Merge pull request #113 from oliver006/add_gc_metrics

Add garbage collection stats

Author: beorn7

prometheus/go_collector.go

@@ -2,10 +2,13 @@ package prometheus
 
 import (
 	"runtime"
+	"runtime/debug"
+	"time"
 )
 
 type goCollector struct {
 	goroutines Gauge
+	gcDesc     *Desc
 }
 
 // NewGoCollector returns a collector which exports metrics about the current
@@ -16,16 +19,32 @@ func NewGoCollector() *goCollector {
 			Name: "process_goroutines",
 			Help: "Number of goroutines that currently exist.",
 		}),
+		gcDesc: NewDesc(
+			"go_gc_duration_seconds",
+			"A summary of the GC invocation durations.",
+			nil, nil),
 	}
 }
 
 // Describe returns all descriptions of the collector.
 func (c *goCollector) Describe(ch chan<- *Desc) {
 	ch <- c.goroutines.Desc()
+	ch <- c.gcDesc
 }
 
 // Collect returns the current state of all metrics of the collector.
 func (c *goCollector) Collect(ch chan<- Metric) {
 	c.goroutines.Set(float64(runtime.NumGoroutine()))
 	ch <- c.goroutines
+
+	var stats debug.GCStats
+	stats.PauseQuantiles = make([]time.Duration, 5)
+	debug.ReadGCStats(&stats)
+
+	quantiles := make(map[float64]float64)
+	for idx, pq := range stats.PauseQuantiles[1:] {
+		quantiles[float64(idx+1)/float64(len(stats.PauseQuantiles)-1)] = pq.Seconds()
+	}
+	quantiles[0.0] = stats.PauseQuantiles[0].Seconds()
+	ch <- MustNewConstSummary(c.gcDesc, uint64(stats.NumGC), float64(stats.PauseTotal.Seconds()), quantiles)
 }

prometheus/go_collector_test.go

@@ -1,7 +1,7 @@
 package prometheus
 
 import (
-	"reflect"
+	"runtime"
 	"testing"
 	"time"
 
@@ -35,6 +35,9 @@ func TestGoCollector(t *testing.T) {
 			case Gauge:
 				pb := &dto.Metric{}
 				m.Write(pb)
+				if pb.GetGauge() == nil {
+					continue
+				}
 
 				if old == -1 {
 					old = int(pb.GetGauge().GetValue())
@@ -48,8 +51,66 @@ func TestGoCollector(t *testing.T) {
 				}
 
 				return
-			default:
-				t.Errorf("want type Gauge, got %s", reflect.TypeOf(metric))
+			}
+		case <-time.After(1 * time.Second):
+			t.Fatalf("expected collect timed out")
+		}
+	}
+}
+
+func TestGCCollector(t *testing.T) {
+	var (
+		c        = NewGoCollector()
+		ch       = make(chan Metric)
+		waitc    = make(chan struct{})
+		closec   = make(chan struct{})
+		oldGC    uint64
+		oldPause float64
+	)
+	defer close(closec)
+
+	go func() {
+		c.Collect(ch)
+		// force GC
+		runtime.GC()
+		<-waitc
+		c.Collect(ch)
+	}()
+
+	first := true
+	for {
+		select {
+		case metric := <-ch:
+			switch m := metric.(type) {
+			case *constSummary, *value:
+				pb := &dto.Metric{}
+				m.Write(pb)
+				if pb.GetSummary() == nil {
+					continue
+				}
+
+				if len(pb.GetSummary().Quantile) != 5 {
+					t.Errorf("expected 4 buckets, got %d", len(pb.GetSummary().Quantile))
+				}
+				for idx, want := range []float64{0.0, 0.25, 0.5, 0.75, 1.0} {
+					if *pb.GetSummary().Quantile[idx].Quantile != want {
+						t.Errorf("bucket #%d is off, got %f, want %f", idx, *pb.GetSummary().Quantile[idx].Quantile, want)
+					}
+				}
+				if first {
+					first = false
+					oldGC = *pb.GetSummary().SampleCount
+					oldPause = *pb.GetSummary().SampleSum
+					close(waitc)
+					continue
+				}
+				if diff := *pb.GetSummary().SampleCount - oldGC; diff != 1 {
+					t.Errorf("want 1 new garbage collection run, got %d", diff)
+				}
+				if diff := *pb.GetSummary().SampleSum - oldPause; diff <= 0 {
+					t.Errorf("want moar pause, got %f", diff)
+				}
+				return
 			}
 		case <-time.After(1 * time.Second):
 			t.Fatalf("expected collect timed out")