Prometheus

Pull-based scraping, the label data model and PromQL, exporters, and Alertmanager

Prometheus inverts the usual model: instead of applications pushing metrics out, Prometheus scrapes HTTP endpoints on a schedule. That single design choice drives most interview questions. Because scraping is pull-based, Prometheus needs service discovery to know what exists, and short-lived jobs need a Pushgateway because they may not survive until the next scrape. The label data model is the other half — every unique combination of label values is a separate time series, so high-cardinality labels are the classic way teams accidentally exhaust memory. Expect to be asked about the four metric types and when a histogram beats a summary, about rate() versus increase() and why rate() must wrap a counter, and about how recording rules trade storage for query speed. Alertmanager, and specifically grouping, inhibition and silencing, is where alert-fatigue questions land.