feat: phase 8 serviceadapter crd rollout (32/33 tasks)

internal/observability/metrics_test.go

@@ -0,0 +1,162 @@
+package observability
+
+import (
+	"strings"
+	"testing"
+	"time"
+)
+
+func TestMetricsRecordRequest(t *testing.T) {
+	m := NewMetrics()
+
+	// Record some requests
+	m.RecordRequest("v1-chat", "reasoning", 200, 500*time.Millisecond)
+	m.RecordRequest("v1-chat", "reasoning", 200, 600*time.Millisecond)
+	m.RecordRequest("v1-chat", "reasoning", 500, 100*time.Millisecond)
+
+	metrics := m.GetMetrics()
+	requestTotal := metrics["request_total"].(map[string]int64)
+
+	if requestTotal["v1-chat:reasoning:200"] != 2 {
+		t.Errorf("expected 2 successful requests, got %d", requestTotal["v1-chat:reasoning:200"])
+	}
+
+	if requestTotal["v1-chat:reasoning:500"] != 1 {
+		t.Errorf("expected 1 error request, got %d", requestTotal["v1-chat:reasoning:500"])
+	}
+}
+
+func TestMetricsRecordBytes(t *testing.T) {
+	m := NewMetrics()
+
+	m.RecordBytesIn("v1-chat", "reasoning", 1024)
+	m.RecordBytesOut("v1-chat", "reasoning", 2048)
+
+	metrics := m.GetMetrics()
+	bytesIn := metrics["bytes_in"].(map[string]int64)
+	bytesOut := metrics["bytes_out"].(map[string]int64)
+
+	if bytesIn["v1-chat:reasoning"] != 1024 {
+		t.Errorf("expected 1024 bytes in, got %d", bytesIn["v1-chat:reasoning"])
+	}
+
+	if bytesOut["v1-chat:reasoning"] != 2048 {
+		t.Errorf("expected 2048 bytes out, got %d", bytesOut["v1-chat:reasoning"])
+	}
+}
+
+func TestMetricsUpstreamHealth(t *testing.T) {
+	m := NewMetrics()
+
+	m.SetUpstreamHealth("reasoning", true)
+	m.SetUpstreamHealth("embedding", false)
+
+	metrics := m.GetMetrics()
+	health := metrics["upstream_health"].(map[string]int)
+
+	if health["reasoning"] != 1 {
+		t.Errorf("expected reasoning upstream healthy (1), got %d", health["reasoning"])
+	}
+
+	if health["embedding"] != 0 {
+		t.Errorf("expected embedding upstream unhealthy (0), got %d", health["embedding"])
+	}
+}
+
+func TestExportPrometheus(t *testing.T) {
+	m := NewMetrics()
+
+	// Record some data
+	m.RecordRequest("v1-chat", "reasoning", 200, 500*time.Millisecond)
+	m.RecordBytesIn("v1-chat", "reasoning", 1024)
+	m.RecordBytesOut("v1-chat", "reasoning", 2048)
+	m.SetUpstreamHealth("reasoning", true)
+
+	export := m.ExportPrometheus()
+
+	// Check for expected metric families
+	if !strings.Contains(export, "# HELP gateway_requests_total") {
+		t.Errorf("missing gateway_requests_total help")
+	}
+
+	if !strings.Contains(export, "# TYPE gateway_requests_total counter") {
+		t.Errorf("missing gateway_requests_total type")
+	}
+
+	if !strings.Contains(export, "gateway_requests_total{route=\"v1-chat\",upstream=\"reasoning\",status=\"200\"} 1") {
+		t.Errorf("missing or incorrect request_total metric")
+	}
+
+	if !strings.Contains(export, "# HELP gateway_bytes_in_total") {
+		t.Errorf("missing gateway_bytes_in_total help")
+	}
+
+	if !strings.Contains(export, "gateway_bytes_in_total{route=\"v1-chat\",upstream=\"reasoning\"} 1024") {
+		t.Errorf("missing or incorrect bytes_in metric")
+	}
+
+	if !strings.Contains(export, "gateway_bytes_out_total{route=\"v1-chat\",upstream=\"reasoning\"} 2048") {
+		t.Errorf("missing or incorrect bytes_out metric")
+	}
+
+	if !strings.Contains(export, "gateway_upstream_health{upstream=\"reasoning\"} 1") {
+		t.Errorf("missing or incorrect upstream_health metric")
+	}
+}
+
+func TestExportPrometheusHistogram(t *testing.T) {
+	m := NewMetrics()
+
+	// Record requests with different durations
+	m.RecordRequest("v1-chat", "reasoning", 200, 50*time.Millisecond)
+	m.RecordRequest("v1-chat", "reasoning", 200, 200*time.Millisecond)
+	m.RecordRequest("v1-chat", "reasoning", 200, 1*time.Second)
+
+	export := m.ExportPrometheus()
+
+	// Check for histogram structure
+	if !strings.Contains(export, "# HELP gateway_request_duration_seconds Request latency in seconds") {
+		t.Errorf("missing duration_seconds help")
+	}
+
+	if !strings.Contains(export, "# TYPE gateway_request_duration_seconds histogram") {
+		t.Errorf("missing histogram type")
+	}
+
+	if !strings.Contains(export, "gateway_request_duration_seconds_bucket") {
+		t.Errorf("missing histogram bucket")
+	}
+
+	if !strings.Contains(export, "gateway_request_duration_seconds_count") {
+		t.Errorf("missing histogram count")
+	}
+}
+
+func TestMetricsThreadSafety(t *testing.T) {
+	m := NewMetrics()
+
+	// Concurrent recordings
+	done := make(chan bool, 2)
+
+	go func() {
+		for i := 0; i < 100; i++ {
+			m.RecordRequest("route1", "upstream1", 200, time.Millisecond)
+		}
+		done <- true
+	}()
+
+	go func() {
+		for i := 0; i < 100; i++ {
+			m.RecordBytesIn("route2", "upstream2", 1024)
+		}
+		done <- true
+	}()
+
+	<-done
+	<-done
+
+	metrics := m.GetMetrics()
+	if len(metrics["request_total"].(map[string]int64)) == 0 {
+		t.Errorf("expected metrics to be recorded")
+	}
+}