Summary

Expose operational metrics for monitoring via Prometheus-compatible /metrics endpoint.

Motivation

Production deployments need observability. Prometheus is the standard for infrastructure metrics.

Metrics to Expose

Counters

• distill_requests_total{endpoint, status} - Total requests by endpoint and status
• distill_tokens_processed_total{stage} - Total tokens processed by stage
• distill_cache_hits_total - Cache hits
• distill_cache_misses_total - Cache misses

Histograms

• distill_request_duration_seconds{endpoint} - Request latency
• distill_reduction_ratio{stage} - Reduction ratio distribution

Gauges

• distill_cache_size_bytes - Current cache size
• distill_active_requests - Currently processing requests

Implementation

import "github.com/prometheus/client_golang/prometheus"

var (
    requestsTotal = prometheus.NewCounterVec(
        prometheus.CounterOpts{
            Name: "distill_requests_total",
            Help: "Total requests by endpoint and status",
        },
        []string{"endpoint", "status"},
    )
    
    requestDuration = prometheus.NewHistogramVec(
        prometheus.HistogramOpts{
            Name:    "distill_request_duration_seconds",
            Help:    "Request latency distribution",
            Buckets: prometheus.DefBuckets,
        },
        []string{"endpoint"},
    )
)

Deliverables

• /metrics endpoint in Prometheus format
• Grafana dashboard template (JSON)
• Documentation for metrics

Acceptance Criteria

• All metrics exposed correctly
• Grafana dashboard works out of box
• No significant performance impact (< 1% overhead)