Module 07 Activity

Scenario

It is 2am and something is wrong. This activity decides now what you will wish you had instrumented.

What you build

Traces answering all seven incident questions, a metric taxonomy separating alerts from gates, redaction on write, and non-uniform sampling.

Steps

1. Write the seven questions you will ask during an incident and the field each requires. Include deploy and config version.
2. Emit a trace per request with per-span duration, success and enough detail to interpret it - including where in the sequence a failure landed.
3. Sort your metrics into alerts, daily reviews and release gates. Justify anything that is a count rather than a rate.
4. Redact structured identifiers on write and state plainly what redaction does not catch.
5. Set a non-uniform sampling policy: all failures, all gated actions, all over-budget runs, and a sample of successes.
6. Compute what fraction of traffic you keep and confirm no failure is sampled away.

Evidence to hand in

• The seven questions with their fields.
• One complete trace.
• The metric taxonomy with the count-versus-rate justification.
• A redacted log line and a statement of what is missed.
• The sampling policy with the retained fraction.

Review checklist

• Deploy and config version appear in every trace.
• Traces record where in the sequence a failure landed.
• Correctness is a gate, not an alert.
• Redaction runs on write and its limits are stated.
• No category of failure is sampled.

Mark complete