Target uses HTTP transport with deadlines

Instead of externally handling timeouts when scraping a target, we set
timeouts on the HTTP connection. This ensures that we don't leak
goroutines on timeouts.

[fixes #181]

retrieval/deadline_client.go

@@ -0,0 +1,30 @@
+package retrieval
+
+import (
+	"net"
+	"net/http"
+	"time"
+)
+
+// NewDeadlineClient returns a new http.Client which will time out long running
+// requests.
+func NewDeadlineClient(timeout time.Duration) http.Client {
+	return http.Client{
+		Transport: &http.Transport{
+			// We need to disable keepalive, becasue we set a deadline on the
+			// underlying connection.
+			DisableKeepAlives: true,
+			Dial: func(netw, addr string) (c net.Conn, err error) {
+				start := time.Now()
+
+				c, err = net.DialTimeout(netw, addr, timeout)
+
+				if err == nil {
+					c.SetDeadline(start.Add(timeout))
+				}
+
+				return
+			},
+		},
+	}
+}

retrieval/target.go

@@ -117,6 +117,7 @@ type target struct {
 	Deadline time.Duration
 	// Any base labels that are added to this target and its metrics.
 	baseLabels model.LabelSet
+	client     http.Client
 }
 
 // Furnish a reasonably configured target for querying.
@@ -125,6 +126,7 @@ func NewTarget(address string, deadline time.Duration, baseLabels model.LabelSet
 		address:    address,
 		Deadline:   deadline,
 		baseLabels: baseLabels,
+		client:     NewDeadlineClient(deadline),
 	}
 
 	scheduler := &healthScheduler{
@@ -162,77 +164,55 @@ func (t *target) recordScrapeHealth(results chan format.Result, timestamp time.T
 
 func (t *target) Scrape(earliest time.Time, results chan format.Result) (err error) {
 	now := time.Now()
+	futureState := t.state
 
-	defer func() {
-		futureState := t.state
-
-		switch err {
-		case nil:
-			t.recordScrapeHealth(results, now, true)
-			futureState = ALIVE
-		default:
-			t.recordScrapeHealth(results, now, false)
-			futureState = UNREACHABLE
-		}
-
-		t.scheduler.Reschedule(earliest, futureState)
-		t.state = futureState
-	}()
-
-	done := make(chan bool)
-
-	go func(start time.Time) {
-		defer func() {
-			ms := float64(time.Since(start)) / float64(time.Millisecond)
-			labels := map[string]string{address: t.Address(), outcome: success}
-			if err != nil {
-				labels[outcome] = failure
-			}
-
-			targetOperationLatencies.Add(labels, ms)
-			targetOperations.Increment(labels)
-		}()
-
-		defer func() {
-			done <- true
-		}()
-
-		var resp *http.Response // Don't shadow "err" from the enclosing function.
-		resp, err = http.Get(t.Address())
-		if err != nil {
-			return
-		}
-
-		defer resp.Body.Close()
-
-		processor, err := format.DefaultRegistry.ProcessorForRequestHeader(resp.Header)
-		if err != nil {
-			return
-		}
-
-		// XXX: This is a wart; we need to handle this more gracefully down the
-		//      road, especially once we have service discovery support.
-		baseLabels := model.LabelSet{model.InstanceLabel: model.LabelValue(t.Address())}
-		for baseLabel, baseValue := range t.baseLabels {
-			baseLabels[baseLabel] = baseValue
-		}
-
-		err = processor.Process(resp.Body, now, baseLabels, results)
-		if err != nil {
-			return
-		}
-	}(time.Now())
-
-	select {
-	case <-done:
-		break
-	case <-time.After(t.Deadline):
-		err = fmt.Errorf("Target %s exceeded %s deadline.", t, t.Deadline)
+	if err = t.scrape(now, results); err != nil {
+		t.recordScrapeHealth(results, now, false)
+		futureState = UNREACHABLE
+	} else {
+		t.recordScrapeHealth(results, now, true)
+		futureState = ALIVE
 	}
 
+	t.scheduler.Reschedule(earliest, futureState)
+	t.state = futureState
+
 	return
 }
 
+func (t *target) scrape(timestamp time.Time, results chan format.Result) (err error) {
+	defer func(start time.Time) {
+		ms := float64(time.Since(start)) / float64(time.Millisecond)
+		labels := map[string]string{address: t.Address(), outcome: success}
+		if err != nil {
+			labels[outcome] = failure
+		}
+
+		targetOperationLatencies.Add(labels, ms)
+		targetOperations.Increment(labels)
+	}(time.Now())
+
+	resp, err := t.client.Get(t.Address())
+	if err != nil {
+		return
+	}
+	defer resp.Body.Close()
+
+	processor, err := format.DefaultRegistry.ProcessorForRequestHeader(resp.Header)
+	if err != nil {
+		return
+	}
+
+	// XXX: This is a wart; we need to handle this more gracefully down the
+	//      road, especially once we have service discovery support.
+	baseLabels := model.LabelSet{model.InstanceLabel: model.LabelValue(t.Address())}
+	for baseLabel, baseValue := range t.baseLabels {
+		baseLabels[baseLabel] = baseValue
+	}
+
+	return processor.Process(resp.Body, timestamp, baseLabels, results)
+}
+
 func (t target) State() TargetState {
 	return t.state
 }

retrieval/target_test.go

@@ -16,6 +16,8 @@ package retrieval
 import (
 	"github.com/prometheus/prometheus/model"
 	"github.com/prometheus/prometheus/retrieval/format"
+	"net/http"
+	"net/http/httptest"
 	"testing"
 	"time"
 )
@@ -63,3 +65,45 @@ func TestTargetRecordScrapeHealth(t *testing.T) {
 		t.Fatalf("Expected and actual samples not equal. Expected: %v, actual: %v", expected, actual)
 	}
 }
+
+func TestTargetScrapeTimeout(t *testing.T) {
+	signal := make(chan bool, 1)
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		<-signal
+		w.Header().Set("X-Prometheus-API-Version", "0.0.1")
+		w.Write([]byte(`[]`))
+	}))
+
+	defer server.Close()
+
+	testTarget := NewTarget(server.URL, 10*time.Millisecond, model.LabelSet{})
+	results := make(chan format.Result, 1024)
+
+	// scrape once without timeout
+	signal <- true
+	if err := testTarget.Scrape(time.Now(), results); err != nil {
+		t.Fatal(err)
+	}
+
+	// let the deadline lapse
+	time.Sleep(15*time.Millisecond)
+
+	// now scrape again
+	signal <- true
+	if err := testTarget.Scrape(time.Now(), results); err != nil {
+		t.Fatal(err)
+	}
+
+	// now timeout
+	if err := testTarget.Scrape(time.Now(), results); err == nil {
+		t.Fatal("expected scrape to timeout")
+	} else {
+		signal <- true // let handler continue
+	}
+
+	// now scrape again without timeout
+	signal <- true
+	if err := testTarget.Scrape(time.Now(), results); err != nil {
+		t.Fatal(err)
+	}
+}