add zones label to cache metrics (#5124)

* add zones to cache metrics

Signed-off-by: Elijah Andrews <elijahcandrews@gmail.com>

plugin/cache/README.md

@@ -73,14 +73,14 @@ Entries with 0 TTL will remain in the cache until randomly evicted when the shar
 
 If monitoring is enabled (via the *prometheus* plugin) then the following metrics are exported:
 
-* `coredns_cache_entries{server, type}` - Total elements in the cache by cache type.
+* `coredns_cache_entries{server, type, zones}` - Total elements in the cache by cache type.
-* `coredns_cache_hits_total{server, type}` - Counter of cache hits by cache type.
+* `coredns_cache_hits_total{server, type, zones}` - Counter of cache hits by cache type.
-* `coredns_cache_misses_total{server}` - Counter of cache misses. - Deprecated, derive misses from cache hits/requests counters.
+* `coredns_cache_misses_total{server, zones}` - Counter of cache misses. - Deprecated, derive misses from cache hits/requests counters.
-* `coredns_cache_requests_total{server}` - Counter of cache requests.
+* `coredns_cache_requests_total{server, zones}` - Counter of cache requests.
-* `coredns_cache_prefetch_total{server}` - Counter of times the cache has prefetched a cached item.
+* `coredns_cache_prefetch_total{server, zones}` - Counter of times the cache has prefetched a cached item.
-* `coredns_cache_drops_total{server}` - Counter of responses excluded from the cache due to request/response question name mismatch.
+* `coredns_cache_drops_total{server, zones}` - Counter of responses excluded from the cache due to request/response question name mismatch.
-* `coredns_cache_served_stale_total{server}` - Counter of requests served from stale cache entries.
+* `coredns_cache_served_stale_total{server, zones}` - Counter of requests served from stale cache entries.
-* `coredns_cache_evictions_total{server, type}` - Counter of cache evictions.
+* `coredns_cache_evictions_total{server, type, zones}` - Counter of cache evictions.
 
 Cache types are either "denial" or "success". `Server` is the server handling the request, see the
 prometheus plugin for documentation.

plugin/cache/cache.go

@@ -21,6 +21,8 @@ type Cache struct {
 	Next  plugin.Handler
 	Zones []string
 
+	zonesMetricLabel string
+
 	ncache  *cache.Cache
 	ncap    int
 	nttl    time.Duration
@@ -162,11 +164,11 @@ func (w *ResponseWriter) WriteMsg(res *dns.Msg) error {
 	if hasKey && duration > 0 {
 		if w.state.Match(res) {
 			w.set(res, key, mt, duration)
-			cacheSize.WithLabelValues(w.server, Success).Set(float64(w.pcache.Len()))
+			cacheSize.WithLabelValues(w.server, Success, w.zonesMetricLabel).Set(float64(w.pcache.Len()))
-			cacheSize.WithLabelValues(w.server, Denial).Set(float64(w.ncache.Len()))
+			cacheSize.WithLabelValues(w.server, Denial, w.zonesMetricLabel).Set(float64(w.ncache.Len()))
 		} else {
 			// Don't log it, but increment counter
-			cacheDrops.WithLabelValues(w.server).Inc()
+			cacheDrops.WithLabelValues(w.server, w.zonesMetricLabel).Inc()
 		}
 	}
 
@@ -195,7 +197,7 @@ func (w *ResponseWriter) set(m *dns.Msg, key uint64, mt response.Type, duration
 	case response.NoError, response.Delegation:
 		i := newItem(m, w.now(), duration)
 		if w.pcache.Add(key, i) {
-			evictions.WithLabelValues(w.server, Success).Inc()
+			evictions.WithLabelValues(w.server, Success, w.zonesMetricLabel).Inc()
 		}
 		// when pre-fetching, remove the negative cache entry if it exists
 		if w.prefetch {
@@ -205,7 +207,7 @@ func (w *ResponseWriter) set(m *dns.Msg, key uint64, mt response.Type, duration
 	case response.NameError, response.NoData, response.ServerError:
 		i := newItem(m, w.now(), duration)
 		if w.ncache.Add(key, i) {
-			evictions.WithLabelValues(w.server, Denial).Inc()
+			evictions.WithLabelValues(w.server, Denial, w.zonesMetricLabel).Inc()
 		}
 
 	case response.OtherError:

plugin/cache/handler.go

@@ -43,7 +43,7 @@ func (c *Cache) ServeDNS(ctx context.Context, w dns.ResponseWriter, r *dns.Msg)
 		return c.doRefresh(ctx, state, crr)
 	}
 	if ttl < 0 {
-		servedStale.WithLabelValues(server).Inc()
+		servedStale.WithLabelValues(server, c.zonesMetricLabel).Inc()
 		// Adjust the time to get a 0 TTL in the reply built from a stale item.
 		now = now.Add(time.Duration(ttl) * time.Second)
 		cw := newPrefetchResponseWriter(server, state, c)
@@ -59,7 +59,7 @@ func (c *Cache) ServeDNS(ctx context.Context, w dns.ResponseWriter, r *dns.Msg)
 }
 
 func (c *Cache) doPrefetch(ctx context.Context, state request.Request, cw *ResponseWriter, i *item, now time.Time) {
-	cachePrefetches.WithLabelValues(cw.server).Inc()
+	cachePrefetches.WithLabelValues(cw.server, c.zonesMetricLabel).Inc()
 	c.doRefresh(ctx, state, cw)
 
 	// When prefetching we loose the item i, and with it the frequency
@@ -91,41 +91,41 @@ func (c *Cache) Name() string { return "cache" }
 
 func (c *Cache) get(now time.Time, state request.Request, server string) (*item, bool) {
 	k := hash(state.Name(), state.QType())
-	cacheRequests.WithLabelValues(server).Inc()
+	cacheRequests.WithLabelValues(server, c.zonesMetricLabel).Inc()
 
 	if i, ok := c.ncache.Get(k); ok && i.(*item).ttl(now) > 0 {
-		cacheHits.WithLabelValues(server, Denial).Inc()
+		cacheHits.WithLabelValues(server, Denial, c.zonesMetricLabel).Inc()
 		return i.(*item), true
 	}
 
 	if i, ok := c.pcache.Get(k); ok && i.(*item).ttl(now) > 0 {
-		cacheHits.WithLabelValues(server, Success).Inc()
+		cacheHits.WithLabelValues(server, Success, c.zonesMetricLabel).Inc()
 		return i.(*item), true
 	}
-	cacheMisses.WithLabelValues(server).Inc()
+	cacheMisses.WithLabelValues(server, c.zonesMetricLabel).Inc()
 	return nil, false
 }
 
 // getIgnoreTTL unconditionally returns an item if it exists in the cache.
 func (c *Cache) getIgnoreTTL(now time.Time, state request.Request, server string) *item {
 	k := hash(state.Name(), state.QType())
-	cacheRequests.WithLabelValues(server).Inc()
+	cacheRequests.WithLabelValues(server, c.zonesMetricLabel).Inc()
 
 	if i, ok := c.ncache.Get(k); ok {
 		ttl := i.(*item).ttl(now)
 		if ttl > 0 || (c.staleUpTo > 0 && -ttl < int(c.staleUpTo.Seconds())) {
-			cacheHits.WithLabelValues(server, Denial).Inc()
+			cacheHits.WithLabelValues(server, Denial, c.zonesMetricLabel).Inc()
 			return i.(*item)
 		}
 	}
 	if i, ok := c.pcache.Get(k); ok {
 		ttl := i.(*item).ttl(now)
 		if ttl > 0 || (c.staleUpTo > 0 && -ttl < int(c.staleUpTo.Seconds())) {
-			cacheHits.WithLabelValues(server, Success).Inc()
+			cacheHits.WithLabelValues(server, Success, c.zonesMetricLabel).Inc()
 			return i.(*item)
 		}
 	}
-	cacheMisses.WithLabelValues(server).Inc()
+	cacheMisses.WithLabelValues(server, c.zonesMetricLabel).Inc()
 	return nil
 }
 

plugin/cache/metrics.go

@@ -14,54 +14,54 @@ var (
 		Subsystem: "cache",
 		Name:      "entries",
 		Help:      "The number of elements in the cache.",
-	}, []string{"server", "type"})
+	}, []string{"server", "type", "zones"})
 	// cacheRequests is a counter of all requests through the cache.
 	cacheRequests = promauto.NewCounterVec(prometheus.CounterOpts{
 		Namespace: plugin.Namespace,
 		Subsystem: "cache",
 		Name:      "requests_total",
 		Help:      "The count of cache requests.",
-	}, []string{"server"})
+	}, []string{"server", "zones"})
 	// cacheHits is counter of cache hits by cache type.
 	cacheHits = promauto.NewCounterVec(prometheus.CounterOpts{
 		Namespace: plugin.Namespace,
 		Subsystem: "cache",
 		Name:      "hits_total",
 		Help:      "The count of cache hits.",
-	}, []string{"server", "type"})
+	}, []string{"server", "type", "zones"})
 	// cacheMisses is the counter of cache misses. - Deprecated
 	cacheMisses = promauto.NewCounterVec(prometheus.CounterOpts{
 		Namespace: plugin.Namespace,
 		Subsystem: "cache",
 		Name:      "misses_total",
 		Help:      "The count of cache misses. Deprecated, derive misses from cache hits/requests counters.",
-	}, []string{"server"})
+	}, []string{"server", "zones"})
 	// cachePrefetches is the number of time the cache has prefetched a cached item.
 	cachePrefetches = promauto.NewCounterVec(prometheus.CounterOpts{
 		Namespace: plugin.Namespace,
 		Subsystem: "cache",
 		Name:      "prefetch_total",
 		Help:      "The number of times the cache has prefetched a cached item.",
-	}, []string{"server"})
+	}, []string{"server", "zones"})
 	// cacheDrops is the number responses that are not cached, because the reply is malformed.
 	cacheDrops = promauto.NewCounterVec(prometheus.CounterOpts{
 		Namespace: plugin.Namespace,
 		Subsystem: "cache",
 		Name:      "drops_total",
 		Help:      "The number responses that are not cached, because the reply is malformed.",
-	}, []string{"server"})
+	}, []string{"server", "zones"})
 	// servedStale is the number of requests served from stale cache entries.
 	servedStale = promauto.NewCounterVec(prometheus.CounterOpts{
 		Namespace: plugin.Namespace,
 		Subsystem: "cache",
 		Name:      "served_stale_total",
 		Help:      "The number of requests served from stale cache entries.",
-	}, []string{"server"})
+	}, []string{"server", "zones"})
 	// evictions is the counter of cache evictions.
 	evictions = promauto.NewCounterVec(prometheus.CounterOpts{
 		Namespace: plugin.Namespace,
 		Subsystem: "cache",
 		Name:      "evictions_total",
 		Help:      "The count of cache evictions.",
-	}, []string{"server", "type"})
+	}, []string{"server", "type", "zones"})
 )

plugin/cache/setup.go

@@ -4,6 +4,7 @@ import (
 	"errors"
 	"fmt"
 	"strconv"
+	"strings"
 	"time"
 
 	"github.com/coredns/caddy"
@@ -185,6 +186,7 @@ func cacheParse(c *caddy.Controller) (*Cache, error) {
 		}
 
 		ca.Zones = origins
+		ca.zonesMetricLabel = strings.Join(origins, ",")
 		ca.pcache = cache.New(ca.pcap)
 		ca.ncache = cache.New(ca.ncap)
 	}