Two weeks of Harborline telemetry sit in the TSDB on obs-01, and this chapter is the language for interrogating them. Five topics build from the two vector shapes to the canonical histogram query to recording rules — and along the way, an offset 7d comparison hands Mara the book's first numeric proof that Saturday mornings really are different.
5 topics
Chapter 3 ended with data and no answers: node_exporter on every host, Redis and blackbox exporters scraping every 15 seconds, and a TSDB full of monotone counter ramps that mean nothing raw. Mara spends this chapter in the query editor turning that raw material into evidence — CPU busy-fraction on app-01, the keyspace hit ratio on cache-01, p99 latency of the probe against the checkout endpoint.
The payoff lands mid-chapter: probe latency on Saturday 09:00–11:00 runs roughly 3× Tuesday's, and the cache hit ratio sags from 0.97 to 0.71 in exactly the same window. The customer was right, and now there is a graph that says so. What the host data cannot say is why — which is the question Chapter 5's code-level instrumentation exists to answer.