Merge pull request #236 from coroot/otel_fix_service_detection

otel: fix service detection for traces and logs

api/views/logs/logs.go

@@ -14,7 +14,6 @@ import (
 	"github.com/coroot/coroot/timeseries"
 	"github.com/coroot/coroot/utils"
 	"github.com/coroot/logparser"
-	"golang.org/x/exp/maps"
 	"k8s.io/klog"
 )
 
@@ -133,22 +132,32 @@ func renderEntries(ctx context.Context, v *View, ch *clickhouse.Client, app *mod
 		return
 	}
 
-	service := ""
-	if appSettings != nil && appSettings.Logs != nil {
-		service = appSettings.Logs.Service
-	} else {
-		service = model.GuessService(maps.Keys(services), app.Id)
-	}
+	var logsFromAgentFound bool
+	var otelServices []string
 	for s := range services {
 		if strings.HasPrefix(s, "/") {
-			v.Sources = append(v.Sources, model.LogSourceAgent)
+			logsFromAgentFound = true
 		} else {
-			v.Services = append(v.Services, s)
-			if s == service {
-				v.Service = s
-				v.Sources = append(v.Sources, model.LogSourceOtel)
-			}
+			otelServices = append(otelServices, s)
+		}
+	}
+	otelService := ""
+	if appSettings != nil && appSettings.Logs != nil {
+		otelService = appSettings.Logs.Service
+	} else {
+		otelService = model.GuessService(otelServices, app.Id)
+	}
+
+	if logsFromAgentFound {
+		v.Sources = append(v.Sources, model.LogSourceAgent)
+	}
+
+	for _, s := range otelServices {
+		if s == otelService {
+			v.Service = s
+			v.Sources = append(v.Sources, model.LogSourceOtel)
 		}
+		v.Services = append(v.Services, s)
 	}
 	sort.Strings(v.Services)
 
@@ -172,15 +181,15 @@ func renderEntries(ctx context.Context, v *View, ch *clickhouse.Client, app *mod
 	var entries []*model.LogEntry
 	switch v.Source {
 	case model.LogSourceOtel:
-		v.Message = fmt.Sprintf("Using OpenTelemetry logs of <i>%s</i>", service)
+		v.Message = fmt.Sprintf("Using OpenTelemetry logs of <i>%s</i>", otelService)
 		v.Severities = services[v.Service]
 		if len(v.Severity) == 0 {
 			v.Severity = v.Severities
 		}
 		if v.View == viewMessages {
-			histogram, err = ch.GetServiceLogsHistogram(ctx, w.Ctx.From, w.Ctx.To, w.Ctx.Step, service, v.Severity, q.Search)
+			histogram, err = ch.GetServiceLogsHistogram(ctx, w.Ctx.From, w.Ctx.To, w.Ctx.Step, otelService, v.Severity, q.Search)
 			if err == nil {
-				entries, err = ch.GetServiceLogs(ctx, w.Ctx.From, w.Ctx.To, service, v.Severity, q.Search, q.Limit)
+				entries, err = ch.GetServiceLogs(ctx, w.Ctx.From, w.Ctx.To, otelService, v.Severity, q.Search, q.Limit)
 			}
 		}
 	case model.LogSourceAgent:

api/views/tracing/tracing.go

@@ -88,35 +88,38 @@ func Render(ctx context.Context, ch *clickhouse.Client, app *model.Application,
 		v.Message = fmt.Sprintf("Clickhouse error: %s", err)
 		return v
 	}
-	service := ""
-	if appSettings != nil && appSettings.Tracing != nil {
-		service = appSettings.Tracing.Service
-	} else {
-		service = model.GuessService(services, app.Id)
-	}
-	var serviceFound, ebpfSpansFound bool
+
+	var ebpfSpansFound bool
+	var otelServices []string
 	for _, s := range services {
 		if strings.HasPrefix(s, "/") {
 			ebpfSpansFound = true
 		} else {
-			if s == service {
-				serviceFound = true
-			}
-			v.Services = append(v.Services, Service{Name: s, Linked: s == service})
+			otelServices = append(otelServices, s)
 		}
 	}
+
+	var otelService string
+	if appSettings != nil && appSettings.Tracing != nil {
+		otelService = appSettings.Tracing.Service
+	} else {
+		otelService = model.GuessService(otelServices, app.Id)
+	}
+	for _, s := range otelServices {
+		v.Services = append(v.Services, Service{Name: s, Linked: s == otelService})
+	}
 	sort.Slice(v.Services, func(i, j int) bool {
 		return v.Services[i].Name < v.Services[j].Name
 	})
 
-	if serviceFound {
+	if len(otelServices) > 0 {
 		v.Sources = append(v.Sources, Source{Type: model.TraceSourceOtel, Name: "OpenTelemetry"})
 	}
 	if ebpfSpansFound {
 		v.Sources = append(v.Sources, Source{Type: model.TraceSourceAgent, Name: "OpenTelemetry (eBPF)"})
 	}
 
-	if !serviceFound && !ebpfSpansFound {
+	if len(v.Sources) == 0 {
 		v.Status = model.UNKNOWN
 		v.Message = "No traces found"
 		return v
@@ -129,7 +132,7 @@ func Render(ctx context.Context, ch *clickhouse.Client, app *model.Application,
 	case traceId != "":
 		spans, err = ch.GetSpansByTraceId(ctx, traceId)
 
-	case (source == "" || source == model.TraceSourceOtel) && serviceFound:
+	case (source == "" || source == model.TraceSourceOtel) && otelService != "":
 		source = model.TraceSourceOtel
 		var ignoredPeerAddrs []string
 		if !app.Category.Monitoring() {
@@ -144,7 +147,7 @@ func Render(ctx context.Context, ch *clickhouse.Client, app *model.Application,
 				Ctx:              w.Ctx,
 				ExcludePeerAddrs: ignoredPeerAddrs,
 			}
-			sq.Filters = append(sq.Filters, clickhouse.NewSpanFilter("ServiceName", "=", service))
+			sq.Filters = append(sq.Filters, clickhouse.NewSpanFilter("ServiceName", "=", otelService))
 			histogram, e = ch.GetSpansByServiceNameHistogram(ctx, sq)
 			if e != nil {
 				err = e
@@ -164,7 +167,7 @@ func Render(ctx context.Context, ch *clickhouse.Client, app *model.Application,
 				Limit:            limit,
 				ExcludePeerAddrs: ignoredPeerAddrs,
 			}
-			sq.Filters = append(sq.Filters, clickhouse.NewSpanFilter("ServiceName", "=", service))
+			sq.Filters = append(sq.Filters, clickhouse.NewSpanFilter("ServiceName", "=", otelService))
 			spans, e = ch.GetSpansByServiceName(ctx, sq)
 			if e != nil {
 				err = e
@@ -222,7 +225,7 @@ func Render(ctx context.Context, ch *clickhouse.Client, app *model.Application,
 
 	switch source {
 	case model.TraceSourceOtel:
-		v.Message = fmt.Sprintf("Using traces of <i>%s</i>", service)
+		v.Message = fmt.Sprintf("Using traces of <i>%s</i>", otelService)
 	case model.TraceSourceAgent:
 		v.Message = "Using data gathered by the eBPF tracer"
 	}