Observability
OTel: 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 unit stick.
Reconstruct the unit’s spine — the four-piece architecture, where vendor-neutrality lives, why Semantic Conventions are the highest-leverage discipline, head vs tail sampling, the agent-to-gateway pattern, and Collector operations — without looking back at the lessons.
- 01Name the four pieces of OTel and the single job of each.
- 02Where exactly does vendor-neutrality live, and what are the two ways it silently erodes?
- 03Why are Semantic Conventions the highest-leverage discipline in OTel, given that neither the SDK nor the Collector enforces them?
- 04Contrast head and tail sampling, and state the textbook production combination.
- 05Describe the agent-to-gateway deployment pattern and why tail sampling requires it.
- 06List the production operations disciplines that keep the Collector from failing silently, and the highest-frequency failure modes.
If you could reconstruct each answer from memory, you hold the unit’s spine: OTel is API + SDK + Collector + OTLP; neutrality lives at the application edge and erodes via vendor SDKs in code or proprietary Collector processors; Semantic Conventions are the governance lever that makes signals correlate; head sampling is cheap-but-blind and tail sampling is curated-but-stateful, combined in the ParentBased + tail pattern; the agent-to-gateway topology with trace_id-hashed routing makes tail sampling possible; and the Collector is critical-path infra that demands HA, persistent queue, and self-monitoring so its own failure is observable.