performance-profiling_skill

Overview

This skill profiles CPU, memory, and I/O usage to identify bottlenecks, analyze execution traces, and diagnose performance issues. It provides a practical workflow for collecting profiles, analyzing flame graphs and memory snapshots, and validating optimizations. The focus is on measurable improvements in latency, resource usage, and scalability.

How this skill works

The skill guides you to establish a baseline, select appropriate profiling tools, and collect data under realistic load. It inspects CPU hot paths, memory allocation patterns, and blocking I/O (database, file, network) using tools like cProfile, py-spy, memory_profiler, and system utilities. Results are analyzed to prioritize changes, implement optimizations, and verify improvements with repeatable measurements.

When to use it

• Investigating slow endpoints or high latency in production-like environments
• Reducing CPU, memory, or I/O resource usage for cost and scalability
• Diagnosing regressions after a deployment or library upgrade
• Validating the impact of architectural or query optimizations
• Capacity planning and load-testing preparations

Best practices

• Profile with realistic data and load, ideally in staging or production-like setup
• Capture both CPU and memory profiles together to correlate causes
• Use flame graphs and call graphs to visualize hot paths and call frequency
• Prioritize fixes by impact and effort; focus on high-cost bottlenecks first
• Measure before and after each change and keep profiling artifacts for comparison
• Prefer statistical profilers in production to limit overhead

Example use cases

• Detect and fix N+1 database queries causing endpoint latency
• Find memory spikes and leaks by comparing snapshots across runs
• Identify slow third-party API calls and move them to async/background jobs
• Reduce CPU hotspots by rewriting hot functions or caching results
• Validate that eager loading or query joins reduce query count and latency