From 097b359b413d84fa7d8d46ed1a23c4227f0c170d Mon Sep 17 00:00:00 2001
From: David N Perkins <david.n.perkins@ibm.com>
Date: Tue, 8 Mar 2022 08:10:45 -0500
Subject: [PATCH] Added a failure counter to the HTTP service discovery
 (#10372)

* Added a failure counter to the http service discovery

Signed-off-by: David N Perkins <David.N.Perkins@ibm.com>
---
 discovery/http/http.go              | 14 ++++++++++++++
 discovery/http/http_test.go         | 30 +++++++++++++++++++++++++++++
 docs/configuration/configuration.md |  5 +++--
 docs/http_sd.md                     |  3 ++-
 4 files changed, 49 insertions(+), 3 deletions(-)

diff --git a/discovery/http/http.go b/discovery/http/http.go
index 6dd949e47..217abb192 100644
--- a/discovery/http/http.go
+++ b/discovery/http/http.go
@@ -28,6 +28,7 @@ import (
 	"github.com/go-kit/log"
 	"github.com/grafana/regexp"
 	"github.com/pkg/errors"
+	"github.com/prometheus/client_golang/prometheus"
 	"github.com/prometheus/common/config"
 	"github.com/prometheus/common/model"
 	"github.com/prometheus/common/version"
@@ -45,10 +46,17 @@ var (
 	}
 	userAgent        = fmt.Sprintf("Prometheus/%s", version.Version)
 	matchContentType = regexp.MustCompile(`^(?i:application\/json(;\s*charset=("utf-8"|utf-8))?)$`)
+
+	failuresCount = prometheus.NewCounter(
+		prometheus.CounterOpts{
+			Name: "prometheus_sd_http_failures_total",
+			Help: "Number of HTTP service discovery refresh failures.",
+		})
 )
 
 func init() {
 	discovery.RegisterConfig(&SDConfig{})
+	prometheus.MustRegister(failuresCount)
 }
 
 // SDConfig is the configuration for HTTP based discovery.
@@ -145,6 +153,7 @@ func (d *Discovery) refresh(ctx context.Context) ([]*targetgroup.Group, error) {
 
 	resp, err := d.client.Do(req.WithContext(ctx))
 	if err != nil {
+		failuresCount.Inc()
 		return nil, err
 	}
 	defer func() {
@@ -153,26 +162,31 @@ func (d *Discovery) refresh(ctx context.Context) ([]*targetgroup.Group, error) {
 	}()
 
 	if resp.StatusCode != http.StatusOK {
+		failuresCount.Inc()
 		return nil, errors.Errorf("server returned HTTP status %s", resp.Status)
 	}
 
 	if !matchContentType.MatchString(strings.TrimSpace(resp.Header.Get("Content-Type"))) {
+		failuresCount.Inc()
 		return nil, errors.Errorf("unsupported content type %q", resp.Header.Get("Content-Type"))
 	}
 
 	b, err := ioutil.ReadAll(resp.Body)
 	if err != nil {
+		failuresCount.Inc()
 		return nil, err
 	}
 
 	var targetGroups []*targetgroup.Group
 
 	if err := json.Unmarshal(b, &targetGroups); err != nil {
+		failuresCount.Inc()
 		return nil, err
 	}
 
 	for i, tg := range targetGroups {
 		if tg == nil {
+			failuresCount.Inc()
 			err = errors.New("nil target group item found")
 			return nil, err
 		}
diff --git a/discovery/http/http_test.go b/discovery/http/http_test.go
index 896eec1be..1ab034dfc 100644
--- a/discovery/http/http_test.go
+++ b/discovery/http/http_test.go
@@ -22,6 +22,8 @@ import (
 	"time"
 
 	"github.com/go-kit/log"
+	"github.com/prometheus/client_golang/prometheus"
+	dto "github.com/prometheus/client_model/go"
 	"github.com/prometheus/common/config"
 	"github.com/prometheus/common/model"
 	"github.com/stretchr/testify/require"
@@ -61,6 +63,7 @@ func TestHTTPValidRefresh(t *testing.T) {
 		},
 	}
 	require.Equal(t, tgs, expectedTargets)
+	require.Equal(t, 0.0, getFailureCount())
 }
 
 func TestHTTPInvalidCode(t *testing.T) {
@@ -82,6 +85,7 @@ func TestHTTPInvalidCode(t *testing.T) {
 	ctx := context.Background()
 	_, err = d.refresh(ctx)
 	require.EqualError(t, err, "server returned HTTP status 400 Bad Request")
+	require.Equal(t, 1.0, getFailureCount())
 }
 
 func TestHTTPInvalidFormat(t *testing.T) {
@@ -103,6 +107,32 @@ func TestHTTPInvalidFormat(t *testing.T) {
 	ctx := context.Background()
 	_, err = d.refresh(ctx)
 	require.EqualError(t, err, `unsupported content type "text/plain; charset=utf-8"`)
+	require.Equal(t, 1.0, getFailureCount())
+}
+
+var lastFailureCount float64
+
+func getFailureCount() float64 {
+	failureChan := make(chan prometheus.Metric)
+
+	go func() {
+		failuresCount.Collect(failureChan)
+		close(failureChan)
+	}()
+
+	var counter dto.Metric
+	for {
+		metric, ok := <-failureChan
+		if ok == false {
+			break
+		}
+		metric.Write(&counter)
+	}
+
+	// account for failures in prior tests
+	count := *counter.Counter.Value - lastFailureCount
+	lastFailureCount = *counter.Counter.Value
+	return count
 }
 
 func TestContentTypeRegex(t *testing.T) {
diff --git a/docs/configuration/configuration.md b/docs/configuration/configuration.md
index 4d730a6b3..2f8ec2751 100644
--- a/docs/configuration/configuration.md
+++ b/docs/configuration/configuration.md
@@ -1448,8 +1448,9 @@ Example response body:
 ]
 ```
 
-The endpoint is queried periodically at the specified
-refresh interval.
+The endpoint is queried periodically at the specified refresh interval.
+The `prometheus_sd_http_failures_total` counter metric tracks the number of 
+refresh failures. 
 
 Each target has a meta label `__meta_url` during the
 [relabeling phase](#relabel_config). Its value is set to the
diff --git a/docs/http_sd.md b/docs/http_sd.md
index ab88886cc..884deb9f3 100644
--- a/docs/http_sd.md
+++ b/docs/http_sd.md
@@ -40,7 +40,8 @@ an empty list `[]`. Target lists are unordered.
 
 Prometheus caches target lists. If an error occurs while fetching an updated
 targets list, Prometheus keeps using the current targets list. The targets list
-is not saved across restart.
+is not saved across restart. The `prometheus_sd_http_failures_total` counter 
+metric tracks the number of refresh failures.
 
 The whole list of targets must be returned on every scrape. There is no support
 for incremental updates. A Prometheus instance does not send its hostname and it