feat(metrics): initial errcode metric

Adds an errcode metric to track the number of errcodes generated by
OSM.

Signed-off-by: jaellio <[email protected]>

cmd/osm-controller/osm-controller.go

@@ -299,6 +299,7 @@ func startMetricsStore() {
 		metricsstore.DefaultMetricsStore.ProxyBroadcastEventCount,
 		metricsstore.DefaultMetricsStore.CertIssuedCount,
 		metricsstore.DefaultMetricsStore.CertIssuedTime,
+		metricsstore.DefaultMetricsStore.ErrCodeCounter,
 	)
 }
 

pkg/catalog/egress.go

@@ -85,7 +85,7 @@ func (mc *MeshCatalog) GetEgressTrafficPolicy(serviceIdentity identity.ServiceId
 	// Deduplicate the list of TrafficMatch objects
 	trafficMatches, err = trafficpolicy.DeduplicateTrafficMatches(trafficMatches)
 	if err != nil {
-		log.Error().Err(err).Str(errcode.Kind, errcode.ErrDedupEgressTrafficMatches.String()).
+		log.Error().Err(err).Str(errcode.Kind, errcode.GetErrCodeWithMetric(errcode.ErrDedupEgressTrafficMatches)).
 			Msgf("Error deduplicating egress traffic matches for service identity %s", serviceIdentity)
 		return nil, err
 	}

pkg/errcode/errcode.go

@@ -8,6 +8,8 @@ import (
 	"strings"
 
 	"github.com/pkg/errors"
+
+	"github.com/openservicemesh/osm/pkg/metricsstore"
 )
 
 // ErrCode defines the type to represent error codes
@@ -373,6 +375,13 @@ func (e ErrCode) String() string {
 	return fmt.Sprintf("E%d", e)
 }
 
+// GetErrCodeWithMetric increments the ErrCodeCounter metric for the given error code
+// Returns the error code as a string
+func GetErrCodeWithMetric(e ErrCode) string {
+	metricsstore.DefaultMetricsStore.ErrCodeCounter.WithLabelValues(e.String()).Inc()
+	return e.String()
+}
+
 // FromStr returns the ErrCode representation for the given error code string
 // Ex. E1000 is converted to ErrInvalidCLIArgument
 func FromStr(e string) (ErrCode, error) {

pkg/errcode/errcode_test.go

@@ -11,6 +11,11 @@ func TestString(t *testing.T) {
 	assert.Equal(ErrInvalidCLIArgument.String(), "E1000")
 }
 
+func TestGetErrorCodeWithMetric(t *testing.T) {
+	assert := tassert.New(t)
+	assert.Equal(GetErrCodeWithMetric(ErrInvalidCLIArgument), "E1000")
+}
+
 func TestFromStr(t *testing.T) {
 	testCases := []struct {
 		name            string

pkg/metricsstore/metricsstore.go

@@ -55,6 +55,12 @@ type MetricsStore struct {
 	// CertXdsIssuedCounter the histogram to track the time to issue a certificates
 	CertIssuedTime *prometheus.HistogramVec
 
+	/*
+	 * ErrCode metrics
+	 */
+	// ErrCodeCounter is the metric counter for the number of errcodes generated by OSM
+	ErrCodeCounter *prometheus.CounterVec
+
 	/*
 	 * MetricsStore internals should be defined below --------------
 	 */
@@ -158,6 +164,19 @@ func init() {
 			Help:      "Histogram to track time spent to issue xds certificate",
 		},
 		[]string{})
+
+	/*
+	 * ErrCode metrics
+	 */
+	defaultMetricsStore.ErrCodeCounter = prometheus.NewCounterVec(
+		prometheus.CounterOpts{
+			Namespace: metricsRootNamespace,
+			Subsystem: "error",
+			Name:      "err_code_count",
+			Help:      "Respesents the number of errcodes generated by OSM",
+		},
+		[]string{"err_code"},
+	)
 	defaultMetricsStore.registry = prometheus.NewRegistry()
 }
 

pkg/metricsstore/metricsstore_test.go

@@ -13,13 +13,15 @@ func setup() {
 	DefaultMetricsStore.Start(
 		DefaultMetricsStore.K8sAPIEventCounter,
 		DefaultMetricsStore.ProxyConnectCount,
+		DefaultMetricsStore.ErrCodeCounter,
 	)
 }
 
 func teardown() {
 	DefaultMetricsStore.Stop(
 		DefaultMetricsStore.K8sAPIEventCounter,
 		DefaultMetricsStore.ProxyConnectCount,
+		DefaultMetricsStore.ErrCodeCounter,
 	)
 }
 
@@ -82,4 +84,30 @@ osm_proxy_connect_count %d
 `, proxiesConnected-proxiesDisconnected)
 		assert.Contains(rr.Body.String(), expectedResp)
 	})
+
+	t.Run("ErrcodeCounter", func(t *testing.T) {
+		assert := tassert.New(t)
+
+		errcodeCount := 3
+
+		for i := 1; i <= errcodeCount; i++ {
+			DefaultMetricsStore.ErrCodeCounter.WithLabelValues("E1100").Inc()
+
+			handler := DefaultMetricsStore.Handler()
+
+			req, err := http.NewRequest("GET", "/metrics", nil)
+			assert.Nil(err)
+
+			rr := httptest.NewRecorder()
+			handler.ServeHTTP(rr, req)
+
+			assert.Equal(http.StatusOK, rr.Code)
+
+			expectedResp := fmt.Sprintf(`# HELP osm_error_err_code_count Respesents the number of errcodes generated by OSM
+# TYPE osm_error_err_code_count counter
+osm_error_err_code_count{err_code="E1100"} %d
+`, i /* api event count */)
+			assert.Contains(rr.Body.String(), expectedResp)
+		}
+	})
 }