Logger: alerts held back too long, retries that extend the wait, and stale tracked counts

Context

Welcome to the platform team. The rate limiter in logger/ implements both levels in README.md. Since the last release, three reports have come in.

Fix as much as you can in 25 minutes. The checks are graded one by one, so each report you fix counts even if the rest is not done.

LOG-11: Alerts wait eleven seconds, not ten

Report

disk full printed at 100. At 110 it was held back again; it only printed at 111.

Acceptance criteria

• A message printed at t prints again at exactly t + 10.

LOG-12: Retrying an alert keeps it silent

Report

A service retries db down every 5 seconds. It printed once at 0 and never again, though it should print again at 10.

Acceptance criteria

• Calls that returned False do not restart the wait. Only a print does.

LOG-13: tracked counts messages that have expired

Report

The dashboard polls tracked() while the stream is quiet. At 0 one alert printed; tracked(15) still says 1.

Acceptance criteria

• tracked(timestamp) counts only messages printed less than 10 seconds before timestamp, even when nothing else has been called since.

Notes

• Run pytest tests/visible. Two tests fail on the current code; not every report has a failing test yet.
• Practice task modelled on a candidate report of an Amazon logger rate limiter interview question. Original StackYard code, not the actual question; not affiliated with Amazon.