Accelerating Innovation with AI-Powered Operations

Evolution of AI-Ops Services

1. Rule-based systems:

   • Precise with low error rates (e.g., AWS Config, Trusted Advisor)

2. Machine Learning for Anomaly Detection:

   • Provides probabilities, not certainties (like weather forecasts)

3. Current Industry State: AI for Incident Management

   • Ensemble of techniques (rule-based, ML, AI, generative AI) for incident correlation, root cause analysis, and remediation suggestions

4. Future Expectations:

   • Predictive insights to forecast future disruptions and capacity needs
   • Fully automated, self-healing systems (long-term goal)

Challenges Faced by Customers

1. Overwhelming data volume: Difficulty in knowing where to start investigations and identifying "what changed" in the application.
2. Proactive detection: Challenges in predicting and identifying unknown issues before they become critical.

Existing AI-Ops Capabilities in Amazon CloudWatch

1. Pattern Analytics: Automatically surfaces key patterns in log query results, helping to synthesize large volumes of data.
2. Compare Mode: Visually compares pattern trends across time periods to identify changes.
3. Anomaly Detection: Continuously evaluates incoming logs, compares to historical baselines, and proactively notifies of anomalous trends.
4. Natural Language Query Generation: Allows users to generate CloudWatch log and metric insights queries using natural language.

Introducing Amazon DevOps Guru

1. Motivation:

   • Lack of comprehensive data for triage
   • Difficulty in correlating signals to identify root causes
   • Alarm and tool fatigue
   • Scattered data across multiple tools

2. Working Backwards Process:

   • Identified key themes for an ideal AI-Ops experience
   • Automatic instrumentation and data collection
   • Detecting and prioritizing alerts
   • Guided root cause analysis and operational troubleshooting
   • Recommended actions
   • Continuous learning and improvement

3. Demonstration of Amazon DevOps Guru

   • Incident scenario: Increased load on a multi-tenant application, causing issues in the visit booking service
   • Amazon DevOps Guru automatically analyzes various telemetry signals and provides guided troubleshooting
   • Identifies root cause, suggests mitigation actions, and provides AWS-managed runbooks

Getting Started with Amazon DevOps Guru

1. Essentials:

   • Create an Investigation Group for common configuration
   • Enable application signals, AWS X-Ray, and upgrade CloudWatch/Fluent Bit agents

2. Best Practices:

   • Enable CloudTrail logs for change detection
   • Configure alarms to automatically create investigations

3. Entry Points:

   • Auto-triggered investigations from CloudWatch alarms
   • Initiate investigations from Q chat or embedded dashboards

Key Takeaways:

• AI-Ops is an evolving landscape, moving from rule-based to AI-powered Incident Management
• Amazon CloudWatch provides AI-Ops capabilities like Pattern Analytics, Compare Mode, and Anomaly Detection
• Amazon DevOps Guru offers a guided, AI-powered operational troubleshooting experience
• Getting started involves setting up the essentials and adopting best practices
• Amazon DevOps Guru can be accessed through alarms, Q chat, and embedded dashboards