Fix blue-green deployment for Kubernetes apps

Author: akclace

docs/content/docs/Container/Config.md

@@ -30,7 +30,6 @@ container.show_logs_for_failure = true
 
 # Kubernetes related settings
 kubernetes.default_volume_size = "10Gi"
-kubernetes.strict_version_check = true
 kubernetes.scaling_threshold_cpu = 80
 ```
 

docs/content/docs/Container/Kubernetes.md

@@ -105,11 +105,4 @@ OpenRun is installed as a Kubernetes Deployment, with a Service for routing API
 
 All Kubernetes resources are created lazily, on the first API call to the app. If app is running version 1, and a code/config change is done which updates it to version 2, the deployment update will happen on the next API call to the app. The API call is blocked, the container image is rebuild if required and the deployment is updated using Server Side Apply API calls.
 
-By default, OpenRun will wait to ensure that the app is running pods with only the new version, before processing further API calls. This behavior can be relaxed by setting
-
-```toml {filename="openrun.toml"}
-[app_config]
-kubernetes.strict_version_check = false
-```
-
-in which case OpenRun just makes sure that the new version started successfully,
+OpenRun waits until Kubernetes reports the expected new version rollout is complete before processing further API calls.

internal/app/app.go

@@ -176,7 +176,7 @@ func NewApp(sourceFS *appfs.SourceFs, workFS *appfs.WorkFs, logger *types.Logger
 func (a *App) Initialize(ctx context.Context, dryRun types.DryRun) error {
 	var reloaded bool
 	var err error
-	if reloaded, err = a.Reload(ctx, false, true, dryRun, true); err != nil {
+	if reloaded, err = a.Reload(ctx, false, true, dryRun, ReloadOptions{ReloadContainer: true, Verify: false}); err != nil {
 		return err
 	}
 
@@ -235,7 +235,20 @@ func (a *App) ResetFS() {
 	a.sourceFS.Reset()
 }
 
-func (a *App) Reload(ctx context.Context, force, immediate bool, dryRun types.DryRun, reloadContainer bool) (bool, error) {
+// ReloadOptions controls how App.Reload handles the prod container.
+type ReloadOptions struct {
+	// ReloadContainer, when true, (re)loads the prod container (rebuild/restart
+	// as needed). It is true for reload and initialize operations; only the
+	// metadata-only paths leave it false. Image-spec apps always reload.
+	ReloadContainer bool
+	// Verify indicates the caller wants a verified, rollback-capable update.
+	// For in-place container managers (Kubernetes) this makes a missing
+	// rollback snapshot a fatal error rather than proceeding with an
+	// irreversible in-place update.
+	Verify bool
+}
+
+func (a *App) Reload(ctx context.Context, force, immediate bool, dryRun types.DryRun, opts ReloadOptions) (bool, error) {
 	requestTime := time.Now()
 
 	a.initMutex.Lock()
@@ -301,7 +314,7 @@ func (a *App) Reload(ctx context.Context, force, immediate bool, dryRun types.Dr
 	}
 
 	// Load Starlark config, AppConfig is updated with the settings contents
-	if err = a.loadStarlarkConfig(ctx, dryRun, reloadContainer); err != nil {
+	if err = a.loadStarlarkConfig(ctx, dryRun, opts); err != nil {
 		return false, fmt.Errorf("error during initial setup: %w", err)
 	}
 	a.Metadata.Name = a.Name
@@ -775,7 +788,7 @@ func (a *App) startWatcher() error {
 
 					inReload.Store(true)
 					defer inReload.Store(false)
-					_, err := a.Reload(context.Background(), true, false, types.DryRun(false), true)
+					_, err := a.Reload(context.Background(), true, false, types.DryRun(false), ReloadOptions{ReloadContainer: true, Verify: false})
 					a.reloadError = err
 					if err != nil {
 						a.Error().Err(err).Msg("Error reloading app")