railway-observability（铁路可观测性）

Railway Observability

Comprehensive guide for Railway.com observability including built-in metrics, customizable dashboards, alerting (Pro plan), and external OTEL integration with Grafana/Prometheus/Loki/Tempo.

Overview

Railway provides multi-tier observability capabilities:

• Built-in metrics (CPU, memory, disk, network)
• Customizable dashboards with drag-and-drop widgets
• Alerting with email and webhook integrations (Pro plan)
• External OTEL integration via Grafana Alloy
• Full observability stack deployment (Grafana template)

Keywords: metrics, monitoring, observability, dashboard, alerts, Grafana, Prometheus, Loki, Tempo, OTEL, Alloy, Railway

When to Use This Skill

• Setting up monitoring for Railway services
• Creating custom observability dashboards
• Configuring alerts for resource thresholds
• Integrating external monitoring systems
• Deploying Grafana observability stack
• Analyzing service performance metrics
• Troubleshooting resource issues
• Setting up distributed tracing

Quick Start

1. Access Built-in Observability Dashboard

Navigate to your Railway project:

Railway Dashboard → Project → Service → Metrics Tab

What you see:

• CPU usage (% of allocated resources)
• Memory usage (MB/GB over time)
• Disk usage (storage consumption)
• Network I/O (ingress/egress)
• 30-day metric retention

2. Configure Custom Dashboard Widgets

Add and customize metric widgets:

Metrics Tab → Add Widget → Select Metric Type

Available widgets:

• Line charts (time-series metrics)
• Bar charts (deployment comparisons)
• Gauges (current values)
• Tables (multi-metric views)

Customization:

• Drag-and-drop layout
• Adjustable time ranges (1h, 6h, 24h, 7d, 30d)
• Multi-replica aggregation
• Deployment markers on charts

3. Set Up Alerting (Pro Plan Required)

Configure alerts for threshold violations:

Service Settings → Alerts → Create Alert Rule

Alert types:

• CPU threshold (e.g., > 80% for 5 minutes)
• Memory threshold (e.g., > 90% for 10 minutes)
• Disk usage (e.g., > 85%)
• Custom metric thresholds

Notification channels:

• Email (default)
• Discord webhook
• Slack webhook
• Custom webhook (JSON payload)

4. Integrate External OTEL Collectors

Export metrics to external systems:

Service Settings → Observability → OTEL Integration

Configure environment variables:

OTEL_EXPORTER_OTLP_ENDPOINT=https://your-collector:4318
OTEL_EXPORTER_OTLP_HEADERS=Authorization=Bearer <token>
OTEL_SERVICE_NAME=my-railway-service

See references/otel-integration.md for complete setup.

5. Deploy Full Grafana Stack

Use Railway template for complete observability:

# Option 1: Deploy via Railway Dashboard
# Template ID: 8TLSQD (Grafana Stack)
# Includes: Grafana, Prometheus, Loki, Tempo, Alloy

# Option 2: Deploy via script
.claude/skills/railway-observability/scripts/deploy-grafana-stack.sh

Stack components:

• Grafana (visualization)
• Prometheus (metrics)
• Loki (logs)
• Tempo (traces)
• Alloy (collector)

Workflow

Step 1: Access Built-in Metrics

Railway provides instant metrics without configuration.

Navigate to metrics:

Project → Service → Metrics

Available metrics:

• CPU: % usage, allocated cores
• Memory: MB used, % of limit
• Disk: GB used, read/write IOPS
• Network: MB/s ingress/egress

Retention: 30 days for all metrics

Step 2: Customize Dashboard

Create personalized monitoring views.

Add widgets:

1. Click "Add Widget"
2. Select metric type
3. Configure aggregation
4. Set time range
5. Position via drag-and-drop

Best practices:

• Group related metrics (CPU + Memory)
• Use deployment markers for correlation
• Configure multi-replica views
• Set appropriate time ranges

Step 3: Configure Alerts (Pro Plan)

Set up proactive monitoring.

Create alert rule:

Service → Settings → Alerts → New Rule

Alert configuration:

Metric: CPU Usage
Condition: Greater than 80%
Duration: 5 minutes
Notification: Slack webhook

Webhook payload example:

{
  "service": "backend-production",
  "metric": "cpu_usage",
  "threshold": 80,
  "current": 87.5,
  "timestamp": "2025-11-26T10:30:00Z"
}

See references/dashboard-widgets.md for all alert types.

Step 4: Integrate OTEL Exporters

Send metrics to external systems.

Configure Alloy collector:

# Use template from templates/alloy-config.river
# Deploy as Railway service
# Configure OTEL endpoints

Environment setup:

# In your Railway service
OTEL_EXPORTER_OTLP_ENDPOINT=http://alloy:4318
OTEL_EXPORTER_OTLP_PROTOCOL=http/protobuf
OTEL_METRICS_EXPORTER=otlp
OTEL_LOGS_EXPORTER=otlp
OTEL_TRACES_EXPORTER=otlp

Verify integration:

# Check Alloy logs
railway logs -s alloy

# Should show: "Successfully received OTLP metrics"

See references/otel-integration.md for complete guide.

Step 5: Deploy Grafana Observability Stack

Full monitoring solution on Railway.

Deploy stack:

# Run deployment script
cd .claude/skills/railway-observability/scripts
./deploy-grafana-stack.sh

# Or deploy manually via Railway Dashboard
# Template: 8TLSQD (Grafana Stack)

Stack includes:

• Grafana (port 3000) - Dashboards
• Prometheus (port 9090) - Metrics storage
• Loki (port 3100) - Log aggregation
• Tempo (port 3200) - Distributed tracing
• Alloy (port 4318) - OTLP collector

Access Grafana:

URL: https://<grafana-service>.up.railway.app
Username: admin
Password: (set during deployment)

Architecture

Progressive Disclosure Structure

1. SKILL.md (this file): Quick start and workflow
2. references/: Detailed documentation

- metrics-reference.md - Complete metrics catalog - dashboard-widgets.md - Widget configuration guide - otel-integration.md - External integration setup

1. scripts/: Automation tools

- deploy-grafana-stack.sh - Deploy observability stack

1. templates/: Configuration files

- alloy-config.river - Grafana Alloy collector config

Key Features

Built-in Metrics (Always Available)

No configuration required - Metrics collected automatically for all services.

Dashboard Customization

Drag-and-drop widgets:

• Resize and reposition
• Multiple chart types
• Custom time ranges
• Deployment markers

Multi-replica support:

• Aggregated views (avg, min, max, sum)
• Per-replica breakdowns
• Replica scaling events

Time range options:

• Last 1 hour
• Last 6 hours
• Last 24 hours
• Last 7 days
• Last 30 days
• Custom range

Alerting (Pro Plan Only)

Threshold alerts:

• CPU > X% for Y minutes
• Memory > X MB for Y minutes
• Disk > X% capacity
• Network > X MB/s

Notification channels:

# Email
alert@example.com

# Discord webhook
https://discord.com/api/webhooks/...

# Slack webhook
https://hooks.slack.com/services/...

# Custom webhook
https://your-api.com/alerts

Alert states:

• Pending (condition met, waiting duration)
• Firing (threshold exceeded)
• Resolved (back to normal)

OTEL Integration

Supported protocols:

• OTLP/HTTP (port 4318)
• OTLP/gRPC (port 4317)

Signal types:

• Metrics (Prometheus format)
• Logs (JSON structured)
• Traces (Jaeger/Zipkin format)

Collector options:

• Grafana Alloy (recommended)
• OpenTelemetry Collector
• Custom OTLP exporters

Grafana Stack Details

Template Deployment (8TLSQD)

One-click deployment:

Railway Dashboard → New Project → Deploy Template → Search "8TLSQD"

Services deployed:

1. Grafana (visualization)
2. Prometheus (time-series DB)
3. Loki (log aggregation)
4. Tempo (trace storage)
5. Alloy (OTLP collector)

Configuration:

• Pre-configured data sources
• Sample dashboards included
• Persistent storage volumes
• Internal networking enabled

Alloy Collector

Purpose: Receive OTLP signals from Railway services and forward to Grafana stack.

Configuration (templates/alloy-config.river):

// OTLP receiver
otelcol.receiver.otlp "default" {
  grpc {
    endpoint = "0.0.0.0:4317"
  }
  http {
    endpoint = "0.0.0.0:4318"
  }

  output {
    metrics = [otelcol.exporter.prometheus.default.input]
    logs    = [otelcol.exporter.loki.default.input]
    traces  = [otelcol.exporter.otlp.tempo.input]
  }
}

// Prometheus exporter
otelcol.exporter.prometheus "default" {
  forward_to = [prometheus.remote_write.railway.receiver]
}

// Loki exporter
otelcol.exporter.loki "default" {
  forward_to = [loki.write.railway.receiver]
}

// Tempo exporter
otelcol.exporter.otlp "tempo" {
  client {
    endpoint = "tempo:4317"
  }
}

Grafana Dashboards

Pre-built dashboards:

• Railway Service Metrics (CPU, Memory, Disk, Network)
• OTLP Metrics Overview
• Log Analytics (Loki)
• Trace Explorer (Tempo)

Custom dashboards:

• Create via Grafana UI
• Import from Grafana.com
• Build with JSON models

Integration Examples

Example 1: Monitor Node.js Service

# Install OTEL SDK
npm install @opentelemetry/sdk-node @opentelemetry/auto-instrumentations-node

# Configure OTEL (in Railway service)
export OTEL_EXPORTER_OTLP_ENDPOINT=http://alloy:4318
export OTEL_SERVICE_NAME=nodejs-backend
export OTEL_METRICS_EXPORTER=otlp
export OTEL_LOGS_EXPORTER=otlp
export OTEL_TRACES_EXPORTER=otlp

# Auto-instrumentation
node --require @opentelemetry/auto-instrumentations-node/register app.js

View in Grafana:

• Metrics: Grafana → Dashboards → Railway Service Metrics
• Logs: Grafana → Explore → Loki
• Traces: Grafana → Explore → Tempo

Example 2: Alert on High Memory

# Pro Plan: Service → Alerts → New Rule
Name: High Memory Alert
Metric: Memory Usage
Condition: Greater than 512 MB
Duration: 10 minutes
Notification: Slack webhook
Webhook URL: https://hooks.slack.com/services/YOUR/WEBHOOK/URL

Slack notification:

{
  "text": "🚨 High Memory Alert",
  "blocks": [
    {
      "type": "section",
      "text": {
        "type": "mrkdwn",
        "text": "*Service*: backend-production\n*Memory*: 567 MB (> 512 MB threshold)\n*Duration*: 12 minutes"
      }
    }
  ]
}

Example 3: Custom Metrics from Python

from opentelemetry import metrics
from opentelemetry.sdk.metrics import MeterProvider
from opentelemetry.exporter.otlp.proto.http.metric_exporter import OTLPMetricExporter

# Configure OTLP exporter
exporter = OTLPMetricExporter(
    endpoint="http://alloy:4318/v1/metrics"
)

# Create meter provider
provider = MeterProvider(metric_readers=[
    PeriodicExportingMetricReader(exporter, export_interval_millis=60000)
])
metrics.set_meter_provider(provider)

# Create custom metrics
meter = metrics.get_meter(__name__)
request_counter = meter.create_counter("api_requests_total")
response_time = meter.create_histogram("api_response_time_seconds")

# Record metrics
request_counter.add(1, {"endpoint": "/api/users", "method": "GET"})
response_time.record(0.125, {"endpoint": "/api/users"})

View in Grafana:

Explore → Prometheus → Metrics Browser → api_requests_total

Cross-References

• railway-api: Query metrics via GraphQL API
• railway-logs: Log aggregation (complements metrics)
• railway-deployment: Deployment marker correlation
• observability-stack-setup: Local LGTM stack (similar to Railway stack)

Troubleshooting

Metrics Not Showing

Check service status:

railway status -s <service-name>

Verify metrics enabled:

• Metrics are always enabled
• Check service is running
• Wait 1-2 minutes for first data points

Alerts Not Firing

Requirements:

• Pro plan subscription required
• Alert rule properly configured
• Condition met for full duration
• Valid notification channel

Debug checklist:

1. Verify Pro plan active
2. Check threshold configuration
3. Confirm duration setting
4. Test webhook URL manually
5. Check Railway dashboard for alert status

OTEL Integration Issues

Common problems:

• Incorrect endpoint URL
• Network connectivity (use internal Railway URLs)
• Authentication headers missing
• Protocol mismatch (HTTP vs gRPC)

Verify Alloy receiving data:

# Check Alloy logs
railway logs -s alloy

# Look for:
# ✅ "OTLP receiver started"
# ✅ "Received X metric points"
# ❌ "Connection refused" = endpoint issue
# ❌ "Unauthorized" = auth issue

See references/otel-integration.md for detailed troubleshooting.

Best Practices

Dashboard Design

1. Group related metrics (CPU + Memory together)
2. Use deployment markers to correlate changes
3. Set appropriate time ranges (7d for trends, 1h for incidents)
4. Create separate dashboards for different services
5. Use multi-replica views for scaled services

Alerting Strategy

1. Set thresholds based on actual usage patterns
2. Use duration to avoid false positives (5-10 min)
3. Test notification channels before production
4. Document escalation procedures
5. Review and tune alerts regularly

OTEL Integration

1. Use internal Railway URLs (no egress costs)
2. Batch metrics for efficiency (60s intervals)
3. Add service tags for filtering
4. Use sampling for high-volume traces
5. Monitor collector resource usage

Grafana Stack

1. Enable persistent volumes for data retention
2. Configure backup strategy
3. Secure Grafana with strong password
4. Use Railway internal networking
5. Monitor stack resource consumption

Quick Reference

Accessing Observability

# Railway Dashboard
https://railway.app/project/<project-id>/service/<service-id>/metrics

# Via Railway CLI
railway status -s <service-name>
railway metrics -s <service-name>

Common Alert Thresholds

OTEL Environment Variables

OTEL_EXPORTER_OTLP_ENDPOINT=http://alloy:4318
OTEL_EXPORTER_OTLP_PROTOCOL=http/protobuf
OTEL_SERVICE_NAME=my-service
OTEL_METRICS_EXPORTER=otlp
OTEL_LOGS_EXPORTER=otlp
OTEL_TRACES_EXPORTER=otlp

Grafana Stack Template

Template ID: 8TLSQD
Components: Grafana, Prometheus, Loki, Tempo, Alloy
Deployment: Railway Dashboard → New Project → Deploy Template
Cost: ~$20-30/month (depends on usage)

Resources

• Railway Observability Docs: https://docs.railway.com/reference/observability
• Grafana Template: https://railway.app/template/8TLSQD
• OTEL Documentation: https://opentelemetry.io/docs/
• Grafana Alloy: https://grafana.com/docs/alloy/

Notes

• Built-in metrics are free for all plans
• Alerting requires Pro plan ($20/month)
• OTEL integration works on all plans
• Grafana stack deploys as separate Railway services
• Metrics retained for 30 days
• Use internal Railway URLs for OTEL (no egress costs)
• Deployment markers automatically added to charts
• Multi-replica metrics aggregated automatically

Updated: November 26, 2025 Template ID: 8TLSQD (Grafana Stack) Retention: 30 days (built-in metrics)