1 AI Ops - Anomaly Detection

Types of Anomalies in API Monitoring

1. Response Time Anomalies

  • When an API suddenly becomes slower than usual.
  • Example: The average response time is 200ms, but suddenly it spikes to 2 seconds.

2. Error Rate Anomalies

  • When the number of failed API calls increases unexpectedly.
  • Example: If an API usually has a 1% error rate but jumps to 20%, it’s an anomaly.

3. Traffic Volume Anomalies

  • Sudden increase or decrease in API request traffic.
  • Example: If an API normally gets 1,000 requests per minute, but suddenly gets 10,000, it might indicate a DDoS attack or bug.

4. Data Pattern Anomalies

  • Unexpected behavior in API responses.
  • Example: A user details API suddenly returns empty data for valid users.

Why Detect Anomalies?

Identify performance issues early before they affect users.

Prevent system failures by acting on unusual trends.

Enhance security by detecting suspicious activities.

Improve reliability of the distributed platform.

Machine Learning-Based Anomaly Detection

Isolation Forest (IForest)

  • Works by randomly splitting the dataset and identifying points that get isolated quickly.
  • Best for detecting sudden spikes or drops in API performance.