Metrics Basics (What to Measure)

Metrics Answer "How Is It Doing?"

Metrics provide time-series signals for alerting and capacity planning. Focus on a small set that maps to user impact.

Core Metric Types

• Counters: requests_total, errors_total
• Gauges: queue_depth, memory_bytes
• Histograms/Summaries: latency distributions

Minimal Metric Plan

# Conceptual metrics:
# requests_total{route, status}
# request_latency_ms{route} histogram
# errors_total{type}
# queue_depth gauge

Operational Checklist

• Define SLO-aligned metrics (latency, error rate).
• Use labels sparingly; avoid high-cardinality labels (user_id).
• Alert on symptoms (error rate, latency) not causes (CPU alone).

Failure Modes

• High cardinality: metrics backends melt under too many label values.
• Vanity metrics: numbers that do not reflect user impact.