In the high-stakes world of software deployment, traditional verification methods fall short of ensuring robust, reliable, and safe releases. Harness's Continuous Verification (CV) leverages Machine Learning and represents a paradigm shift in approaching deployment reliability. This talk will unveil how Harness CV leverages advanced ML algorithms to transform deployment strategies, providing unprecedented insights into service performance, error detection, and risk mitigation.

By integrating seamlessly with APM and logging tools, Harness CV goes beyond simple threshold monitoring. It creates an intelligent, adaptive verification framework that learns from each deployment, automatically identifies anomalies, and can trigger immediate rollbacks when potential issues are detected. From canary and blue-green deployments to rolling updates, we'll explore how machine learning is revolutionizing the way organizations ensure software quality and minimize deployment risks.

Talk Overview

• Duration: 35 minutes
• Target Audience: DevOps engineers, SREs, Software Architects, Technology Leaders
• Technical Level: Intermediate to Advanced

Analytical Progression

1. The Deployment Verification Challenge (5 minutes)

• Current limitations of traditional deployment verification
• The cost of deployment failures in modern distributed systems
• Why manual verification is no longer sustainable

1. Introduction to Continuous Verification (10 minutes)

• Defining Continuous Verification
• Core principles of ML-driven deployment verification
• Harness CV architecture and design philosophy

Key Technical Highlights:
- Machine learning techniques for time series analysis
- Symbolic Aggregate Approximation (SAA) for metric comparison
- Log clustering and anomaly detection algorithms

1. ML Techniques in Continuous Verification (5 minutes)

• Metric Analysis Techniques
• Comparing time series data using standard deviation
• Early trend detection before threshold breaches

• Automated performance deviation identification

• Log Analysis Strategies

• Error clustering mechanisms
• Automatic detection of:
  • New error types
  • Frequency changes in existing errors
  • Similar log pattern identification

1. Deployment Strategy Variations (3 minutes)
   - Canary Deployments
   - Blue-Green Deployments
   - Rolling Updates
   - Auto Deployments

2. Practical Implementation and Best Practices (2 minutes)
   - Configuring sensitivity levels
   - Metric and log feedback mechanisms
   - Integrating CV into existing CI/CD pipelines

Key Takeaways
- Practical strategies for implementing intelligent deployment checks
- How to reduce deployment risks and improve software reliability
- Understanding the transformative potential of ML in deployment verification

Technical Demonstration (10 minutes)
- Real-world CV scenario with metric and log analysis for Datadog as an example APM tool
- Automatic anomaly detection
- Rollback trigger mechanism