Slide 3 of 5

Key takeaways

What you learned

  1. Two approaches to cloud observability:

    • Store in Grafana Cloud: Full platform features, unified alerting, long-term retention
    • Query directly: Faster setup, data stays in cloud provider

  2. Collection methods for storing data:

    • Metrics: Scrape jobs (simple), Metric Streams (low latency), Alloy (flexible)
    • Logs: Lambda/Functions (serverless), Firehose (streaming), Alloy (universal)

  3. Cloud provider data sources:

    • AWS CloudWatch (metrics + logs)
    • Azure Monitor (metrics + logs)
    • GCP Cloud Monitoring (metrics) + Cloud Logging (logs)

The one thing to remember

Store data in Grafana Cloud for unified alerting, PromQL/LogQL, and full platform features. Query directly for quick setup when you don’t need those capabilities.

Script

Let’s bring it all together. You have two fundamental approaches to cloud provider observability.

The first is storing data in Grafana Cloud. Your metrics go to Mimir, your logs go to Loki, and you get the full power of the platform: unified querying with PromQL and LogQL, long-term retention, cross-cloud alerting, and all the features Grafana Cloud offers.

The second is querying cloud APIs directly through data sources: CloudWatch, Azure Monitor, GCP Cloud Monitoring and Cloud Logging. Data stays where it is, you’re just visualizing it. Faster to set up, but you miss some platform features.

For metrics collection, you learned about scrape jobs for simplicity, metric streams for low latency, and Grafana Alloy for flexibility.

For logs, similar patterns with Lambda functions, Firehose, Azure Functions, and Alloy.

Choose based on your latency needs, scale, and operational preferences. Most production environments go with the storage approach for the unified experience, but the direct query approach works great for quick wins or when you want to keep things simple.

/
Previous Next