chore(orch): add more info for sandboxes (#2174)

* chore(orch): add more telemetry for sandboxes

* chore: add more event data

* chore: auto-commit generated changes

* chore: refactor to prevent accidentally using old fc version

* chore: namespace to sandbox.*

* fix: test

---------

Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>

Author: jakubno

packages/api/internal/handlers/sandbox_create.go

@@ -116,8 +116,8 @@ func (a *APIStore) PostSandboxes(c *gin.Context) {
 		attribute.String("env.team.id", teamInfo.Team.ID.String()),
 		telemetry.WithTemplateID(env.TemplateID),
 		attribute.String("env.alias", alias),
-		attribute.String("env.kernel.version", build.KernelVersion),
-		attribute.String("env.firecracker.version", build.FirecrackerVersion),
+		telemetry.WithKernelVersion(build.KernelVersion),
+		telemetry.WithFirecrackerVersion(build.FirecrackerVersion),
 	)
 
 	autoPause := sharedUtils.DerefOrDefault(body.AutoPause, sandbox.AutoPauseDefault)

packages/orchestrator/internal/sandbox/map.go

@@ -93,6 +93,9 @@ func (m *Map) Insert(ctx context.Context, sbx *Sandbox) {
 		logger.WithTemplateID(sbx.Runtime.TemplateID),
 		logger.WithBuildID(sbx.Runtime.BuildID),
 		logger.WithSandboxIP(sbx.Slot.HostIPString()),
+		logger.WithEnvdVersion(sbx.Config.Envd.Version),
+		logger.WithKernelVersion(sbx.Config.FirecrackerConfig.KernelVersion),
+		logger.WithFirecrackerVersion(sbx.Config.FirecrackerConfig.FirecrackerVersion),
 	)
 
 	m.sandboxes.Insert(sbx.Runtime.SandboxID, sbx)

packages/orchestrator/internal/server/sandboxes.go

@@ -78,11 +78,11 @@ func (s *Server) Create(ctx context.Context, req *orchestrator.SandboxCreateRequ
 	}()
 
 	childSpan.SetAttributes(
+		telemetry.WithBuildID(req.GetSandbox().GetBuildId()),
 		telemetry.WithTemplateID(req.GetSandbox().GetTemplateId()),
-		attribute.String("kernel.version", req.GetSandbox().GetKernelVersion()),
+		telemetry.WithKernelVersion(req.GetSandbox().GetKernelVersion()),
 		telemetry.WithSandboxID(req.GetSandbox().GetSandboxId()),
-		attribute.String("client.id", s.info.ClientId),
-		attribute.String("envd.version", req.GetSandbox().GetEnvdVersion()),
+		telemetry.WithEnvdVersion(req.GetSandbox().GetEnvdVersion()),
 	)
 
 	// setup launch darkly
@@ -156,46 +156,52 @@ func (s *Server) Create(ctx context.Context, req *orchestrator.SandboxCreateRequ
 	}
 
 	resolvedFCVersion := featureflags.ResolveFirecrackerVersion(ctx, s.featureFlags, req.GetSandbox().GetFirecrackerVersion())
-
 	volumeMounts, err := createVolumeMountModelsFromAPI(req.GetSandbox().GetVolumeMounts())
 	if err != nil {
 		return nil, fmt.Errorf("failed to convert volume mounts: %w", err)
 	}
 
-	sbx, err := s.sandboxFactory.ResumeSandbox(
-		ctx,
-		template,
-		sandbox.NewConfig(sandbox.Config{
-			BaseTemplateID: req.GetSandbox().GetBaseTemplateId(),
+	config := sandbox.NewConfig(sandbox.Config{
+		BaseTemplateID: req.GetSandbox().GetBaseTemplateId(),
 
-			Vcpu:            req.GetSandbox().GetVcpu(),
-			RamMB:           req.GetSandbox().GetRamMb(),
-			TotalDiskSizeMB: req.GetSandbox().GetTotalDiskSizeMb(),
-			HugePages:       req.GetSandbox().GetHugePages(),
+		Vcpu:            req.GetSandbox().GetVcpu(),
+		RamMB:           req.GetSandbox().GetRamMb(),
+		TotalDiskSizeMB: req.GetSandbox().GetTotalDiskSizeMb(),
+		HugePages:       req.GetSandbox().GetHugePages(),
 
-			Network: network,
+		Network: network,
 
-			Envd: sandbox.EnvdMetadata{
-				Version:     req.GetSandbox().GetEnvdVersion(),
-				AccessToken: req.GetSandbox().EnvdAccessToken,
-				Vars:        req.GetSandbox().GetEnvVars(),
-			},
-
-			FirecrackerConfig: fc.Config{
-				KernelVersion:      req.GetSandbox().GetKernelVersion(),
-				FirecrackerVersion: resolvedFCVersion,
-			},
+		Envd: sandbox.EnvdMetadata{
+			Version:     req.GetSandbox().GetEnvdVersion(),
+			AccessToken: req.GetSandbox().EnvdAccessToken,
+			Vars:        req.GetSandbox().GetEnvVars(),
+		},
 
-			VolumeMounts: volumeMounts,
-		}),
-		sandbox.RuntimeMetadata{
-			TemplateID:  req.GetSandbox().GetTemplateId(),
-			SandboxID:   req.GetSandbox().GetSandboxId(),
-			ExecutionID: req.GetSandbox().GetExecutionId(),
-			TeamID:      req.GetSandbox().GetTeamId(),
-			BuildID:     req.GetSandbox().GetBuildId(),
-			SandboxType: sandbox.SandboxTypeSandbox,
+		FirecrackerConfig: fc.Config{
+			KernelVersion:      req.GetSandbox().GetKernelVersion(),
+			FirecrackerVersion: resolvedFCVersion,
 		},
+
+		VolumeMounts: volumeMounts,
+	})
+	childSpan.SetAttributes(
+		telemetry.WithFirecrackerVersion(config.FirecrackerConfig.FirecrackerVersion),
+	)
+
+	runtime := sandbox.RuntimeMetadata{
+		TemplateID:  req.GetSandbox().GetTemplateId(),
+		SandboxID:   req.GetSandbox().GetSandboxId(),
+		ExecutionID: req.GetSandbox().GetExecutionId(),
+		TeamID:      req.GetSandbox().GetTeamId(),
+		BuildID:     req.GetSandbox().GetBuildId(),
+		SandboxType: sandbox.SandboxTypeSandbox,
+	}
+
+	sbx, err := s.sandboxFactory.ResumeSandbox(
+		ctx,
+		template,
+		config,
+		runtime,
 		req.GetStartTime().AsTime(),
 		req.GetEndTime().AsTime(),
 		req.GetSandbox(),
@@ -210,6 +216,14 @@ func (s *Server) Create(ctx context.Context, req *orchestrator.SandboxCreateRequ
 
 		err = errors.Join(err, context.Cause(ctx))
 		telemetry.ReportCriticalError(ctx, "failed to create sandbox", err)
+		logger.L().Error(ctx, "failed to create sandbox", zap.Error(err),
+			logger.WithSandboxID(runtime.SandboxID),
+			logger.WithBuildID(runtime.BuildID),
+			logger.WithTemplateID(runtime.TemplateID),
+			logger.WithEnvdVersion(config.Envd.Version),
+			logger.WithKernelVersion(config.FirecrackerConfig.KernelVersion),
+			logger.WithFirecrackerVersion(config.FirecrackerConfig.FirecrackerVersion),
+		)
 
 		return nil, status.Errorf(codes.Internal, "failed to create sandbox: %s", err)
 	}

packages/orchestrator/internal/server/sandboxes_test.go

@@ -50,6 +50,7 @@ func Test_server_List(t *testing.T) {
 						Runtime: sandbox.RuntimeMetadata{
 							SandboxID: id.Generate(),
 						},
+						Config: sandbox.NewConfig(sandbox.Config{}),
 					},
 					Resources: &sandbox.Resources{
 						Slot: &network.Slot{HostIP: net.IPv4(127, 0, 0, 1)},

packages/shared/pkg/logger/fields.go

@@ -53,6 +53,14 @@ func WithEnvdVersion(envdVersion string) zap.Field {
 	return zap.String("envd.version", envdVersion)
 }
 
+func WithKernelVersion(kernelVersion string) zap.Field {
+	return zap.String("sandbox.kernel.version", kernelVersion)
+}
+
+func WithFirecrackerVersion(firecrackerVersion string) zap.Field {
+	return zap.String("sandbox.firecracker.version", firecrackerVersion)
+}
+
 func WithClientIP(clientIP string) zap.Field {
 	return zap.String("http.client_ip", clientIP)
 }

packages/shared/pkg/telemetry/fields.go

@@ -48,6 +48,14 @@ func WithEnvdVersion(envdVersion string) attribute.KeyValue {
 	return zapFieldToOTELAttribute(logger.WithEnvdVersion(envdVersion))
 }
 
+func WithKernelVersion(kernelVersion string) attribute.KeyValue {
+	return zapFieldToOTELAttribute(logger.WithKernelVersion(kernelVersion))
+}
+
+func WithFirecrackerVersion(firecrackerVersion string) attribute.KeyValue {
+	return zapFieldToOTELAttribute(logger.WithFirecrackerVersion(firecrackerVersion))
+}
+
 func WithMaskedAPIKey(maskedAPIKey string) attribute.KeyValue {
 	return zapFieldToOTELAttribute(logger.WithMaskedAPIKey(maskedAPIKey))
 }