upaas/internal/service/deploy/deploy.go

// Package deploy provides deployment services.
package deploy

import (
	"context"
	"database/sql"
	"errors"
	"fmt"
	"log/slog"
	"os"
	"path/filepath"
	"time"

	"go.uber.org/fx"

	"git.eeqj.de/sneak/upaas/internal/config"
	"git.eeqj.de/sneak/upaas/internal/database"
	"git.eeqj.de/sneak/upaas/internal/docker"
	"git.eeqj.de/sneak/upaas/internal/logger"
	"git.eeqj.de/sneak/upaas/internal/models"
	"git.eeqj.de/sneak/upaas/internal/service/notify"
)

// Time constants.
const (
	healthCheckDelaySeconds = 60
	// upaasLabelCount is the number of upaas-specific labels added to containers.
	upaasLabelCount = 1
	// buildsDirPermissions is the permission mode for the builds directory.
	buildsDirPermissions = 0o750
)

// Sentinel errors for deployment failures.
var (
	// ErrContainerUnhealthy indicates the container failed health check.
	ErrContainerUnhealthy = errors.New("container unhealthy after 60 seconds")
)

// ServiceParams contains dependencies for Service.
type ServiceParams struct {
	fx.In

	Logger   *logger.Logger
	Config   *config.Config
	Database *database.Database
	Docker   *docker.Client
	Notify   *notify.Service
}

// Service provides deployment functionality.
type Service struct {
	log    *slog.Logger
	db     *database.Database
	docker *docker.Client
	notify *notify.Service
	config *config.Config
	params *ServiceParams
}

// New creates a new deploy Service.
func New(_ fx.Lifecycle, params ServiceParams) (*Service, error) {
	return &Service{
		log:    params.Logger.Get(),
		db:     params.Database,
		docker: params.Docker,
		notify: params.Notify,
		config: params.Config,
		params: &params,
	}, nil
}

// GetBuildDir returns the build directory path for an app.
func (svc *Service) GetBuildDir(appID string) string {
	return filepath.Join(svc.config.DataDir, "builds", appID)
}

// Deploy deploys an app.
func (svc *Service) Deploy(
	ctx context.Context,
	app *models.App,
	webhookEventID *int64,
) error {
	deployment, err := svc.createDeploymentRecord(ctx, app, webhookEventID)
	if err != nil {
		return err
	}

	err = svc.updateAppStatusBuilding(ctx, app)
	if err != nil {
		return err
	}

	svc.notify.NotifyBuildStart(ctx, app, deployment)

	imageID, err := svc.buildImage(ctx, app, deployment)
	if err != nil {
		return err
	}

	svc.notify.NotifyBuildSuccess(ctx, app, deployment)

	err = svc.updateDeploymentDeploying(ctx, deployment)
	if err != nil {
		return err
	}

	svc.removeOldContainer(ctx, app, deployment)

	_, err = svc.createAndStartContainer(ctx, app, deployment, imageID)
	if err != nil {
		return err
	}

	err = svc.updateAppRunning(ctx, app, imageID)
	if err != nil {
		return err
	}

	// Use context.WithoutCancel to ensure health check completes even if
	// the parent context is cancelled (e.g., HTTP request ends).
	go svc.checkHealthAfterDelay(context.WithoutCancel(ctx), app, deployment)

	return nil
}

func (svc *Service) createDeploymentRecord(
	ctx context.Context,
	app *models.App,
	webhookEventID *int64,
) (*models.Deployment, error) {
	deployment := models.NewDeployment(svc.db)
	deployment.AppID = app.ID

	if webhookEventID != nil {
		deployment.WebhookEventID = sql.NullInt64{
			Int64: *webhookEventID,
			Valid: true,
		}
	}

	deployment.Status = models.DeploymentStatusBuilding

	saveErr := deployment.Save(ctx)
	if saveErr != nil {
		return nil, fmt.Errorf("failed to create deployment: %w", saveErr)
	}

	return deployment, nil
}

func (svc *Service) updateAppStatusBuilding(
	ctx context.Context,
	app *models.App,
) error {
	app.Status = models.AppStatusBuilding

	saveErr := app.Save(ctx)
	if saveErr != nil {
		return fmt.Errorf("failed to update app status: %w", saveErr)
	}

	return nil
}

func (svc *Service) buildImage(
	ctx context.Context,
	app *models.App,
	deployment *models.Deployment,
) (string, error) {
	tempDir, cleanup, err := svc.cloneRepository(ctx, app, deployment)
	if err != nil {
		return "", err
	}

	defer cleanup()

	imageTag := "upaas/" + app.Name + ":latest"

	imageID, err := svc.docker.BuildImage(ctx, docker.BuildImageOptions{
		ContextDir:     tempDir,
		DockerfilePath: app.DockerfilePath,
		Tags:           []string{imageTag},
	})
	if err != nil {
		svc.notify.NotifyBuildFailed(ctx, app, deployment, err)
		svc.failDeployment(
			ctx,
			app,
			deployment,
			fmt.Errorf("failed to build image: %w", err),
		)

		return "", fmt.Errorf("failed to build image: %w", err)
	}

	deployment.ImageID = sql.NullString{String: imageID, Valid: true}
	_ = deployment.AppendLog(ctx, "Image built: "+imageID)

	return imageID, nil
}

func (svc *Service) cloneRepository(
	ctx context.Context,
	app *models.App,
	deployment *models.Deployment,
) (string, func(), error) {
	// Use a subdirectory of DataDir for builds since it's mounted from the host
	// and accessible to Docker for bind mounts (unlike /tmp inside the container).
	// Structure: builds/<appname>/<deployment-id>-<random>
	appBuildsDir := filepath.Join(svc.config.DataDir, "builds", app.Name)

	err := os.MkdirAll(appBuildsDir, buildsDirPermissions)
	if err != nil {
		svc.failDeployment(ctx, app, deployment, fmt.Errorf("failed to create builds dir: %w", err))

		return "", nil, fmt.Errorf("failed to create builds dir: %w", err)
	}

	tempDir, err := os.MkdirTemp(appBuildsDir, fmt.Sprintf("%d-*", deployment.ID))
	if err != nil {
		svc.failDeployment(ctx, app, deployment, fmt.Errorf("failed to create temp dir: %w", err))

		return "", nil, fmt.Errorf("failed to create temp dir: %w", err)
	}

	cleanup := func() { _ = os.RemoveAll(tempDir) }

	cloneErr := svc.docker.CloneRepo(ctx, app.RepoURL, app.Branch, app.SSHPrivateKey, tempDir)
	if cloneErr != nil {
		cleanup()
		svc.failDeployment(ctx, app, deployment, fmt.Errorf("failed to clone repo: %w", cloneErr))

		return "", nil, fmt.Errorf("failed to clone repo: %w", cloneErr)
	}

	_ = deployment.AppendLog(ctx, "Repository cloned successfully")

	return tempDir, cleanup, nil
}

func (svc *Service) updateDeploymentDeploying(
	ctx context.Context,
	deployment *models.Deployment,
) error {
	deployment.Status = models.DeploymentStatusDeploying

	saveErr := deployment.Save(ctx)
	if saveErr != nil {
		return fmt.Errorf("failed to update deployment status: %w", saveErr)
	}

	return nil
}

func (svc *Service) removeOldContainer(
	ctx context.Context,
	app *models.App,
	deployment *models.Deployment,
) {
	containerInfo, err := svc.docker.FindContainerByAppID(ctx, app.ID)
	if err != nil || containerInfo == nil {
		return
	}

	svc.log.Info("removing old container", "id", containerInfo.ID)

	removeErr := svc.docker.RemoveContainer(ctx, containerInfo.ID, true)
	if removeErr != nil {
		svc.log.Warn("failed to remove old container", "error", removeErr)
	}

	_ = deployment.AppendLog(ctx, "Old container removed")
}

func (svc *Service) createAndStartContainer(
	ctx context.Context,
	app *models.App,
	deployment *models.Deployment,
	imageID string,
) (string, error) {
	containerOpts, err := svc.buildContainerOptions(ctx, app, imageID)
	if err != nil {
		svc.failDeployment(ctx, app, deployment, err)

		return "", err
	}

	containerID, err := svc.docker.CreateContainer(ctx, containerOpts)
	if err != nil {
		svc.notify.NotifyDeployFailed(ctx, app, deployment, err)
		svc.failDeployment(
			ctx,
			app,
			deployment,
			fmt.Errorf("failed to create container: %w", err),
		)

		return "", fmt.Errorf("failed to create container: %w", err)
	}

	deployment.ContainerID = sql.NullString{String: containerID, Valid: true}
	_ = deployment.AppendLog(ctx, "Container created: "+containerID)

	startErr := svc.docker.StartContainer(ctx, containerID)
	if startErr != nil {
		svc.notify.NotifyDeployFailed(ctx, app, deployment, startErr)
		svc.failDeployment(
			ctx,
			app,
			deployment,
			fmt.Errorf("failed to start container: %w", startErr),
		)

		return "", fmt.Errorf("failed to start container: %w", startErr)
	}

	_ = deployment.AppendLog(ctx, "Container started")

	return containerID, nil
}

func (svc *Service) buildContainerOptions(
	ctx context.Context,
	app *models.App,
	_ string,
) (docker.CreateContainerOptions, error) {
	envVars, err := app.GetEnvVars(ctx)
	if err != nil {
		return docker.CreateContainerOptions{}, fmt.Errorf("failed to get env vars: %w", err)
	}

	labels, err := app.GetLabels(ctx)
	if err != nil {
		return docker.CreateContainerOptions{}, fmt.Errorf("failed to get labels: %w", err)
	}

	volumes, err := app.GetVolumes(ctx)
	if err != nil {
		return docker.CreateContainerOptions{}, fmt.Errorf("failed to get volumes: %w", err)
	}

	envMap := make(map[string]string, len(envVars))
	for _, envVar := range envVars {
		envMap[envVar.Key] = envVar.Value
	}

	network := ""
	if app.DockerNetwork.Valid {
		network = app.DockerNetwork.String
	}

	return docker.CreateContainerOptions{
		Name:    "upaas-" + app.Name,
		Image:   "upaas/" + app.Name + ":latest",
		Env:     envMap,
		Labels:  buildLabelMap(app, labels),
		Volumes: buildVolumeMounts(volumes),
		Network: network,
	}, nil
}

func buildLabelMap(app *models.App, labels []*models.Label) map[string]string {
	labelMap := make(map[string]string, len(labels)+upaasLabelCount)
	for _, label := range labels {
		labelMap[label.Key] = label.Value
	}

	// Add the upaas.id label to identify this container
	labelMap[docker.LabelUpaasID] = app.ID

	return labelMap
}

func buildVolumeMounts(volumes []*models.Volume) []docker.VolumeMount {
	mounts := make([]docker.VolumeMount, 0, len(volumes))
	for _, vol := range volumes {
		mounts = append(mounts, docker.VolumeMount{
			HostPath:      vol.HostPath,
			ContainerPath: vol.ContainerPath,
			ReadOnly:      vol.ReadOnly,
		})
	}

	return mounts
}

func (svc *Service) updateAppRunning(
	ctx context.Context,
	app *models.App,
	imageID string,
) error {
	app.ImageID = sql.NullString{String: imageID, Valid: true}
	app.Status = models.AppStatusRunning

	saveErr := app.Save(ctx)
	if saveErr != nil {
		return fmt.Errorf("failed to update app: %w", saveErr)
	}

	return nil
}

func (svc *Service) checkHealthAfterDelay(
	ctx context.Context,
	app *models.App,
	deployment *models.Deployment,
) {
	svc.log.Info(
		"waiting 60 seconds to check container health",
		"app", app.Name,
	)
	time.Sleep(healthCheckDelaySeconds * time.Second)

	// Reload app to get current state
	reloadedApp, err := models.FindApp(ctx, svc.db, app.ID)
	if err != nil || reloadedApp == nil {
		svc.log.Error("failed to reload app for health check", "error", err)

		return
	}

	containerInfo, containerErr := svc.docker.FindContainerByAppID(ctx, app.ID)
	if containerErr != nil || containerInfo == nil {
		return
	}

	healthy, err := svc.docker.IsContainerHealthy(ctx, containerInfo.ID)
	if err != nil {
		svc.log.Error("failed to check container health", "error", err)
		svc.notify.NotifyDeployFailed(ctx, reloadedApp, deployment, err)
		_ = deployment.MarkFinished(ctx, models.DeploymentStatusFailed)

		return
	}

	if healthy {
		svc.log.Info("container healthy after 60 seconds", "app", reloadedApp.Name)
		svc.notify.NotifyDeploySuccess(ctx, reloadedApp, deployment)
		_ = deployment.MarkFinished(ctx, models.DeploymentStatusSuccess)
	} else {
		svc.log.Warn(
			"container unhealthy after 60 seconds",
			"app", reloadedApp.Name,
		)
		svc.notify.NotifyDeployFailed(ctx, reloadedApp, deployment, ErrContainerUnhealthy)
		_ = deployment.MarkFinished(ctx, models.DeploymentStatusFailed)
		reloadedApp.Status = models.AppStatusError
		_ = reloadedApp.Save(ctx)
	}
}

func (svc *Service) failDeployment(
	ctx context.Context,
	app *models.App,
	deployment *models.Deployment,
	deployErr error,
) {
	svc.log.Error("deployment failed", "app", app.Name, "error", deployErr)
	_ = deployment.AppendLog(ctx, "ERROR: "+deployErr.Error())
	_ = deployment.MarkFinished(ctx, models.DeploymentStatusFailed)
	app.Status = models.AppStatusError
	_ = app.Save(ctx)
}