Prometheus scrapes a plain-text page (/metrics) every few seconds. Each line is name{label="value"} number, preceded by # HELP and # TYPE lines.

counter: only goes up (restarts reset it). Name it ..._total; ask for rate(x[5m]), never the raw value.
gauge: goes up and down (queue depth, memory).
histogram: counts observations in cumulative le ("less than or equal") buckets, plus _sum and _count, so you can compute percentiles.

Every label value creates a new time series. Never use user ids or request ids as labels. Nothing here runs a monitoring server: you write the real files and the checks read them the way the tools would.

Metrics: the Prometheus text format