correct EDNS responses (#96)

Tests updated as well and all the middleware. And Prometheus renamed to
metrics (directive is still prometheus).

middleware/metrics/README.md

@@ -0,0 +1,34 @@
+# prometheus
+
+This module enables prometheus metrics for CoreDNS.
+
+The following metrics are exported:
+
+* coredns_dns_request_count_total
+* coredns_dns_request_duration_seconds
+* coredns_dns_response_size_bytes
+* coredns_dns_response_rcode_count_total
+
+Each counter has a label `zone` which is the zonename used for the request/response,
+and a label `qtype` which old the query type.
+The `response_rcode_count_total` has an extra label `rcode` which holds the rcode
+of the response.
+
+If monitoring is enabled queries that do not enter the middleware chain are exported
+under the fake domain "dropped" (without a closing dot).
+
+Restarting CoreDNS will stop the monitoring. This is a bug. Also [this upstream
+Caddy bug](https://github.com/mholt/caddy/issues/675).
+
+## Syntax
+
+~~~
+prometheus
+~~~
+
+For each zone that you want to see metrics for.
+
+It optionally takes an address where the metrics are exported, the default
+is `localhost:9135`. The metrics path is fixed to `/metrics`.
+
+## Examples

middleware/metrics/handler.go

@@ -0,0 +1,42 @@
+package metrics
+
+import (
+	"time"
+
+	"golang.org/x/net/context"
+
+	"github.com/miekg/coredns/middleware"
+	"github.com/miekg/dns"
+)
+
+func (m Metrics) ServeDNS(ctx context.Context, w dns.ResponseWriter, r *dns.Msg) (int, error) {
+	state := middleware.State{W: w, Req: r}
+	qname := state.Name()
+	qtype := state.Type()
+	zone := middleware.Zones(m.ZoneNames).Matches(qname)
+	if zone == "" {
+		zone = "."
+	}
+
+	// Record response to get status code and size of the reply.
+	rw := middleware.NewResponseRecorder(w)
+	status, err := m.Next.ServeDNS(ctx, rw, r)
+
+	Report(zone, qtype, rw.Rcode(), rw.Size(), rw.Start())
+
+	return status, err
+}
+
+// Report is a plain reporting function that the server can use for REFUSED and other
+// queries that are turned down because they don't match any middleware.
+func Report(zone, qtype, rcode string, size int, start time.Time) {
+	if requestCount == nil {
+		// no metrics are enabled
+		return
+	}
+
+	requestCount.WithLabelValues(zone, qtype).Inc()
+	requestDuration.WithLabelValues(zone, qtype).Observe(float64(time.Since(start) / time.Second))
+	responseSize.WithLabelValues(zone, qtype).Observe(float64(size))
+	responseRcode.WithLabelValues(zone, rcode, qtype).Inc()
+}

middleware/metrics/metrics.go

@@ -0,0 +1,86 @@
+package metrics
+
+import (
+	"log"
+	"net/http"
+	"sync"
+
+	"github.com/miekg/coredns/middleware"
+	"github.com/prometheus/client_golang/prometheus"
+)
+
+const namespace = "coredns"
+
+var (
+	requestCount    *prometheus.CounterVec
+	requestDuration *prometheus.HistogramVec
+	responseSize    *prometheus.HistogramVec
+	responseRcode   *prometheus.CounterVec
+)
+
+const path = "/metrics"
+
+// Metrics holds the prometheus configuration. The metrics' path is fixed to be /metrics
+type Metrics struct {
+	Next      middleware.Handler
+	Addr      string // where to we listen
+	Once      sync.Once
+	ZoneNames []string
+}
+
+func (m *Metrics) Start() error {
+	m.Once.Do(func() {
+		define("")
+
+		prometheus.MustRegister(requestCount)
+		prometheus.MustRegister(requestDuration)
+		prometheus.MustRegister(responseSize)
+		prometheus.MustRegister(responseRcode)
+
+		http.Handle(path, prometheus.Handler())
+		go func() {
+			if err := http.ListenAndServe(m.Addr, nil); err != nil {
+				log.Printf("[ERROR] Failed to start prometheus handler: %s", err)
+			}
+		}()
+	})
+	return nil
+}
+
+func define(subsystem string) {
+	if subsystem == "" {
+		subsystem = "dns"
+	}
+	requestCount = prometheus.NewCounterVec(prometheus.CounterOpts{
+		Namespace: namespace,
+		Subsystem: subsystem,
+		Name:      "request_count_total",
+		Help:      "Counter of DNS requests made per zone and type and opcode.",
+	}, []string{"zone", "qtype"})
+
+	requestDuration = prometheus.NewHistogramVec(prometheus.HistogramOpts{
+		Namespace: namespace,
+		Subsystem: subsystem,
+		Name:      "request_duration_seconds",
+		Buckets:   append([]float64{.0001, .0005, .001, .0025}, prometheus.DefBuckets...),
+		Help:      "Histogram of the time (in seconds) each request took.",
+	}, []string{"zone", "qtype"})
+
+	responseSize = prometheus.NewHistogramVec(prometheus.HistogramOpts{
+		Namespace: namespace,
+		Subsystem: subsystem,
+		Name:      "response_size_bytes",
+		Help:      "Size of the returns response in bytes.",
+		Buckets:   []float64{0, 100, 200, 300, 400, 511, 1023, 2047, 4095, 8291, 16e3, 32e3, 48e3, 64e3},
+	}, []string{"zone", "qtype"})
+
+	responseRcode = prometheus.NewCounterVec(prometheus.CounterOpts{
+		Namespace: namespace,
+		Subsystem: subsystem,
+		Name:      "response_rcode_count_total",
+		Help:      "Counter of response status codes.",
+	}, []string{"zone", "rcode", "qtype"})
+}
+
+// Dropped indicates we dropped the query before any handling. It has no closing dot, so it can not be a valid zone.
+const Dropped = "dropped"