fix(agent): start devcontainers through agentcontainers package (#18471)

Fixes https://github.com/coder/internal/issues/706

Context for the implementation here
https://github.com/coder/internal/issues/706#issuecomment-2990490282

Synchronously starts dev containers defined in terraform with our
`DevcontainerCLI` abstraction, instead of piggybacking off of our
`agentscripts` package. This gives us more control over logs, instead of
being reliant on packages which may or may not exist in the
user-provided image.

agent/agentcontainers/api.go

@@ -207,6 +207,10 @@ func WithDevcontainers(devcontainers []codersdk.WorkspaceAgentDevcontainer, scri
 		api.devcontainerNames = make(map[string]bool, len(devcontainers))
 		api.devcontainerLogSourceIDs = make(map[string]uuid.UUID)
 		for _, dc := range devcontainers {
+			if dc.Status == "" {
+				dc.Status = codersdk.WorkspaceAgentDevcontainerStatusStarting
+			}
+
 			api.knownDevcontainers[dc.WorkspaceFolder] = dc
 			api.devcontainerNames[dc.Name] = true
 			for _, script := range scripts {
@@ -265,8 +269,6 @@ func NewAPI(logger slog.Logger, options ...Option) *API {
 	api := &API{
 		ctx:                         ctx,
 		cancel:                      cancel,
-		watcherDone:                 make(chan struct{}),
-		updaterDone:                 make(chan struct{}),
 		initialUpdateDone:           make(chan struct{}),
 		updateTrigger:               make(chan chan error),
 		updateInterval:              defaultUpdateInterval,
@@ -315,10 +317,28 @@ func NewAPI(logger slog.Logger, options ...Option) *API {
 		api.subAgentClient.Store(&c)
 	}
 
+	return api
+}
+
+// Init applies a final set of options to the API and then
+// begins the watcherLoop and updaterLoop. This function
+// must only be called once.
+func (api *API) Init(opts ...Option) {
+	api.mu.Lock()
+	defer api.mu.Unlock()
+	if api.closed {
+		return
+	}
+
+	for _, opt := range opts {
+		opt(api)
+	}
+
+	api.watcherDone = make(chan struct{})
+	api.updaterDone = make(chan struct{})
+
 	go api.watcherLoop()
 	go api.updaterLoop()
-
-	return api
 }
 
 func (api *API) watcherLoop() {
@@ -909,8 +929,9 @@ func (api *API) handleDevcontainerRecreate(w http.ResponseWriter, r *http.Reques
 	dc.Status = codersdk.WorkspaceAgentDevcontainerStatusStarting
 	dc.Container = nil
 	api.knownDevcontainers[dc.WorkspaceFolder] = dc
-	api.asyncWg.Add(1)
-	go api.recreateDevcontainer(dc, configPath)
+	go func() {
+		_ = api.CreateDevcontainer(dc.WorkspaceFolder, configPath, WithRemoveExistingContainer())
+	}()
 
 	api.mu.Unlock()
 
@@ -920,15 +941,29 @@ func (api *API) handleDevcontainerRecreate(w http.ResponseWriter, r *http.Reques
 	})
 }
 
-// recreateDevcontainer should run in its own goroutine and is responsible for
+// createDevcontainer should run in its own goroutine and is responsible for
 // recreating a devcontainer based on the provided devcontainer configuration.
 // It updates the devcontainer status and logs the process. The configPath is
 // passed as a parameter for the odd chance that the container being recreated
 // has a different config file than the one stored in the devcontainer state.
 // The devcontainer state must be set to starting and the asyncWg must be
 // incremented before calling this function.
-func (api *API) recreateDevcontainer(dc codersdk.WorkspaceAgentDevcontainer, configPath string) {
+func (api *API) CreateDevcontainer(workspaceFolder, configPath string, opts ...DevcontainerCLIUpOptions) error {
+	api.mu.Lock()
+	if api.closed {
+		api.mu.Unlock()
+		return nil
+	}
+
+	dc, found := api.knownDevcontainers[workspaceFolder]
+	if !found {
+		api.mu.Unlock()
+		return xerrors.Errorf("devcontainer not found")
+	}
+
+	api.asyncWg.Add(1)
 	defer api.asyncWg.Done()
+	api.mu.Unlock()
 
 	var (
 		err    error
@@ -969,12 +1004,15 @@ func (api *API) recreateDevcontainer(dc codersdk.WorkspaceAgentDevcontainer, con
 
 	logger.Debug(ctx, "starting devcontainer recreation")
 
-	_, err = api.dccli.Up(ctx, dc.WorkspaceFolder, configPath, WithUpOutput(infoW, errW), WithRemoveExistingContainer())
+	upOptions := []DevcontainerCLIUpOptions{WithUpOutput(infoW, errW)}
+	upOptions = append(upOptions, opts...)
+
+	_, err = api.dccli.Up(ctx, dc.WorkspaceFolder, configPath, upOptions...)
 	if err != nil {
 		// No need to log if the API is closing (context canceled), as this
 		// is expected behavior when the API is shutting down.
 		if !errors.Is(err, context.Canceled) {
-			logger.Error(ctx, "devcontainer recreation failed", slog.Error(err))
+			logger.Error(ctx, "devcontainer creation failed", slog.Error(err))
 		}
 
 		api.mu.Lock()
@@ -983,10 +1021,11 @@ func (api *API) recreateDevcontainer(dc codersdk.WorkspaceAgentDevcontainer, con
 		api.knownDevcontainers[dc.WorkspaceFolder] = dc
 		api.recreateErrorTimes[dc.WorkspaceFolder] = api.clock.Now("agentcontainers", "recreate", "errorTimes")
 		api.mu.Unlock()
-		return
+
+		return xerrors.Errorf("start devcontainer: %w", err)
 	}
 
-	logger.Info(ctx, "devcontainer recreated successfully")
+	logger.Info(ctx, "devcontainer created successfully")
 
 	api.mu.Lock()
 	dc = api.knownDevcontainers[dc.WorkspaceFolder]
@@ -1009,8 +1048,11 @@ func (api *API) recreateDevcontainer(dc codersdk.WorkspaceAgentDevcontainer, con
 	// Ensure an immediate refresh to accurately reflect the
 	// devcontainer state after recreation.
 	if err := api.RefreshContainers(ctx); err != nil {
-		logger.Error(ctx, "failed to trigger immediate refresh after devcontainer recreation", slog.Error(err))
+		logger.Error(ctx, "failed to trigger immediate refresh after devcontainer creation", slog.Error(err))
+		return xerrors.Errorf("refresh containers: %w", err)
 	}
+
+	return nil
 }
 
 // markDevcontainerDirty finds the devcontainer with the given config file path
@@ -1609,8 +1651,12 @@ func (api *API) Close() error {
 	err := api.watcher.Close()
 
 	// Wait for loops to finish.
-	<-api.watcherDone
-	<-api.updaterDone
+	if api.watcherDone != nil {
+		<-api.watcherDone
+	}
+	if api.updaterDone != nil {
+		<-api.updaterDone
+	}
 
 	// Wait for all async tasks to complete.
 	api.asyncWg.Wait()