# Self-Check: Mixed Revision — Logging, Metrics, Tracing & Alerting

Source: https://www.skillbyai.com/en/observability/ob-self-check-mixed

> Test your understanding across logs, metrics, tracing, SLOs and alerting, and know which earlier topic to revisit for any gap.

## How to use this self-check

Revision works best when you **retrieve** answers from memory before looking. For each prompt below, say your answer aloud or write it down, then check the topic named in brackets.

1. Name the three main signals and the question each answers best. *(Logs, Metrics and Traces)*
2. Why should a log message be constant with variable data in fields? *(Structured Logging)*
3. Estimate the series count of a counter with labels of 4, 10 and 20 values. *(Labels and Cardinality)*
4. Write a PromQL query for the 5xx ratio of one job. *(PromQL Essentials)*
5. What are the four parts of a `traceparent` header? *(Context Propagation)*
6. Which Collector processor should come first, and why? *(The OpenTelemetry Collector)*
7. With a 99.9% SLO, what burn rate empties a 30-day budget in about two days? *(Burn-Rate Alerting)*
8. What is the difference between inhibition and a silence? *(Routing, Grouping and Silencing Alerts)*

If you hesitate on more than two, reread those topics and try again tomorrow; spaced repetition beats one long session.

## Answers to check yourself

3: 4 x 10 x 20 = 800 series. 5: version, trace ID, parent span ID, trace flags. 6: `memory_limiter`, so overload is detected before other work. 7: about 14.4x. 8: inhibition mutes alerts automatically while another alert fires; a silence is a manual, time-limited mute.

## Practice nets before the match

A batter does not learn to play spin by reading about it; they face hundreds of balls in the nets. Answering questions from memory, getting some wrong and correcting them is the nets session for this course.

**Quiz:** Which combination gives the most reliable fleet-wide picture of checkout latency and error rate?

- [x] Counters and histograms aggregated with PromQL, because metrics are unsampled and cheap to aggregate
- [ ] Counting ERROR lines in sampled logs
- [ ] Averaging p99 values reported by each pod's summary
- [ ] Reading a few traces chosen at random

*Answer:* Counters and histograms aggregated with PromQL, because metrics are unsampled and cheap to aggregate. Metrics cover all traffic without sampling and aggregate correctly when you use counters with rate() and histogram buckets. Logs may be sampled, summaries cannot be averaged, and a handful of traces is anecdotal.
