fix(runtime): retry incomplete materialization

Author: MarcStdt

apps/druid/adapters/cli/client/dev.go

@@ -132,7 +132,7 @@ func runDevServer() error {
 	}
 	auth := devAuth{runtimeID: devRuntimeID, ownerID: devOwnerID}
 	if devAuthJWKSURL != "" {
-		auth.user, err = coreservices.NewAuthorizer(devAuthJWKSURL, "")
+		auth.user, err = coreservices.NewAuthorizer([]string{devAuthJWKSURL}, "")
 		if err != nil {
 			return err
 		}

apps/druid/adapters/cli/daemon.go

@@ -148,7 +148,7 @@ func runRuntimeDaemon() error {
 		return err
 	}
 
-	authorizer, err := services.NewAuthorizer(runtimeAuthJWKSURL, "")
+	authorizer, err := services.NewAuthorizer(buildJWKSURLs([]string{runtimeAuthJWKSURL}), "")
 	if err != nil {
 		return err
 	}
@@ -269,3 +269,22 @@ func listenRuntimeDaemon(app *fiber.App, stateDir string) error {
 	logger.Log().Info("Starting runtime daemon", zap.String("socket", runtimeSocket), zap.String("stateDir", stateDir))
 	return app.Listener(listener)
 }
+
+func buildJWKSURLs(values []string) []string {
+	urls := make([]string, 0, len(values))
+	seen := make(map[string]struct{}, len(values))
+
+	for _, url := range values {
+		url = strings.TrimSpace(url)
+		if url == "" {
+			continue
+		}
+		if _, ok := seen[url]; ok {
+			continue
+		}
+		seen[url] = struct{}{}
+		urls = append(urls, url)
+	}
+
+	return urls
+}

apps/druid/adapters/cli/serve.go

@@ -101,6 +101,13 @@ func loadWorkerRegistryStore() *registry.CredentialStore {
 	if len(config.Registries) == 0 {
 		_ = viper.UnmarshalKey("registries", &config.Registries)
 	}
+	if len(config.Registries) == 0 {
+		if path := viper.ConfigFileUsed(); path != "" {
+			if raw, err := os.ReadFile(path); err == nil {
+				_ = json.Unmarshal(raw, &config)
+			}
+		}
+	}
 	return registry.NewCredentialStore(config.Registries)
 }
 

apps/druid/adapters/cli/worker_pull.go

@@ -143,15 +143,6 @@ func (s *RuntimeSupervisor) EnsureWithOwner(artifact string, name string, ownerI
 				}
 			}
 			if runtimeScroll.ScrollYAML == "" {
-				if runtimeScroll.Status == domain.RuntimeScrollStatusError {
-					if ownerID != "" && runtimeScroll.OwnerID != ownerID {
-						runtimeScroll.OwnerID = ownerID
-						if err := s.store.UpdateScroll(runtimeScroll); err != nil {
-							return nil, err
-						}
-					}
-					return runtimeScroll, nil
-				}
 				if artifact == "" {
 					artifact = runtimeScroll.Artifact
 				}

apps/druid/core/services/runtime_supervisor.go

@@ -28,6 +28,7 @@ app_version: "1.0"
 serve: start
 commands:
   start:
+    run: restart
     procedures:
       - image: alpine:3.20
         command: ["true"]
@@ -459,7 +460,7 @@ func TestRuntimeSupervisorCreateUsesRequestedNamespaceForRoot(t *testing.T) {
 	}
 }
 
-func TestRuntimeSupervisorEnsureMaterializationFailureIsRemembered(t *testing.T) {
+func TestRuntimeSupervisorEnsureRetriesIncompleteMaterializationFailure(t *testing.T) {
 	store := newTestStateStore(t)
 	callbacks := NewWorkerCallbackManager()
 	backend := &fakeWorkerBackend{callbacks: callbacks, workerErr: errors.New("pull image failed")}
@@ -481,12 +482,15 @@ func TestRuntimeSupervisorEnsureMaterializationFailureIsRemembered(t *testing.T)
 		t.Fatalf("failed scroll = %#v", failed)
 	}
 
+	backend.workerErr = nil
+	backend.scrollYAML = cachedScrollYAML("start")
+	backend.digest = "sha256:recovered"
 	runtimeScroll, err := supervisor.Ensure("registry.local/missing:1.0", "broken-scroll", nil)
 	if err != nil {
-		t.Fatalf("second Ensure error = %v, want remembered runtime scroll", err)
+		t.Fatalf("second Ensure error = %v, want recovery", err)
 	}
-	if runtimeScroll.Status != domain.RuntimeScrollStatusError || backend.spawnCount != 1 {
-		t.Fatalf("runtimeScroll=%#v spawnCount=%d, want remembered error and no respawn", runtimeScroll, backend.spawnCount)
+	if runtimeScroll.Status != domain.RuntimeScrollStatusCreated || runtimeScroll.ScrollYAML == "" || runtimeScroll.ArtifactDigest != "sha256:recovered" || backend.spawnCount != 2 {
+		t.Fatalf("runtimeScroll=%#v spawnCount=%d, want recovered materialization", runtimeScroll, backend.spawnCount)
 	}
 }
 
@@ -1005,6 +1009,7 @@ app_version: "2.0"
 serve: start
 commands:
   start:
+    run: restart
     procedures:
       - image: alpine:3.20
         command: ["true"]

apps/druid/core/services/runtime_supervisor_test.go

@@ -13,6 +13,7 @@ import (
 
 	"github.com/gofiber/fiber/v2"
 	"github.com/golang-jwt/jwt/v4"
+	"github.com/highcard-dev/daemon/internal/core/ports"
 )
 
 func TestAuthorizerService_CheckHeaderAcceptsMultipleJWKS(t *testing.T) {
@@ -76,7 +77,7 @@ func TestAuthorizerService_CheckHeaderRejectsWrongSub(t *testing.T) {
 }
 
 func checkAuthorizationHeader(authorizer interface {
-	CheckHeader(*fiber.Ctx) (*time.Time, error)
+	CheckHeader(*fiber.Ctx) (*ports.AuthContext, error)
 }, token string) error {
 	app := fiber.New()
 	var authErr error