Plugins: create the settings collection instead of waiting for it forever

01a8fec fixed the advice that led operators into this, but advice is not a
guard: a stack still running PB_BOOTSTRAP=false gets no app_settings
collection on upgrade, and the plugin panel sits at 503 while the retry
loop reads a collection that does not exist.

The fix is not to soften the reading. A missing collection stays "not
ready" rather than "no plugins configured", because the alternative lets
the first save write a fresh document over settings the server merely
failed to find - the failure this whole line of work exists to prevent.
Instead the server now fixes the cause: on a missing collection it creates
that collection and reads again.

Three pieces:

bootstrap.EnsureCollection creates one named collection from the desired
schema if absent, and nothing else. Deliberately narrower than Run - no
field reconcile elsewhere, no super-admin - so it is safe to call on a
deployment that turned the full bootstrap off. It creates the collection
the server cannot start without, not the schema the operator declined.

The store tells a missing collection apart from an outage. A 404 from a
list means the collection itself is gone: an existing but empty one answers
200 with no items. That is tagged errNoCollection, which wraps errNotReady
so every write is still refused, and IsMissingCollection narrows it. The
distinction matters because the remedies are opposites - creating
collections against a flaky database is exactly the wrong reflex, and a
test pins that an outage does not trigger it.

loadPlugins acts on the tag once, then re-reads. Failing to create is
reported as the original read error rather than the repair's, so the log
names the real problem.

Six tests: the tag and its negative in internal/plugins, and three in
internal/api against a fake PocketBase covering the collection being
created exactly once, an existing collection not being recreated, and an
outage creating nothing.

Docs from 01a8fec are corrected in the same pass - they said the panel
would answer 503 forever, which is no longer true. They now say what still
depends on the bootstrap (every other collection and field) and what does
not (app_settings alone).

go build, go vet and go test ./... pass; compose files still parse. Not
verified: no Docker CLI here, so the repair has not been exercised against
a real PocketBase, only the fake.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

API Server/internal/api/plugins_bootstrap_test.go

@@ -0,0 +1,157 @@
+package api
+
+import (
+	"context"
+	"encoding/json"
+	"net/http"
+	"net/http/httptest"
+	"strings"
+	"sync"
+	"testing"
+
+	"drivervault/apiserver/internal/config"
+	"drivervault/apiserver/internal/pb"
+)
+
+// fakeSchemaPB is a PocketBase stand-in for the "app_settings does not exist"
+// case: a stack upgraded with PB_BOOTSTRAP off, where the on-boot schema pass
+// never created the collection the plugin settings live in.
+type fakeSchemaPB struct {
+	mu sync.Mutex
+
+	created     bool // app_settings exists
+	createCalls int  // POST /api/collections
+	unreachable bool // every records call fails with a 502
+}
+
+func (f *fakeSchemaPB) server(t *testing.T) *httptest.Server {
+	t.Helper()
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		f.mu.Lock()
+		defer f.mu.Unlock()
+
+		switch {
+		case strings.HasSuffix(r.URL.Path, "/auth-with-password"):
+			writeTestJSON(w, http.StatusOK, map[string]any{"token": "test-token"})
+
+		// The settings read. 404 until the collection exists — PocketBase
+		// answers a list against a missing collection that way, while an
+		// existing but empty one answers 200 with no items.
+		case r.Method == http.MethodGet && r.URL.Path == "/api/collections/app_settings/records":
+			if f.unreachable {
+				http.Error(w, "connection refused", http.StatusBadGateway)
+				return
+			}
+			if !f.created {
+				http.Error(w, `{"message":"Missing collection context."}`, http.StatusNotFound)
+				return
+			}
+			writeTestJSON(w, http.StatusOK, map[string]any{
+				"page": 1, "perPage": 1, "totalItems": 0, "totalPages": 1,
+				"items": []any{},
+			})
+
+		// The schema calls EnsureCollection makes.
+		case r.Method == http.MethodGet && r.URL.Path == "/api/collections":
+			items := []any{
+				map[string]any{"id": "col_users", "name": "users", "fields": []any{
+					map[string]any{"name": "email", "type": "text"},
+				}},
+			}
+			if f.created {
+				items = append(items, map[string]any{"id": "col_app", "name": "app_settings"})
+			}
+			writeTestJSON(w, http.StatusOK, map[string]any{"items": items})
+
+		case r.Method == http.MethodPost && r.URL.Path == "/api/collections":
+			var body struct {
+				Name string `json:"name"`
+			}
+			_ = json.NewDecoder(r.Body).Decode(&body)
+			if body.Name != colAppSettings {
+				t.Errorf("created collection %q, want %q", body.Name, colAppSettings)
+			}
+			f.createCalls++
+			f.created = true
+			writeTestJSON(w, http.StatusOK, map[string]any{"id": "col_app"})
+
+		default:
+			t.Errorf("unexpected request: %s %s", r.Method, r.URL.Path)
+			http.Error(w, "unexpected", http.StatusInternalServerError)
+		}
+	}))
+	t.Cleanup(srv.Close)
+	return srv
+}
+
+func writeTestJSON(w http.ResponseWriter, status int, v any) {
+	w.Header().Set("Content-Type", "application/json")
+	w.WriteHeader(status)
+	_ = json.NewEncoder(w).Encode(v)
+}
+
+func newSchemaTestServer(t *testing.T, f *fakeSchemaPB) *Server {
+	t.Helper()
+	pbSrv := f.server(t)
+	// PluginsFile empty: no legacy import in play here.
+	return New(config.Config{UsersCollection: "users", PluginsFile: ""},
+		pb.New(pbSrv.URL, "admin@test.local", "pw"))
+}
+
+// The guard: a missing settings collection is created rather than retried
+// forever, so a stack upgraded with the bootstrap off still comes up.
+func TestLoadPluginsCreatesMissingSettingsCollection(t *testing.T) {
+	f := &fakeSchemaPB{}
+	s := newSchemaTestServer(t, f)
+
+	if err := s.loadPlugins(context.Background()); err != nil {
+		t.Fatalf("loadPlugins should recover by creating the collection, got %v", err)
+	}
+	if !s.plugins.Ready() {
+		t.Fatal("plugins should be loaded after the collection was created")
+	}
+
+	f.mu.Lock()
+	calls := f.createCalls
+	f.mu.Unlock()
+	if calls != 1 {
+		t.Fatalf("expected the collection to be created once, got %d", calls)
+	}
+}
+
+// Once the collection exists, nothing is created again.
+func TestLoadPluginsDoesNotRecreateExistingCollection(t *testing.T) {
+	f := &fakeSchemaPB{created: true}
+	s := newSchemaTestServer(t, f)
+
+	if err := s.loadPlugins(context.Background()); err != nil {
+		t.Fatalf("loadPlugins: %v", err)
+	}
+	f.mu.Lock()
+	calls := f.createCalls
+	f.mu.Unlock()
+	if calls != 0 {
+		t.Fatalf("an existing collection must not be recreated, got %d creates", calls)
+	}
+}
+
+// A database that is merely unreachable must NOT trigger schema surgery: the
+// remedy there is to wait, and creating collections against a flaky database is
+// exactly the wrong reflex.
+func TestLoadPluginsDoesNotCreateOnOutage(t *testing.T) {
+	f := &fakeSchemaPB{created: true, unreachable: true}
+	s := newSchemaTestServer(t, f)
+
+	if err := s.loadPlugins(context.Background()); err == nil {
+		t.Fatal("expected an error while the database is unreachable")
+	}
+	if s.plugins.Ready() {
+		t.Fatal("plugins must not report ready after an outage")
+	}
+	f.mu.Lock()
+	calls := f.createCalls
+	f.mu.Unlock()
+	if calls != 0 {
+		t.Fatalf("an outage must not create collections, got %d creates", calls)
+	}
+}

API Server/internal/api/server.go

@@ -127,6 +127,7 @@ import (
 	"sync"
 	"time"
 
+	"drivervault/apiserver/internal/bootstrap"
 	"drivervault/apiserver/internal/config"
 	"drivervault/apiserver/internal/ocpp"
 	"drivervault/apiserver/internal/pb"
@@ -222,9 +223,34 @@ func (s *Server) StartPlugins() error {
 	return err
 }
 
-// loadPlugins imports a pre-PocketBase plugins.json if one is still lying around
-// and the database holds no settings yet, then reads the settings.
+// loadPlugins reads the settings, creating the collection they live in if it
+// turns out not to exist.
+//
+// That happens on a stack upgraded with PB_BOOTSTRAP off: the on-boot schema
+// pass never ran, so app_settings was never created, and because a missing
+// collection is read as "not ready" (never as "no plugins configured", which
+// would let the first save overwrite settings the server merely failed to find)
+// the retry below would spin forever with the plugin panel stuck at 503. So
+// create just that one collection — not a full schema reconcile, which an
+// operator who turned the bootstrap off has not asked for — and read again.
 func (s *Server) loadPlugins(ctx context.Context) error {
+	err := s.readPlugins(ctx)
+	if err == nil || !plugins.IsMissingCollection(err) {
+		return err
+	}
+
+	log.Printf("plugins: the %s collection does not exist; creating it", colAppSettings)
+	if repairErr := bootstrap.EnsureCollection(ctx, s.pb, colAppSettings); repairErr != nil {
+		log.Printf("plugins: could not create %s: %v", colAppSettings, repairErr)
+		return err // report the original problem, not the repair's
+	}
+	log.Printf("plugins: created %s", colAppSettings)
+	return s.readPlugins(ctx)
+}
+
+// readPlugins imports a pre-PocketBase plugins.json if one is still lying around
+// and the database holds no settings yet, then reads the settings.
+func (s *Server) readPlugins(ctx context.Context) error {
 	switch migrated, err := plugins.MigrateLegacyFile(ctx, s.pluginStore, s.legacyPlugins); {
 	case err != nil:
 		// Not fatal: the read below reports the real problem if there is one.

API Server/internal/bootstrap/bootstrap.go

@@ -193,6 +193,43 @@ func Run(ctx context.Context, client *pb.Client, opts Options) error {
 	return nil
 }
 
+// EnsureCollection creates one collection from the desired schema if it is not
+// already present, and does nothing otherwise.
+//
+// It is deliberately narrower than Run: no field reconcile on other
+// collections, no super-admin. That makes it safe to call on a deployment that
+// turned the full bootstrap off, which is the case it exists for — a collection
+// the server cannot run without (app_settings, holding the plugin settings) is
+// missing, and no amount of retrying a read will conjure it.
+func EnsureCollection(ctx context.Context, client *pb.Client, name string) error {
+	if _, known := collectionsSchema[name]; !known {
+		return fmt.Errorf("unknown collection %q", name)
+	}
+	if err := client.Authenticate(ctx); err != nil {
+		return fmt.Errorf("authenticate: %w", err)
+	}
+	existing, err := listCollections(ctx, client)
+	if err != nil {
+		return fmt.Errorf("list collections: %w", err)
+	}
+
+	idByName := make(map[string]string, len(existing))
+	present := false
+	for _, c := range existing {
+		idByName[c.Name] = c.ID
+		if c.Name == name {
+			present = true
+		}
+	}
+	if present {
+		return nil
+	}
+	if err := createCollection(ctx, client, name, detectFormat(existing), idByName); err != nil {
+		return fmt.Errorf("create %s: %w", name, err)
+	}
+	return nil
+}
+
 // --- PocketBase collection calls -------------------------------------------
 
 type collectionMeta struct {

API Server/internal/plugins/README.md

@@ -263,6 +263,11 @@ it runs as its own process/container, an external plugin is also the
 - **Upgrading from a pre-PocketBase install**: an existing `plugins.json` (see
   `PLUGINS_FILE`) is imported into the database on the first boot that finds no
   settings there, and renamed to `plugins.json.migrated`.
+- **If the `app_settings` collection is missing** — an upgrade on a stack that
+  runs with `PB_BOOTSTRAP=false`, so the on-boot schema pass never created it —
+  the server creates that one collection itself and reads again. A missing
+  collection is told apart from a database that is merely unreachable, because
+  the remedy differs: creating collections is the wrong reflex during an outage.
 - **Secrets** (`Secret: true` fields) are returned masked. On save, a field still
   equal to the mask keeps its stored value; send a new value to change it, or an
   empty string to clear it.

API Server/internal/plugins/manager.go

@@ -576,6 +576,12 @@ var (
 	errUnknown  = errors.New("unknown plugin")
 	errPersist  = errors.New("plugin settings could not be saved")
 	errNotReady = errors.New("plugin settings are not loaded yet")
+
+	// errNoCollection is a not-ready that waiting alone will never resolve: the
+	// collection the settings live in does not exist. It wraps errNotReady, so
+	// the safety behaviour (refuse every write) is unchanged; the extra tag only
+	// lets the caller fix the cause instead of retrying forever.
+	errNoCollection = fmt.Errorf("%w: the settings collection does not exist", errNotReady)
 )
 
 // IsUnknown reports whether err came from addressing a plugin that doesn't exist.
@@ -590,3 +596,8 @@ func IsPersist(err error) bool { return errors.Is(err, errPersist) }
 // the store was unreachable at boot and is still being retried. Callers should
 // answer 503 rather than present the plugin list as empty.
 func IsNotReady(err error) bool { return errors.Is(err, errNotReady) }
+
+// IsMissingCollection reports whether err means the settings collection does not
+// exist. It implies IsNotReady, and narrows it: retrying cannot help, so the
+// caller should create the collection and read again.
+func IsMissingCollection(err error) bool { return errors.Is(err, errNoCollection) }

API Server/internal/plugins/store.go

@@ -103,8 +103,15 @@ func (s *pbStore) find(ctx context.Context) (settingsRecord, bool, error) {
 
 	res, err := s.client.List(ctx, s.collection, q)
 	if err != nil {
-		// A missing collection means bootstrap has not run yet — still "not
-		// ready" rather than "no settings", so nothing gets overwritten.
+		if isNotFound(err) {
+			// A 404 from a list means the collection itself is absent: listing an
+			// existing but empty collection answers 200 with no items. Tagged
+			// apart from a plain outage because retrying a read against a
+			// collection that does not exist can never succeed — the caller has
+			// to create it. Still a flavour of not-ready, so nothing is
+			// overwritten in the meantime.
+			return settingsRecord{}, false, fmt.Errorf("%w: %v", errNoCollection, err)
+		}
 		return settingsRecord{}, false, fmt.Errorf("%w: %v", errNotReady, err)
 	}
 	var items []settingsRecord

API Server/internal/plugins/store_pb_test.go

@@ -192,17 +192,35 @@ func TestPBStoreUnreachableStaysUnloaded(t *testing.T) {
 }
 
 // A missing collection (bootstrap has not run) is "not ready", never "empty" —
-// otherwise the first save would write a fresh document over nothing.
+// otherwise the first save would write a fresh document over nothing. It is also
+// tagged separately, because retrying alone can never resolve it.
 func TestPBStoreMissingCollectionIsNotReady(t *testing.T) {
 	m := NewManager(newPBStore(t, &fakePB{missingCol: true}))
-	if err := m.Load(context.Background()); !IsNotReady(err) {
+	err := m.Load(context.Background())
+	if !IsNotReady(err) {
 		t.Fatalf("expected not-ready for a missing collection, got %v", err)
 	}
+	if !IsMissingCollection(err) {
+		t.Fatalf("a missing collection must be distinguishable, got %v", err)
+	}
 	if m.Ready() {
 		t.Fatal("manager must not be ready without its collection")
 	}
 }
 
+// A plain outage must NOT look like a missing collection: creating the
+// collection is not the remedy for a database that is merely unreachable.
+func TestPBStoreOutageIsNotAMissingCollection(t *testing.T) {
+	m := NewManager(newPBStore(t, &fakePB{listErr: true}))
+	err := m.Load(context.Background())
+	if !IsNotReady(err) {
+		t.Fatalf("expected not-ready, got %v", err)
+	}
+	if IsMissingCollection(err) {
+		t.Fatalf("an outage must not be reported as a missing collection: %v", err)
+	}
+}
+
 // The legacy file is imported into PocketBase exactly once.
 func TestPBStoreLegacyImport(t *testing.T) {
 	f := &fakePB{}