Gateway Gus saysMetrics come as counters, gauges, and histograms — each measuring differently.
Counters only go up (total requests, errors) — you rate them to get req/sec. Gauges go up and down (memory in use, active connections). Histograms bucket observations (request latency) so you can compute percentiles (p50, p95, p99) — far more useful than averages, which hide tail latency. Pick the right type for what you're measuring.
Power-ups you unlock
Counter: monotonic (totals) → rate it
Gauge: up/down (current value)
Histogram: distribution → percentiles
p95/p99 beat averages for latency
Timeout Titan attacks — common mistakes
Using a gauge for a monotonic total
Reporting average latency (hides the tail)
Too many high-cardinality label combinations
Boss battleChoose counter/gauge/histogram for: total errors, active users, request latency.