prometheus-go-code-review

# Prometheus Go Code Review

## Review Checklist

- [ ] Metric types match measurement semantics (Counter/Gauge/Histogram)
- [ ] Labels have low cardinality (no user IDs, timestamps, paths)
- [ ] Metric names follow conventions (snake_case, unit suffix)
- [ ] Histograms use appropriate bucket boundaries
- [ ] Metrics registered once, not per-request
- [ ] Collectors don't panic on race conditions
- [ ] /metrics endpoint exposed and accessible

## Metric Type Selection

| Measurement | Type | Example |
|-------------|------|---------|
| Requests processed | Counter | `requests_total` |
| Items in queue | Gauge | `queue_length` |
| Request duration | Histogram | `request_duration_seconds` |
| Concurrent connections | Gauge | `active_connections` |
| Errors since start | Counter | `errors_total` |
| Memory usage | Gauge | `memory_bytes` |

## Critical Anti-Patterns

### 1. High Cardinality Labels

```go
// BAD - unique per user/request
counter := promauto.NewCounterVec(
    prometheus.CounterOpts{Name: "requests_total"},
    []string{"user_id", "path"},  // millions of series!
)
counter.WithLabelValues(userID, request.URL.Path).Inc()

// GOOD - bounded label values
counter := promauto.NewCounterVec(
    prometheus.CounterOpts{Name: "requests_total"},
    []string{"method", "status_code"},  // <100 series
)
counter.WithLabelValues(r.Method, statusCode).Inc()
```

### 2. Wrong Metric Type

```go
// BAD - using gauge for monotonic value
requestCount := promauto.NewGauge(prometheus.GaugeOpts{
    Name: "http_requests",
})
requestCount.Inc()  // should be Counter!

// GOOD
requestCount := promauto.NewCounter(prometheus.CounterOpts{
    Name: "http_requests_total",
})
requestCount.Inc()
```

### 3. Registering Per-Request

```go
// BAD - new metric per request
func handler(w http.ResponseWriter, r *http.Request) {
    counter := prometheus.NewCounter(...)  // creates new each time!
    prometheus.MustRegister(counter)       // panics on duplicate!
}

// GOOD - register once