Observability
Three pillars: free-recall review
Retrieval beats re-reading. For each prompt, say or write a full answer from memory before you open the model answer — the effort of recall is what makes the cost models stick.
Reconstruct the unit’s spine — the three cost models, cardinality, the sampling tradeoff, join keys, the 2.0 shift, and the per-signal failure modes — without looking back at the lessons.
- 01Why are metrics, logs, and traces not interchangeable — what does each preserve, discard, and pay for?
- 02Explain why cardinality is the metrics cost cliff but high-cardinality fields are nearly free in logs.
- 03Contrast head-based and tail-based sampling: where is each cheap, and which traces does each risk losing?
- 04What are join keys, why does OpenTelemetry Semantic Conventions matter even on 1.0 backends, and how do exemplars fit?
- 05State the observability 2.0 thesis and name one workload where 1.0 still wins.
- 06List the characteristic failure mode of each signal and one guardrail to build before the incident.
If you could reconstruct each answer from memory, you hold the unit’s spine: three signals are three cost models (cardinality, ingestion bytes, sampling); cardinality is the metrics cliff while logs absorb high cardinality cheaply; head vs tail sampling trade collector cost against never losing error traces; join keys via OpenTelemetry Semantic Conventions plus exemplars make the three compose; observability 2.0 collapses them with wide events when the economics justify it; and each signal has a failure mode whose guardrail you build before the 03:00 page, not after.