Project · /echo and /users — at 1k rps with documented p99

Hints

• Pick ONE language. Don't do this in four — you'll ship none.
• Use the framework's built-in JSON parsing. Don't write your own.
• The in-memory user store is a map<int, User> + a lock. A concurrent map is overkill for this project.
• wrk -c100 -t4 -d10s http://localhost:8080/users/42 is the standard benchmark. Save its output verbatim in your README.
• If p99 is more than 10x p50, you have a tail problem — note it honestly.

Expected output

$ wrk -c100 -t4 -d10s http://localhost:8080/users/42
Running 10s test @ http://localhost:8080/users/42
  4 threads and 100 connections
  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency    11.23ms    2.11ms   35.14ms   74.31%
    Req/Sec     2.24k     0.14k    2.61k    82.00%
  Requests/sec:   8954.02
  Transfer/sec:   1.23MB

Measured:
  p50:  9.8ms
  p95:  14.7ms
  p99:  22.1ms
  throughput: ~9k rps on an M2 laptop

Stretch goals

• Implement the same service in a second language. Compare the p99 numbers, honestly.
• Add a /users GET that lists all users. Put it behind pagination (?limit=10&cursor=...).
• Add rate limiting — token bucket at 100 req/sec per IP. Measure the effect on p99 under overload.
• Containerize it. Run it under --cpus=1 and re-measure. How much did you lose?