Skip to content

Level 8 / Project 06 - Response Time Profiler

Home: README

Learn Your Way

Read Build Watch Test Review Visualize Try
— This project — — Flashcards — —

Focus

  • Context managers and decorators for timing code blocks
  • Percentile calculation (p50, p90, p95, p99) with linear interpolation
  • Statistical profiling: mean, median, standard deviation
  • Bottleneck identification across multiple profiled functions
  • time.perf_counter() for high-resolution measurements

Why this project exists

Understanding where time is spent is essential for optimization. A function that runs in 2ms on average but 200ms at p99 causes intermittent user-visible slowdowns that are invisible to mean-based monitoring. This project builds a profiling toolkit that measures function execution times, computes percentile distributions, and identifies bottlenecks — the same approach used by APM tools like New Relic, Datadog, and Jaeger.

Run (copy/paste)

cd <repo-root>/projects/level-8/06-response-time-profiler
python project.py --demo
pytest -q

Expected terminal output

{
  "profiles": [
    {"function": "fast_operation", "calls": 20, "mean_ms": ..., "p99_ms": ...},
    ...
  ],
  "bottleneck": {"function": "slow_operation", ...}
}
7 passed

Expected artifacts

  • Console JSON output with profiling reports
  • Passing tests
  • Updated notes.md

Alter it (required)

  1. Add a reset() method to ResponseTimeProfiler that clears all recorded timings.
  2. Change report() to include a slowest_operations list (top 3 by p99).
  3. Add a @profile class method that works as a decorator without needing an instance reference.

Break it (required)

  1. Call report() with no recorded timings — does percentile() handle an empty list?
  2. Record a negative duration (simulate a clock skew) — what happens to the percentile calculations?
  3. Use the context manager but raise an exception inside it — does the timing still get recorded?

Fix it (required)

  1. Add a guard in percentile() that returns 0.0 for empty data.
  2. Validate that recorded durations are non-negative.
  3. Add a test that the context manager records timing even when the block raises.

Explain it (teach-back)

  1. How does time.perf_counter() differ from time.time() and why is it better for profiling?
  2. What is the p95 vs p99 percentile and why do production systems track both?
  3. How does the @profile_function decorator preserve the original function's name?
  4. Why is a context manager useful for profiling blocks of code that aren't single functions?

Mastery check

You can move on when you can: - explain the difference between mean, median, p95, and p99 response times, - profile a real function and interpret the report output, - describe when p99 matters more than average (tail latency), - add a new profiling metric (e.g. throughput) without modifying existing code.

← Prev Home Next →