Counter Metrics

The CEO walks over and asks: "How many orders did we process today?" If your answer involves SSHing into a server and running grep "order.completed" app.log | wc -l, you're doing it wrong.

A counter is the simplest and most useful metric type. It starts at zero and only goes up. Every time an event happens, a request arrives, an error occurs, a user signs up, you increment the counter. At any point, you can look at the counter's value and know exactly how many times that event has happened since the process started.

More importantly, you can calculate the rate, how fast is the counter increasing? "We processed 12,400 orders today" is useful. "We're processing 300 orders per minute, up from 200 per minute last Tuesday" is actionable. Counters give you both.

A counter has three rules:

1. It starts at zero when the process starts
2. It can only increase (or reset to zero on restart)
3. You derive rates by calculating the difference between two samples

In Prometheus, you'd use rate(http_requests_total[5m]) to get the per-second request rate averaged over the last 5 minutes. The rate() function handles counter resets (when a process restarts and the counter goes back to zero) automatically.

Common counters in production:

• http_requests_total, total requests, labeled by method, path, status code
• http_errors_total, total error responses
• orders_completed_total, business event count
• bytes_transferred_total, network throughput
• db_queries_total, database call count

The key insight: you almost never look at the raw counter value. You always look at the rate. "We have served 47 million requests" is trivia. "We're serving 3,200 requests per second" is operational data.