Memory Leak Detection
Automatically detects memory leak patterns before OOM failures, enabling planned responses instead of emergency restarts.
Overview
Memory Leak Detection continuously monitors pod memory usage patterns to identify gradual or accelerating memory increases that indicate potential memory leaks. By detecting these patterns before Out-of-Memory (OOM) failures occur, it gives you time to investigate and address the root cause proactively.
Wave uses statistical regression analysis with confidence scoring (R²) to identify anomalous growth patterns. By analyzing up to 5 days of historical memory metrics with minimum 6-hour data requirements, it provides reliable early detection of memory leaks in production workloads, preventing unexpected pod restarts and service disruptions.
Why Better Than Reactive OOM Monitoring
| Reactive OOM Alerts | Wave Memory Leak Detection | |
|---|---|---|
| Detection Method | Alert only after OOM kill events occur | Proactive pattern detection before OOM failures |
| Response Time | Emergency restarts and incident firefighting | Planned investigation and fix during business hours |
| Pattern Recognition | Cannot distinguish normal growth from leaks | Statistical regression analysis with confidence scoring (R² ≥ 0.7) for accurate leak detection |
Memory Leak Patterns Detected
Steady Growth Pattern
Detects continuous memory growth with minimal garbage collection (≤3 decreases) and ≥1% total increase. Common in applications with unbounded caches, connection pools, or event listener accumulation. Uses decrease count heuristic for reliable detection.
Trend-Based Growth Pattern
Identifies statistically significant memory increases using linear regression on hourly data: ≥15% hourly increase rate OR ≥50% total increase with confidence score (R²) ≥0.7. Catches both gradual and accelerating leaks through regression slope analysis.
Key Outcomes & Benefits
Catch memory leaks before they cause production outages
Detect leaks with minimum 6 hours of data, up to 5 days historical analysis
Configurable high utilization threshold alerts (default: 85% memory usage)
How It Works
1. Metrics Collection & Aggregation
WA Metrics Agent collects memory usage for all containers. System aggregates data into hourly buckets for statistical analysis (up to 5 days history, minimum 6 hours required).
2. Statistical Analysis & Pattern Detection
Regression analysis calculates memory trend slope and R² confidence. Detects two patterns: Steady Growth (decrease count ≤3, increase ≥1%) and Trend-Based Growth (hourly rate ≥15% OR total ≥50% with R²≥0.7). Also monitors OOM risk when utilization exceeds configurable threshold (default 85%).
3. Leak Classification & Logging
System classifies detected patterns and stores logs with baseline/recent memory, increase percentages, confidence scores, and hourly trend data. Tracks first and latest detection timestamps for each workload container.
4. Visualization & Proactive Alerts
UI displays memory leak detection results with pattern badges, memory trend charts, and OOM risk indicators. Teams receive advance notice to investigate and fix leaks during planned maintenance windows.
Prevent OOM Failures with Statistical Leak Detection
See how regression-based Memory Leak Detection protects your workloads.
Book a Demo