Add Errors Reporting Prometheus Metric (#345)

Author: maria-reynoso

pkg/controller/sync.go

@@ -93,6 +93,8 @@ func (c *Controller) checkContainer(ctx context.Context, log *logrus.Entry,
 
 	result, err := c.checker.Container(ctx, log, pod, container, opts)
 	if err != nil {
+		// Report the error using ErrorsReporting
+		c.metrics.ErrorsReporting(pod.Namespace, pod.Name, container.Name, container.Image)
 		return err
 	}
 

pkg/metrics/metrics.go

@@ -25,6 +25,7 @@ type Metrics struct {
 	registry               *prometheus.Registry
 	containerImageVersion  *prometheus.GaugeVec
 	containerImageDuration *prometheus.GaugeVec
+	containerImageErrors   *prometheus.CounterVec
 
 	// Contains all metrics for the roundtripper
 	roundTripper *RoundTripper
@@ -62,12 +63,23 @@ func NewServer(log *logrus.Entry) *Metrics {
 		},
 		[]string{"namespace", "pod", "container", "image"},
 	)
+	containerImageErrors := promauto.With(reg).NewCounterVec(
+		prometheus.CounterOpts{
+			Namespace: "version_checker",
+			Name:      "image_failures_total",
+			Help:      "Total number of errors where the version-checker was unable to get the latest upstream registry version",
+		},
+		[]string{
+			"namespace", "pod", "container", "image",
+		},
+	)
 
 	return &Metrics{
 		log:                    log.WithField("module", "metrics"),
 		registry:               reg,
 		containerImageVersion:  containerImageVersion,
 		containerImageDuration: containerImageDuration,
+		containerImageErrors:   containerImageErrors,
 		containerCache:         make(map[string]cacheItem),
 		roundTripper:           NewRoundTripper(reg),
 	}
@@ -160,6 +172,11 @@ func (m *Metrics) latestImageIndex(namespace, pod, container, containerType stri
 	return strings.Join([]string{namespace, pod, container, containerType}, "")
 }
 
+func (m *Metrics) ErrorsReporting(namespace, pod, container, imageURL string) {
+
+	m.containerImageErrors.WithLabelValues(namespace, pod, container, imageURL).Inc()
+}
+
 func (m *Metrics) buildLabels(namespace, pod, container, containerType, imageURL, currentVersion, latestVersion string) prometheus.Labels {
 	return prometheus.Labels{
 		"namespace":       namespace,

pkg/metrics/metrics_test.go

@@ -4,8 +4,10 @@ import (
 	"fmt"
 	"testing"
 
+	"github.com/prometheus/client_golang/prometheus"
 	"github.com/prometheus/client_golang/prometheus/testutil"
 	"github.com/sirupsen/logrus"
+	"github.com/stretchr/testify/assert"
 )
 
 func TestCache(t *testing.T) {
@@ -37,3 +39,43 @@ func TestCache(t *testing.T) {
 		}
 	}
 }
+
+// TestErrorsReporting verifies that the error metric increments correctly
+func TestErrorsReporting(t *testing.T) {
+	m := NewServer(logrus.NewEntry(logrus.New()))
+
+	// Reset the metrics before testing
+	m.containerImageErrors.Reset()
+
+	testCases := []struct {
+		namespace string
+		pod       string
+		container string
+		image     string
+		expected  int
+	}{
+		{"namespace", "pod", "container", "url", 1},
+		{"namespace", "pod", "container", "url", 2},
+		{"namespace2", "pod2", "container2", "url2", 1},
+	}
+
+	for i, tc := range testCases {
+		t.Run(fmt.Sprintf("Case %d", i+1), func(t *testing.T) {
+			// Report an error
+			m.ErrorsReporting(tc.namespace, tc.pod, tc.container, tc.image)
+
+			// Retrieve metric
+			metric, err := m.containerImageErrors.GetMetricWith(prometheus.Labels{
+				"namespace": tc.namespace,
+				"pod":       tc.pod,
+				"container": tc.container,
+				"image":     tc.image,
+			})
+			assert.NoError(t, err, "Failed to get metric with labels")
+
+			// Validate metric count
+			fetchErrorCount := testutil.ToFloat64(metric)
+			assert.Equal(t, float64(tc.expected), fetchErrorCount, "Expected error count to increment correctly")
+		})
+	}
+}