AWS re:Invent 2025 - From Metrics to Management: Practical Observability on EKS (DEV202)

Summary of "From Metrics to Management: Practical Observability on EKS" (AWS re:Invent 2025)

Introduction

• Presenters: Dale Orers (Software Engineer) and Emil Labinsky (AWS DevOps Engineer)
• Agenda: Define observability, discuss making applications observable, explore AWS services for observability (Amazon Managed Service for Prometheus, Amazon Managed Grafana)

Understanding Observability

• Observability is about extracting actionable insights to assess and improve application performance, health, and behavior
• Observability is a continuous cycle of detecting metrics, investigating, remediating issues, and assessing improvements
• Key benefits of observability:
  • Gain visibility into application health
  • Improve troubleshooting and issue resolution
  • Deliver superior customer experience
  • Control costs

Observability Signals

• Metrics: Indicate the presence of a problem (e.g. 25% slower application)
• Traces: Identify the location of a problem (e.g. dependencies between services)
• Logs: Determine the cause of a problem (e.g. database startup issue after update)
• Profiles: Provide insight into code behavior (e.g. potential memory leak)
• Together, these signals enable better SLO/SLA compliance

Making Applications Observable

Amazon Managed Service for Prometheus

• Fully managed, serverless Prometheus-compatible monitoring service
• Uses Prometheus data model and querying language (PromQL)
• Enables seamless monitoring of containerized workloads
• Best practices:
  • Use private link and IAM for secure data transfer
  • Optimize scrape interval and relabel configs to reduce costs
  • Adjust retention period based on data requirements
  • Run multiple collector containers for high availability
  • Use consistent identifiers for data correlation

Amazon Managed Grafana

• Fully managed, open-source analytics and visualization platform
• Integrates with Amazon Managed Service for Prometheus
• Best practices for dashboards:
  • Each dashboard should tell a clear, specific story to answer a question
  • Design dashboards to be simple and responsive

Demonstration

• Provisioned a Kubernetes cluster with sample application (web-2048)
• Configured Amazon Managed Service for Prometheus:
  • Created a scraper to collect Prometheus metrics
  • Defined a workspace to store and query the collected data
  • Set appropriate retention period for cost management
• Configured Amazon Managed Grafana:
  • Created a data source pointing to Amazon Managed Service for Prometheus
  • Designed dashboards to visualize cluster and application metrics
  • Applied filters and labels to focus on specific data of interest

Key Takeaways

• Observability is a critical practice for ensuring application health, performance, and customer experience
• Amazon Managed Service for Prometheus and Amazon Managed Grafana provide fully managed, scalable solutions for implementing observability
• Best practices around scrape intervals, relabeling, retention, and dashboard design help optimize observability for cost and effectiveness
• Comprehensive observability, leveraging metrics, traces, logs, and profiles, enables better troubleshooting, SLO/SLA compliance, and overall application management