Metrics and expvar

Canonical guidance

• instrument where ownership is clear
• choose stable names and units
• use expvar when simple built-in publishing is enough

Use when

• basic counters and gauges
• debug endpoints
• small services that do not need heavier observability stacks

Avoid

• publishing metrics from arbitrary call sites with no owner
• unstable metric names
• assuming expvar replaces logs, traces, and profiles

Preferred pattern

• publish a small set of process and request metrics under one debug surface

Anti-pattern

• adding many unlabeled counters with vague names because metrics might be useful later

Explanation: This is tempting because counters are easy to add, but low-quality metrics create noise without helping operations.

Why

• observability helps only when metric meaning is owned and stable