browser/platform/service/distribution_build_execution.go

package service

import (
	"encoding/base64"
	"encoding/json"
	"errors"
	"strings"

	"browser.local/platform/domain"
	"browser.local/platform/repo"
	"browser.local/platform/validator"
)

const platformDistributionBuilderEndpointID = "platform-distribution-builder"

// unconfiguredDistributionBuilder stands in when no platform builder has been
// installed. It reports a platform-builder reason so an operator is not sent
// looking at run endpoint capabilities.
type unconfiguredDistributionBuilder struct{}

func (unconfiguredDistributionBuilder) Readiness() (bool, string) {
	return false, "platform builder is not configured"
}

func (unconfiguredDistributionBuilder) Build(domain.DistributionBuildInput) ([]byte, error) {
	return nil, validationError("platform builder is not configured")
}

func (svc *CoreService) configuredDistributionBuilder() DistributionBuilder {
	svc.distributionBuildMu.Lock()
	defer svc.distributionBuildMu.Unlock()
	if svc.distributionBuilder == nil {
		return unconfiguredDistributionBuilder{}
	}
	return svc.distributionBuilder
}

// distributionBuilderReadiness reports platform builder readiness. The reason
// always names the platform builder, never a run endpoint capability.
func (svc *CoreService) distributionBuilderReadiness() (bool, string) {
	builder := svc.configuredDistributionBuilder()
	ready, reason := builder.Readiness()
	if ready {
		return true, ""
	}
	reason = strings.TrimSpace(reason)
	if reason == "" {
		reason = "platform builder is unavailable"
	} else if !strings.Contains(strings.ToLower(reason), "platform builder") {
		reason = "platform builder is unavailable: " + reason
	}
	return false, reason
}

func (svc *CoreService) enqueueDistributionBuild(job domain.Job) {
	svc.distributionBuildMu.Lock()
	if _, exists := svc.distributionBuilds[job.ID]; exists {
		svc.distributionBuildMu.Unlock()
		return
	}
	svc.distributionBuilds[job.ID] = struct{}{}
	svc.distributionBuildMu.Unlock()
	go func() {
		defer func() {
			svc.distributionBuildMu.Lock()
			delete(svc.distributionBuilds, job.ID)
			svc.distributionBuildMu.Unlock()
		}()
		_ = svc.executeDistributionBuild(job)
	}()
}

// executeDistributionBuild claims the build job internally and completes it with
// the artifact produced by the platform builder. Build work is never dispatched
// to a machine-side run endpoint, so the component auth key stays on the
// platform.
func (svc *CoreService) executeDistributionBuild(job domain.Job) error {
	if job.Capability != domain.JobCapabilityDistributionBuild {
		return validationError("job is not a distribution build")
	}
	input, err := svc.platformDistributionBuildInput(job)
	if err != nil {
		return svc.failDistributionBuildJob(job, "platform builder could not assemble build input")
	}
	if err := svc.markDistributionBuildRunning(&job); err != nil {
		return err
	}
	if isTerminalJobState(job.State) {
		return nil
	}
	builder := svc.configuredDistributionBuilder()
	var payload []byte
	var buildErr error
	if progressBuilder, ok := builder.(distributionBuilderWithProgress); ok {
		payload, buildErr = progressBuilder.BuildWithProgress(input, func(progress DistributionBuildProgress) {
			_ = svc.updateDistributionBuildProgress(job, progress)
		})
	} else {
		payload, buildErr = builder.Build(input)
	}
	if buildErr != nil {
		return svc.failDistributionBuildJob(job, builderJobFailureMessage(buildErr))
	}
	if _, err := svc.platformDistributionBuildInput(job); err != nil {
		return svc.failDistributionBuildJob(job, "platform builder discarded output because the component key is no longer current")
	}
	if err := svc.storeDistributionBuildArtifact(input.ArtifactID, job.ID, payload); err != nil {
		return svc.failDistributionBuildJob(job, "platform builder could not record the distribution artifact")
	}
	return svc.succeedDistributionBuildJob(job, input.ArtifactID)
}

// platformDistributionBuildInput resolves the build input, including the
// component auth key, inside the platform. Unlike GetDistributionBuildInput it
// never crosses the job channel.
func (svc *CoreService) platformDistributionBuildInput(job domain.Job) (domain.DistributionBuildInput, error) {
	runDistributions, err := svc.store.RunDistributions().List(domain.RunDistributionFilter{ServerInstanceID: job.ServerInstanceID})
	if err != nil {
		return domain.DistributionBuildInput{}, err
	}
	for _, distribution := range runDistributions {
		if distribution.BuildJobID != job.ID {
			continue
		}
		key, err := svc.activeComponentKey(distribution.ServerInstanceID, domain.DistributionComponentRun, "")
		if err != nil {
			return domain.DistributionBuildInput{}, err
		}
		if key.Generation != distribution.KeyGeneration {
			return domain.DistributionBuildInput{}, validationError("run build key generation is no longer current")
		}
		plainKey, err := svc.decryptRuntimeKey(key.EncryptedKey)
		if err != nil {
			return domain.DistributionBuildInput{}, err
		}
		packageInput, err := svc.runDistributionPackageContext(distribution)
		if err != nil {
			return domain.DistributionBuildInput{}, err
		}
		return domain.DistributionBuildInput{
			JobID:               job.ID,
			ComponentKind:       domain.DistributionComponentRun,
			ServerInstanceID:    distribution.ServerInstanceID,
			PluginID:            distribution.PluginID,
			RunEndpointID:       distribution.RunEndpointID,
			ProfileKey:          packageInput.profileKey,
			TargetOS:            distribution.TargetOS,
			TargetArch:          distribution.TargetArch,
			TargetRelease:       distribution.ID,
			PlatformURL:         runReleasePlatformURL(),
			PackageFormat:       distribution.PackageFormat,
			ArtifactID:          distribution.ArtifactID,
			OutputFilename:      executableFilename("run", distribution.TargetOS),
			SecretRef:           distribution.SecretRef,
			KeyGeneration:       distribution.KeyGeneration,
			AuthKey:             plainKey,
			WorkspaceSeed:       packageInput.workspaceSeed,
			AutonomousLifecycle: packageInput.autonomousLifecycle,
		}, nil
	}

	clientDistributions, err := svc.store.ClientManagerDistributions().List(domain.ClientManagerDistributionFilter{ServerInstanceID: job.ServerInstanceID})
	if err != nil {
		return domain.DistributionBuildInput{}, err
	}
	for _, distribution := range clientDistributions {
		if distribution.BuildJobID != job.ID {
			continue
		}
		key, err := svc.activeComponentKey(distribution.ServerInstanceID, domain.DistributionComponentClientManager, distribution.ProfileKey)
		if err != nil {
			return domain.DistributionBuildInput{}, err
		}
		if key.Generation != distribution.KeyGeneration {
			return domain.DistributionBuildInput{}, validationError("client-manager build key generation is no longer current")
		}
		plainKey, err := svc.decryptRuntimeKey(key.EncryptedKey)
		if err != nil {
			return domain.DistributionBuildInput{}, err
		}
		return domain.DistributionBuildInput{
			JobID:            job.ID,
			ComponentKind:    domain.DistributionComponentClientManager,
			ServerInstanceID: distribution.ServerInstanceID,
			PluginID:         distribution.PluginID,
			RunEndpointID:    job.RunEndpointID,
			ProfileKey:       distribution.ProfileKey,
			TargetOS:         distribution.TargetOS,
			TargetArch:       distribution.TargetArch,
			PlatformURL:      runReleasePlatformURL(),
			PackageFormat:    packageFormatForTarget(distribution.TargetOS),
			RepositoryURL:    distribution.RepositoryURL,
			SourceRevision:   distribution.SourceRevision,
			ArtifactID:       distribution.ArtifactID,
			OutputFilename:   clientManagerOutputName(distribution.ProfileKey, distribution.TargetOS),
			SecretRef:        distribution.SecretRef,
			KeyGeneration:    distribution.KeyGeneration,
			AuthKey:          plainKey,
		}, nil
	}
	return domain.DistributionBuildInput{}, repo.ErrNotFound
}

type runDistributionPackageContext struct {
	profileKey          string
	workspaceSeed       string
	autonomousLifecycle *domain.RunAutonomousLifecyclePlan
}

func (svc *CoreService) runDistributionPackageContext(distribution domain.RunDistribution) (runDistributionPackageContext, error) {
	instance, err := svc.store.ServerInstances().Get(distribution.ServerInstanceID)
	if err != nil {
		return runDistributionPackageContext{}, err
	}
	plugin, err := svc.store.GamePlugins().Get(distribution.PluginID)
	if err != nil {
		return runDistributionPackageContext{}, err
	}
	profileKey := instance.Deployment.ProfileKey
	bindings := map[string]string(nil)
	if binding, bindingErr := svc.runtimeBindingForServer(distribution.ServerInstanceID); bindingErr == nil {
		profileKey = binding.ProfileKey
		bindings = domain.CopyStringMap(binding.Bindings)
	} else if !errors.Is(bindingErr, repo.ErrNotFound) {
		return runDistributionPackageContext{}, bindingErr
	}
	plan, err := runAutonomousLifecyclePlan(distribution, instance, plugin, profileKey, bindings)
	if err != nil {
		return runDistributionPackageContext{}, err
	}
	seed, err := encodeRunWorkspaceSeed(plugin.LifecycleAssets, plan)
	if err != nil {
		return runDistributionPackageContext{}, err
	}
	return runDistributionPackageContext{profileKey: profileKey, workspaceSeed: seed, autonomousLifecycle: plan}, nil
}

func runAutonomousLifecyclePlan(distribution domain.RunDistribution, instance domain.ServerInstance, plugin domain.GamePlugin, profileKey string, bindings map[string]string) (*domain.RunAutonomousLifecyclePlan, error) {
	plan := &domain.RunAutonomousLifecyclePlan{
		SchemaVersion:      "1",
		ServerInstanceID:   instance.ID,
		PluginID:           plugin.ID,
		PluginVersion:      plugin.Version,
		RunEndpointID:      distribution.RunEndpointID,
		ProfileKey:         profileKey,
		TargetOS:           distribution.TargetOS,
		TargetArch:         distribution.TargetArch,
		TargetRelease:      distribution.ID,
		DeploymentRevision: instance.Deployment.Revision,
		RuntimeBindings:    domain.CopyStringMap(bindings),
		Deployment:         autonomousDeploymentFromDefinition(instance.Deployment, profileKey, bindings),
	}
	profile, hasProfile := runtimeLifecycleProfileForKey(plugin.RuntimeProfiles, profileKey)
	for _, action := range []domain.ServerLifecycleAction{domain.ServerLifecycleActionCreate, domain.ServerLifecycleActionStart, domain.ServerLifecycleActionStop, domain.ServerLifecycleActionStatus} {
		entry := autonomousLifecycleAction(plugin, profile, hasProfile, action)
		if entry.TargetKey == "" {
			continue
		}
		plan.Actions = append(plan.Actions, entry)
	}
	bootstrap := autonomousLifecycleAction(plugin, profile, hasProfile, autonomousBootstrapLifecycleAction(plugin, profile, hasProfile))
	if bootstrap.TargetKey != "" {
		plan.Bootstrap = &bootstrap
	}
	for _, probe := range plugin.RuntimeProfiles.DependencyProbes {
		if runtimePlatformsContain(probe.Platforms, distribution.TargetOS) {
			plan.DependencyProbes = append(plan.DependencyProbes, autonomousDependencyProbe(probe))
		}
	}
	for _, installPlan := range plugin.RuntimeProfiles.InstallPlans {
		if runtimePlatformsContain(installPlan.Platforms, distribution.TargetOS) {
			plan.InstallPlans = append(plan.InstallPlans, autonomousInstallPlan(installPlan))
		}
	}
	for _, source := range plugin.RuntimeProfiles.LogSources {
		if autonomousRunLogSourceKind(source.Kind) {
			plan.LogSources = append(plan.LogSources, autonomousLogSource(source))
		}
	}
	if hasProfile && len(profile.DLLExtensionRefs) > 0 {
		endpoint := domain.RunEndpoint{ID: distribution.RunEndpointID, Platform: distribution.TargetOS, Architecture: distribution.TargetArch}
		extensions, err := lifecycleDLLExtensionPlans(plugin.RuntimeProfiles, profile, endpoint)
		if err != nil {
			return nil, err
		}
		for _, extension := range extensions {
			plan.DLLExtensions = append(plan.DLLExtensions, autonomousDLLExtension(extension))
		}
	}
	return domain.CopyRunAutonomousLifecyclePlanPtr(plan), nil
}

func autonomousLifecycleAction(plugin domain.GamePlugin, profile domain.RuntimeLifecycleProfile, hasProfile bool, action domain.ServerLifecycleAction) domain.RunAutonomousLifecycleAction {
	targetKey := ""
	if hasProfile {
		targetKey = runtimeProfileActionRef(profile.ActionRefs, action)
	}
	if targetKey == "" {
		targetKey = lifecycleActionRef(plugin, action)
	}
	if targetKey == "" {
		return domain.RunAutonomousLifecycleAction{}
	}
	return domain.RunAutonomousLifecycleAction{Action: action, Operation: lifecycleExecutionOperation(action), Capability: domain.LifecycleCapabilityForAction(action), TargetKey: targetKey}
}

func autonomousBootstrapLifecycleAction(plugin domain.GamePlugin, profile domain.RuntimeLifecycleProfile, hasProfile bool) domain.ServerLifecycleAction {
	if autonomousLifecycleAction(plugin, profile, hasProfile, domain.ServerLifecycleActionStart).TargetKey != "" {
		return domain.ServerLifecycleActionStart
	}
	return domain.ServerLifecycleActionCreate
}

func autonomousDependencyProbe(probe domain.RuntimeDependencyProbe) domain.RunAutonomousDependencyProbe {
	return domain.RunAutonomousDependencyProbe{Key: probe.Key, Kind: probe.Kind, TargetKey: probe.TargetKey, Required: probe.Required, MinimumVersion: probe.MinimumVersion, Platforms: domain.CopyStringSlice(probe.Platforms)}
}

func autonomousInstallPlan(plan domain.RuntimeInstallPlan) domain.RunAutonomousInstallPlan {
	steps := make([]domain.RunAutonomousInstallStep, len(plan.Steps))
	for i, step := range plan.Steps {
		steps[i] = domain.RunAutonomousInstallStep{Type: step.Type, TargetKey: step.TargetKey, PackageManager: step.PackageManager, PackageName: step.PackageName, Version: step.Version, DownloadRef: step.DownloadRef, Checksum: step.Checksum}
	}
	return domain.RunAutonomousInstallPlan{Key: plan.Key, Title: plan.Title, Platforms: domain.CopyStringSlice(plan.Platforms), Steps: steps}
}

func autonomousLogSource(source domain.RuntimeLogSource) domain.RunAutonomousLogSource {
	return domain.RunAutonomousLogSource{Key: source.Key, Kind: source.Kind, TargetKey: source.TargetKey, StreamKey: source.StreamKey, CursorKind: source.CursorKind, RetentionDays: source.RetentionDays}
}

func autonomousRunLogSourceKind(kind string) bool {
	return kind == "process.stdout" || kind == "process.stderr" || kind == "file.tail"
}

func autonomousDLLExtension(extension domain.RuntimeDLLExtensionPlan) domain.RunAutonomousDLLExtension {
	return domain.RunAutonomousDLLExtension{Key: extension.Key, Version: extension.Version, ReleaseURL: extension.ReleaseURL, Checksum: extension.Checksum, SizeBytes: extension.SizeBytes, TargetKey: extension.TargetKey, ModKey: extension.ModKey, DLLRef: extension.DLLRef, SCUMExecutableChecksum: extension.SCUMExecutableChecksum, UE4SSABI: extension.UE4SSABI, RCONPort: extension.RCONPort}
}

func autonomousDeploymentFromDefinition(definition domain.ServerDeploymentDefinition, profileKey string, bindings map[string]string) *domain.RunAutonomousDeployment {
	if definition.Mode == "" {
		return nil
	}
	copy := domain.CopyServerDeploymentDefinition(definition)
	if copy.ProfileKey == "" {
		copy.ProfileKey = profileKey
	}
	if len(copy.RuntimeBindings) == 0 {
		copy.RuntimeBindings = domain.CopyStringMap(bindings)
	}
	return &domain.RunAutonomousDeployment{SchemaVersion: "1", Mode: copy.Mode, ProfileKey: copy.ProfileKey, RuntimeBindings: copy.RuntimeBindings, CreateInputs: copy.CreateInputs, ServerRoot: copy.ServerRoot, WorkingDirectory: copy.WorkingDirectory, InstallCommand: copy.InstallCommand, StartCommand: copy.StartCommand, StopCommand: copy.StopCommand, StatusCommand: copy.StatusCommand, Shell: copy.Shell, Revision: copy.Revision}
}

func encodeRunWorkspaceSeed(files []domain.PluginAssetFile, plan *domain.RunAutonomousLifecyclePlan) (string, error) {
	seedFiles := append([]domain.PluginAssetFile(nil), files...)
	if plan != nil {
		payload, err := json.Marshal(plan)
		if err != nil {
			return "", err
		}
		seedFiles = append(seedFiles, domain.PluginAssetFile{Path: ".platform/autonomous-lifecycle-plan.json", Content: string(payload), Mode: 0o600})
	}
	return encodePluginWorkspaceSeed(seedFiles)
}

func encodePluginWorkspaceSeed(files []domain.PluginAssetFile) (string, error) {
	if len(files) == 0 {
		return "", nil
	}
	type seedFile struct {
		Path    string `json:"path"`
		Content string `json:"content"`
		Mode    int    `json:"mode,omitempty"`
	}
	seed := make([]seedFile, len(files))
	for i, file := range files {
		seed[i] = seedFile{Path: file.Path, Content: file.Content, Mode: file.Mode}
	}
	body, err := json.Marshal(seed)
	if err != nil {
		return "", err
	}
	return base64.StdEncoding.EncodeToString(body), nil
}

// storeDistributionBuildArtifact records the built package under job ownership
// so the existing ArtifactOwnerKindJob scope assertions keep guarding it.
func (svc *CoreService) storeDistributionBuildArtifact(artifactID string, jobID string, payload []byte) error {
	if strings.TrimSpace(artifactID) == "" {
		return validationError("distribution build artifact id is required")
	}
	if len(payload) == 0 {
		return validationError("distribution build produced no package bytes")
	}
	stamp := svc.now()
	svc.artifactMu.Lock()
	defer svc.artifactMu.Unlock()

	artifact, err := svc.store.Artifacts().Get(artifactID)
	if err != nil && !errors.Is(err, repo.ErrNotFound) {
		return err
	}
	create := errors.Is(err, repo.ErrNotFound)
	if create {
		artifact = domain.Artifact{ID: artifactID, OwnerKind: domain.ArtifactOwnerKindJob, OwnerID: jobID, CreatedAt: stamp}
	}
	if artifact.OwnerKind != domain.ArtifactOwnerKindJob || artifact.OwnerID != jobID {
		return validationError("distribution build artifact is outside the job scope")
	}
	artifact.SizeBytes = int64(len(payload))
	artifact.Checksum = validator.BytesChecksum(payload)
	artifact.State = domain.ArtifactStateAvailable
	artifact.UpdatedAt = stamp
	if err := validator.ValidateArtifact(artifact); err != nil {
		return err
	}
	if err := svc.artifactStore.PutPayload(artifact.ID, payload); err != nil {
		return err
	}
	if create {
		if err := svc.store.Artifacts().Create(artifact); err != nil {
			return err
		}
	} else if err := svc.store.Artifacts().Update(artifact); err != nil {
		return err
	}
	svc.artifactPayloads[artifact.ID] = domain.CopyBytes(payload)
	return nil
}

func (svc *CoreService) markDistributionBuildRunning(job *domain.Job) error {
	stamp := svc.now()
	svc.jobMu.Lock()
	defer svc.jobMu.Unlock()

	current, err := svc.store.Jobs().Get(job.ID)
	if err != nil {
		return err
	}
	if isTerminalJobState(current.State) {
		*job = current
		return nil
	}
	current.State = domain.JobStateRunning
	current.Attempt = maxInt(current.Attempt, 1)
	current.Progress = domain.JobProgress{Percent: 5, Phase: current.Progress.Phase, Message: "env_check: platform builder started"}
	current.UpdatedAt = stamp
	if err := svc.updateScheduledJob(current); err != nil {
		return err
	}
	if err := svc.projectDistributionBuildProgress(current, stamp); err != nil {
		return err
	}
	*job = current
	return nil
}

func (svc *CoreService) updateDistributionBuildProgress(job domain.Job, progress DistributionBuildProgress) error {
	stamp := svc.now()
	svc.jobMu.Lock()
	defer svc.jobMu.Unlock()

	current, err := svc.store.Jobs().Get(job.ID)
	if err != nil {
		return err
	}
	if isTerminalJobState(current.State) {
		return nil
	}
	if progress.Percent < current.Progress.Percent {
		progress.Percent = current.Progress.Percent
	}
	current.Progress = domain.JobProgress{Percent: progress.Percent, Phase: current.Progress.Phase, Message: strings.TrimSpace(progress.Message)}
	current.UpdatedAt = stamp
	if err := svc.updateScheduledJob(current); err != nil {
		return err
	}
	return svc.projectDistributionBuildProgress(current, stamp)
}

func (svc *CoreService) succeedDistributionBuildJob(job domain.Job, artifactID string) error {
	stamp := svc.now()
	svc.jobMu.Lock()
	current, err := svc.store.Jobs().Get(job.ID)
	if err != nil {
		svc.jobMu.Unlock()
		return err
	}
	if isTerminalJobState(current.State) {
		if current.State == domain.JobStateSucceeded && current.ResultRef == "artifact://"+artifactID {
			svc.jobMu.Unlock()
			return nil
		}
		svc.jobMu.Unlock()
		_ = svc.expireDistributionArtifact(artifactID)
		return nil
	}
	current.State = domain.JobStateSucceeded
	current.ResultRef = "artifact://" + artifactID
	current.Progress = domain.JobProgress{Percent: 100, Phase: current.Progress.Phase, Message: "platform builder completed"}
	current.LeaseTokenHash = ""
	current.TerminalAt = stamp
	current.TerminalFingerprint = "platform-builder:succeeded:" + artifactID
	current.UpdatedAt = stamp
	if err := svc.updateScheduledJob(current); err != nil {
		svc.jobMu.Unlock()
		return err
	}
	svc.jobMu.Unlock()
	return svc.projectDistributionBuildResult(current, stamp)
}

func (svc *CoreService) failDistributionBuildJob(job domain.Job, reason string) error {
	stamp := svc.now()
	if strings.TrimSpace(reason) == "" {
		reason = "platform builder failed"
	}
	svc.jobMu.Lock()
	current, err := svc.store.Jobs().Get(job.ID)
	if err != nil {
		svc.jobMu.Unlock()
		return err
	}
	if isTerminalJobState(current.State) {
		svc.jobMu.Unlock()
		return nil
	}
	current.State = domain.JobStateFailed
	current.Progress = domain.JobProgress{Percent: current.Progress.Percent, Phase: current.Progress.Phase, Message: reason}
	current.LeaseTokenHash = ""
	current.TerminalAt = stamp
	current.TerminalFingerprint = "platform-builder:failed:" + reason
	current.UpdatedAt = stamp
	if err := svc.updateScheduledJob(current); err != nil {
		svc.jobMu.Unlock()
		return err
	}
	svc.jobMu.Unlock()
	if err := svc.projectDistributionBuildResult(current, stamp); err != nil && !errors.Is(err, repo.ErrNotFound) {
		return err
	}
	return validationError(reason)
}