perf-profile

To find performance bottlenecks, you establish reproducible baselines, identify hot spots using profilers to generate flamegraphs, form testable hypotheses, and verify improvements by re-running benchmarks without introducing regressions.

Profiling Rust, Python, and JavaScript services requires targeted profiler recipes like cProfile, py-spy, and Node profiling to identify functions consuming significant resources and generate flamegraphs for allocation profiling.

Establishing a performance baseline involves running benchmarks to record p50/p99 latency, throughput, and memory characteristics, creating a measurable starting point to detect regressions and compare optimization results over time.

Yes, you can persist profiling results and benchmark data for cross-session comparison using optional SeleneDB persistence, enabling traceability of hypothesis-driven optimization outcomes across multiple analysis sessions.

Hypothesis-driven optimization ranks potential fixes by expected impact, tests changes one at a time, and records outcomes to systematically resolve performance issues while ensuring no new regressions are introduced.

You need to add instrumentation and tracing, such as spans and context fields, when requested to increase observability for performance issues, helping pinpoint slow operations and verify that optimizations resolve the bottlenecks.