telemetry: use TRACEPARENT environment variable

Start traces using TRACEPARENT instead of parsing a custom string.

cmd/deploy/main.go

@@ -9,6 +9,8 @@ import (
 	"github.com/nais/deploy/pkg/pb"
 	"github.com/nais/deploy/pkg/telemetry"
 	"github.com/nais/deploy/pkg/version"
+	"go.opentelemetry.io/otel/attribute"
+	otrace "go.opentelemetry.io/otel/trace"
 	"google.golang.org/protobuf/encoding/protojson"
 
 	log "github.com/sirupsen/logrus"
@@ -52,15 +54,29 @@ func run() error {
 		}
 	}()
 
+	// Inherit traceparent from pipeline, if any
+	ctx := telemetry.WithTraceParent(programContext, cfg.Traceparent)
+	ctx, span := telemetry.Tracer().Start(ctx, "NAIS deploy", otrace.WithSpanKind(otrace.SpanKindClient))
+	defer span.End()
+
+	span.SetAttributes(attribute.KeyValue{
+		Key:   "deploy.client.version",
+		Value: attribute.StringValue(version.Version()),
+	})
+
 	// Welcome
 	log.Infof("NAIS deploy %s", version.Version())
 	ts, err := version.BuildTime()
 	if err == nil {
+		span.SetAttributes(attribute.KeyValue{
+			Key:   "deploy.client.build-time",
+			Value: attribute.StringValue(ts.Local().String()),
+		})
 		log.Infof("This version was built %s", ts.Local())
 	}
 
 	// Prepare request
-	request, err := deployclient.Prepare(programContext, cfg)
+	request, err := deployclient.Prepare(ctx, cfg)
 	if err != nil {
 		return err
 	}
@@ -89,5 +105,5 @@ func run() error {
 		return nil
 	}
 
-	return d.Deploy(programContext, cfg, request)
+	return d.Deploy(ctx, cfg, request)
 }

pkg/deployclient/config.go

@@ -32,6 +32,7 @@ type Config struct {
 	Retry                     bool
 	RetryInterval             time.Duration
 	Team                      string
+	Traceparent               string
 	TelemetryInput            string
 	Telemetry                 *telemetry.PipelineTimings
 	Timeout                   time.Duration
@@ -61,6 +62,7 @@ func InitConfig(cfg *Config) {
 	flag.BoolVar(&cfg.Retry, "retry", getEnvBool("RETRY", true), "Retry deploy when encountering transient errors. (env RETRY)")
 	flag.StringVar(&cfg.Team, "team", os.Getenv("TEAM"), "Team making the deployment. Auto-detected from nais.yaml if possible. (env TEAM)")
 	flag.StringVar(&cfg.OpenTelemetryCollectorURL, "otel-collector-endpoint", getEnv("OTEL_COLLECTOR_ENDPOINT", DefaultOtelCollectorEndpoint), "OpenTelemetry collector endpoint. (env OTEL_COLLECTOR_ENDPOINT)")
+	flag.StringVar(&cfg.Traceparent, "traceparent", os.Getenv("TRACEPARENT"), "The W3C Trace Context traceparent value for the workflow run. (env TRACEPARENT)")
 	flag.StringVar(&cfg.TelemetryInput, "telemetry", os.Getenv("TELEMETRY"), "Telemetry data from CI pipeline. (env TELEMETRY)")
 	flag.DurationVar(&cfg.Timeout, "timeout", getEnvDuration("TIMEOUT", DefaultDeployTimeout), "Time to wait for successful deployment. (env TIMEOUT)")
 	flag.StringVar(&cfg.TracingDashboardURL, "tracing-dashboard-url", getEnv("TRACING_DASHBOARD_URL", DefaultTracingDashboardURL), "Base URL to Grafana tracing dashboard onto which the trace ID can be appended (env TRACING_DASHBOARD_URL)")

pkg/deployclient/deployclient.go

@@ -161,43 +161,43 @@ func (d *Deployer) Deploy(ctx context.Context, cfg *Config, deployRequest *pb.De
 
 	// Root span for tracing.
 	// All sub-spans must be created from this context.
-	ctx, rootSpan := cfg.Telemetry.StartTracing(ctx)
-	defer rootSpan.End()
+	ctx, span := telemetry.Tracer().Start(ctx, "Send deploy request and wait for completion")
+	defer span.End()
 	deployRequest.TraceParent = telemetry.TraceParentHeader(ctx)
 
 	log.Infof("Sending deployment request to NAIS deploy at %s...", cfg.DeployServerURL)
 
 	sendDeploymentRequest := func() error {
-		ctx, span := telemetry.Tracer().Start(ctx, "Send to deploy server")
-		defer span.End()
+		requestContext, requestSpan := telemetry.Tracer().Start(ctx, "Send to deploy server")
+		defer requestSpan.End()
 
 		err = retryUnavailable(cfg.RetryInterval, cfg.Retry, func() error {
-			deployStatus, err = d.Client.Deploy(ctx, deployRequest)
+			deployStatus, err = d.Client.Deploy(requestContext, deployRequest)
 			return err
 		})
 
 		if err != nil {
 			code := grpcErrorCode(err)
 			err = fmt.Errorf(formatGrpcError(err))
-			if ctx.Err() != nil {
-				span.SetStatus(ocodes.Error, ctx.Err().Error())
-				return Errorf(ExitTimeout, "deployment timed out: %s", ctx.Err())
+			if requestContext.Err() != nil {
+				requestSpan.SetStatus(ocodes.Error, requestContext.Err().Error())
+				return Errorf(ExitTimeout, "deployment timed out: %s", requestContext.Err())
 			}
 			if code == codes.Unauthenticated {
 				if !strings.HasSuffix(cfg.Environment, ":"+cfg.Team) {
 					log.Warnf("hint: team %q does not match namespace in %q", cfg.Team, cfg.Environment)
 				}
 			}
-			span.SetStatus(ocodes.Error, err.Error())
+			requestSpan.SetStatus(ocodes.Error, err.Error())
 			return ErrorWrap(ExitNoDeployment, err)
 		}
 
 		log.Infof("Deployment request accepted by NAIS deploy and dispatched to cluster '%s'.", deployStatus.GetRequest().GetCluster())
 
 		deployRequest.ID = deployStatus.GetRequest().GetID()
-		telemetry.AddDeploymentRequestSpanAttributes(rootSpan, deployStatus.GetRequest())
 		telemetry.AddDeploymentRequestSpanAttributes(span, deployStatus.GetRequest())
-		traceID := telemetry.TraceID(ctx)
+		telemetry.AddDeploymentRequestSpanAttributes(requestSpan, deployStatus.GetRequest())
+		traceID := telemetry.TraceID(requestContext)
 
 		urlPrefix := "https://" + strings.Split(cfg.DeployServerURL, ":")[0]
 		log.Infof("Deployment information:")
@@ -214,7 +214,12 @@ func (d *Deployer) Deploy(ctx context.Context, cfg *Config, deployRequest *pb.De
 	}
 
 	err = sendDeploymentRequest()
+
+	// First handle errors that might have occurred with the request itself.
+	// Errors from underlying systems are handled later.
 	if err != nil {
+		span.SetStatus(ocodes.Error, err.Error())
+		span.RecordError(err)
 		return err
 	}
 

pkg/telemetry/telemetry.go

@@ -124,6 +124,9 @@ func AddDeploymentRequestSpanAttributes(span otrace.Span, request *pb.Deployment
 		}, attribute.KeyValue{
 			Key:   "deploy.repository",
 			Value: attribute.StringValue(request.GetRepository().FullName()),
+		}, attribute.KeyValue{
+			Key:   "deploy.deadline",
+			Value: attribute.StringValue(request.GetDeadline().AsTime().Local().Format(time.RFC3339)),
 		},
 	)
 }