefactory-prod-001 \
  --query "[?properties.active].{name:name, replicas:properties.replicas}" \
  -o table

Check Scaling Events¶

az monitor activity-log list \
  --resource-group rg-slidefactory-prod-001 \
  --namespace Microsoft.App/containerApps \
  --max-events 50 \
  --query "[?contains(resourceId, 'slidefactory-n8n-worker')]" \
  -o table

Set Budget Alert¶

az consumption budget create \
  --budget-name n8n-workers-monthly \
  --amount 300 \
  --time-grain Monthly \
  --resource-group rg-slidefactory-prod-001 \
  --notifications enabled=true threshold=80 contactEmails=your-email@example.com

Cold Start Deep Dive¶

What Happens with Min=0?¶

When no workers are running and a workflow is triggered:

1. N8N Main receives workflow execution request
2. N8N Main publishes job to Redis queue (bull:n8n:waiting)
3. Azure Container Apps detects queue activity (if using queue-based scaling) or other trigger
4. Azure spins up first worker container (~30-60 seconds)
5. Pull N8N Docker image (if not cached)
6. Start container
7. N8N worker connects to Redis
8. Worker connects to PostgreSQL
9. Worker starts listening for jobs
10. Worker pulls job from queue and executes workflow

Total delay: 30-90 seconds for first workflow after idle period

Reducing Cold Start Impact¶

Option 1: Warm-up Workflow Create a scheduled N8N workflow that runs every 15 minutes during business hours:

Trigger: Cron (*/15 * * * * Mon-Fri 8-18)
Action: Ping (simple HTTP request)

Option 2: Pre-warming Script Run before known workflow executions:

# Execute a dummy workflow to warm up workers
curl -X POST https://slidefactory-n8n.../api/v1/workflows/dummy-warmup/execute \
  -H "X-N8N-API-KEY: $N8N_API_KEY"

Option 3: Use Min=1 During Business Hours Via scheduled Azure Automation or GitHub Actions.

Decision Matrix¶

Quick Commands¶

Deploy with Min=0 (Cost Optimized)¶

export WORKER_MIN_REPLICAS=0
export WORKER_MAX_REPLICAS=10
./scripts/deploy-n8n-queue-mode.sh deploy-workers

Deploy with Min=1 (Balanced)¶

export WORKER_MIN_REPLICAS=1
export WORKER_MAX_REPLICAS=10
./scripts/deploy-n8n-queue-mode.sh deploy-workers

Change Scaling On-the-Fly¶

az containerapp update \
  --name slidefactory-n8n-worker \
  --resource-group rg-slidefactory-prod-001 \
  --min-replicas 0

Summary¶

💰 Maximum Cost Savings: Min=0, Max=10 - Annual savings: ~$1,440/year vs Min=2 - Best for: Intermittent usage, development/staging - Trade-off: 30-60 second cold start

⚖️ Balanced: Min=1, Max=10 (Recommended for Production) - Annual savings: ~$540/year vs Min=2 - Best for: Production with time-sensitive workflows - Trade-off: Small baseline cost, no cold start

🚀 High Availability: Min=2, Max=10 - Best for: Critical production workloads - Trade-off: Higher cost, immediate redundancy

Last Updated: 2025-11-25