refactor: separate metrics code for reuse (#13514)

Signed-off-by: Alan Clucas <alan@clucas.org>

util/telemetry/attributes.go

@@ -0,0 +1,43 @@
+package telemetry
+
+const (
+	AttribBuildVersion      string = `version`
+	AttribBuildPlatform     string = `platform`
+	AttribBuildGoVersion    string = `go_version`
+	AttribBuildDate         string = `build_date`
+	AttribBuildCompiler     string = `compiler`
+	AttribBuildGitCommit    string = `git_commit`
+	AttribBuildGitTreeState string = `git_treestate`
+	AttribBuildGitTag       string = `git_tag`
+
+	AttribCronWFName string = `name`
+
+	AttribErrorCause string = "cause"
+
+	AttribLogLevel string = `level`
+
+	AttribNodePhase string = `node_phase`
+
+	AttribPodPhase         string = `phase`
+	AttribPodNamespace     string = `namespace`
+	AttribPodPendingReason string = `reason`
+
+	AttribQueueName string = `queue_name`
+
+	AttribRecentlyStarted string = `recently_started`
+
+	AttribRequestKind = `kind`
+	AttribRequestVerb = `verb`
+	AttribRequestCode = `status_code`
+
+	AttribTemplateName      string = `name`
+	AttribTemplateNamespace string = `namespace`
+	AttribTemplateCluster   string = `cluster_scope`
+
+	AttribWorkerType string = `worker_type`
+
+	AttribWorkflowNamespace string = `namespace`
+	AttribWorkflowPhase     string = `phase`
+	AttribWorkflowStatus           = `status`
+	AttribWorkflowType             = `type`
+)

workflow/metrics/exporter_prometheus.go → util/telemetry/exporter_prometheus.go

@@ -1,4 +1,4 @@
-package metrics
+package telemetry
 
 import (
 	"context"
@@ -20,8 +20,8 @@ import (
 )
 
 const (
-	defaultPrometheusServerPort = 9090
-	defaultPrometheusServerPath = "/metrics"
+	DefaultPrometheusServerPort = 9090
+	DefaultPrometheusServerPath = "/metrics"
 )
 
 func (config *Config) prometheusMetricsExporter(namespace string) (*prometheus.Exporter, error) {
@@ -39,14 +39,14 @@ func (config *Config) prometheusMetricsExporter(namespace string) (*prometheus.E
 
 func (config *Config) path() string {
 	if config.Path == "" {
-		return defaultPrometheusServerPath
+		return DefaultPrometheusServerPath
 	}
 	return config.Path
 }
 
 func (config *Config) port() int {
 	if config.Port == 0 {
-		return defaultPrometheusServerPort
+		return DefaultPrometheusServerPort
 	}
 	return config.Port
 }

workflow/metrics/exporter_prometheus_test.go → util/telemetry/exporter_prometheus_test.go

@@ -1,6 +1,6 @@
 //go:build !windows
 
-package metrics
+package telemetry
 
 import (
 	"context"
@@ -14,19 +14,22 @@ import (
 	"github.com/stretchr/testify/require"
 )
 
+// testScopeName is the name that the metrics running under test will have
+const testScopeName string = "argo-workflows-test"
+
 func TestDisablePrometheusServer(t *testing.T) {
 	config := Config{
 		Enabled: false,
-		Path:    defaultPrometheusServerPath,
-		Port:    defaultPrometheusServerPort,
+		Path:    DefaultPrometheusServerPath,
+		Port:    DefaultPrometheusServerPort,
 	}
 	ctx, cancel := context.WithCancel(context.Background())
 	defer cancel()
-	m, err := New(ctx, TestScopeName, &config, Callbacks{})
+	m, err := NewMetrics(ctx, testScopeName, testScopeName, &config)
 	require.NoError(t, err)
 	go m.RunPrometheusServer(ctx, false)
 	time.Sleep(1 * time.Second) // to confirm that the server doesn't start, even if we wait
-	resp, err := http.Get(fmt.Sprintf("http://localhost:%d%s", defaultPrometheusServerPort, defaultPrometheusServerPath))
+	resp, err := http.Get(fmt.Sprintf("http://localhost:%d%s", DefaultPrometheusServerPort, DefaultPrometheusServerPath))
 	if resp != nil {
 		defer resp.Body.Close()
 	}
@@ -37,16 +40,16 @@ func TestDisablePrometheusServer(t *testing.T) {
 func TestPrometheusServer(t *testing.T) {
 	config := Config{
 		Enabled: true,
-		Path:    defaultPrometheusServerPath,
-		Port:    defaultPrometheusServerPort,
+		Path:    DefaultPrometheusServerPath,
+		Port:    DefaultPrometheusServerPort,
 	}
 	ctx, cancel := context.WithCancel(context.Background())
 	defer cancel()
-	m, err := New(ctx, TestScopeName, &config, Callbacks{})
+	m, err := NewMetrics(ctx, testScopeName, testScopeName, &config)
 	require.NoError(t, err)
 	go m.RunPrometheusServer(ctx, false)
 	time.Sleep(1 * time.Second)
-	resp, err := http.Get(fmt.Sprintf("http://localhost:%d%s", defaultPrometheusServerPort, defaultPrometheusServerPath))
+	resp, err := http.Get(fmt.Sprintf("http://localhost:%d%s", DefaultPrometheusServerPort, DefaultPrometheusServerPath))
 	require.NoError(t, err)
 	assert.Equal(t, http.StatusOK, resp.StatusCode)
 
@@ -57,22 +60,25 @@ func TestPrometheusServer(t *testing.T) {
 
 	bodyString := string(bodyBytes)
 	assert.NotEmpty(t, bodyString)
+
+	cancel()                    // Explicit cancel as sometimes in github CI port 9090 is still busy
+	time.Sleep(1 * time.Second) // Wait for prometheus server
 }
 
 func TestDummyPrometheusServer(t *testing.T) {
 	config := Config{
 		Enabled: true,
-		Path:    defaultPrometheusServerPath,
-		Port:    defaultPrometheusServerPort,
+		Path:    DefaultPrometheusServerPath,
+		Port:    DefaultPrometheusServerPort,
 		Secure:  false,
 	}
 	ctx, cancel := context.WithCancel(context.Background())
 	defer cancel()
-	m, err := New(ctx, TestScopeName, &config, Callbacks{})
+	m, err := NewMetrics(ctx, testScopeName, testScopeName, &config)
 	require.NoError(t, err)
 	go m.RunPrometheusServer(ctx, true)
 	time.Sleep(1 * time.Second)
-	resp, err := http.Get(fmt.Sprintf("http://localhost:%d%s", defaultPrometheusServerPort, defaultPrometheusServerPath))
+	resp, err := http.Get(fmt.Sprintf("http://localhost:%d%s", DefaultPrometheusServerPort, DefaultPrometheusServerPath))
 	require.NoError(t, err)
 	assert.Equal(t, http.StatusOK, resp.StatusCode)
 
@@ -84,4 +90,7 @@ func TestDummyPrometheusServer(t *testing.T) {
 	bodyString := string(bodyBytes)
 
 	assert.Empty(t, bodyString) // expect the dummy metrics server to provide no metrics responses
+
+	cancel()                    // Explicit cancel as sometimes in github CI port 9090 is still busy
+	time.Sleep(1 * time.Second) // Wait for prometheus server
 }

util/telemetry/helpers_test.go

@@ -0,0 +1,65 @@
+package telemetry
+
+import (
+	"context"
+
+	"go.opentelemetry.io/otel/sdk/metric"
+)
+
+func createDefaultTestMetrics() (*Metrics, *TestMetricsExporter, error) {
+	config := Config{
+		Enabled: true,
+	}
+	return createTestMetrics(&config)
+}
+
+func createTestMetrics(config *Config) (*Metrics, *TestMetricsExporter, error) {
+	ctx /* with cancel*/ := context.Background()
+	te := NewTestMetricsExporter()
+
+	m, err := NewMetrics(ctx, TestScopeName, TestScopeName, config, metric.WithReader(te))
+	if err != nil {
+		return nil, nil, err
+	}
+	err = m.Populate(ctx, AddVersion, addTestingCounter, addTestingHistogram)
+	return m, te, err
+}
+
+const (
+	nameTestingHistogram = `testing_histogram`
+	nameTestingCounter   = `testing_counter`
+	errorCauseTestingA   = "TestingA"
+	errorCauseTestingB   = "TestingB"
+)
+
+func addTestingHistogram(_ context.Context, m *Metrics) error {
+	// The buckets here are only the 'defaults' and can be overridden with configmap defaults
+	return m.CreateInstrument(Float64Histogram,
+		nameTestingHistogram,
+		"Testing Metric",
+		"s",
+		WithDefaultBuckets([]float64{0.0, 1.0, 5.0, 10.0}),
+		WithAsBuiltIn(),
+	)
+}
+
+func (m *Metrics) TestingHistogramRecord(ctx context.Context, value float64) {
+	m.Record(ctx, nameTestingHistogram, value, InstAttribs{})
+}
+
+func addTestingCounter(ctx context.Context, m *Metrics) error {
+	return m.CreateInstrument(Int64Counter,
+		nameTestingCounter,
+		"Testing Error Counting Metric",
+		"{errors}",
+		WithAsBuiltIn(),
+	)
+}
+
+func (m *Metrics) TestingErrorA(ctx context.Context) {
+	m.AddInt(ctx, nameTestingCounter, 1, InstAttribs{{Name: AttribErrorCause, Value: errorCauseTestingB}})
+}
+
+func (m *Metrics) TestingErrorB(ctx context.Context) {
+	m.AddInt(ctx, nameTestingCounter, 1, InstAttribs{{Name: AttribErrorCause, Value: errorCauseTestingB}})
+}