feat(metaschema): refresh the schema cache on a timer and make it concurrency-safe

cmd/migrate.go

@@ -34,7 +34,7 @@ func RunMigrations(logger *slog.Logger, config db.Config) error {
 		return errors.Wrap(err, "failed to connect to db")
 	}
 	metaschemaRepository := postgres.NewMetaSchemaRepository(logger, dbc)
-	metaschemaService := metaschema.NewService(metaschemaRepository)
+	metaschemaService := metaschema.NewService(metaschemaRepository, logger, 0)
 	if err = metaschemaService.MigrateDefault(context.Background()); err != nil {
 		return errors.Wrap(err, "failed to add default schemas to db")
 	}

cmd/serve.go

@@ -163,12 +163,16 @@ func StartServer(logger *slog.Logger, cfg *config.Frontier) error {
 		return err
 	}
 
-	// load metadata schema in memory from db
-	if schemas, err := deps.MetaSchemaService.List(context.Background()); err != nil {
+	// prime the metaschema cache and start its periodic refresh
+	if err := deps.MetaSchemaService.Init(ctx); err != nil {
 		logger.Warn("metaschemas initialization failed", "err", err)
-	} else {
-		logger.Info("metaschemas loaded", "count", len(schemas))
 	}
+	defer func() {
+		logger.Debug("cleaning up metaschemas")
+		if err := deps.MetaSchemaService.Close(); err != nil {
+			logger.Warn("metaschema service cleanup failed", "err", err)
+		}
+	}()
 
 	// apply schema
 	if err = deps.BootstrapService.MigrateSchema(ctx); err != nil {
@@ -489,7 +493,7 @@ func buildAPIDependencies(
 	domainService := domain.NewService(logger, domainRepository, userService, organizationService, membershipService)
 
 	metaschemaRepository := postgres.NewMetaSchemaRepository(logger, dbc)
-	metaschemaService := metaschema.NewService(metaschemaRepository)
+	metaschemaService := metaschema.NewService(metaschemaRepository, logger, cfg.App.Metaschema.RefreshInterval)
 
 	userPATService := userpat.NewService(logger, userPATRepo, cfg.App.PAT, organizationService, roleService, membershipService, projectService, auditRecordRepository)
 	membershipService.SetUserPATService(userPATService)

config/sample.config.yaml

@@ -204,6 +204,14 @@ app:
     # this is used to validate the webhook payloads
     encryption_key: "hash-secret-should-be-32-chars--"
 
+  # metaschema cache configuration
+  metaschema:
+    # how often each server reloads the metaschema cache from the database, so a
+    # change made on one server reaches the others. 0 disables the background
+    # refresh; the cache is still primed once at startup.
+    # e.g. 30s, 1m, 5m
+    refresh_interval: 1m
+
 db:
   driver: postgres
   url: postgres://frontier:@localhost:5432/frontier?sslmode=disable

core/metaschema/config.go

@@ -0,0 +1,11 @@
+package metaschema
+
+import "time"
+
+// Config holds runtime configuration for the metaschema service.
+type Config struct {
+	// RefreshInterval is how often each server reloads the metaschema cache from
+	// the database, so a change made on one pod reaches the others. A value of 0
+	// disables the background refresh; the cache is still primed once at startup.
+	RefreshInterval time.Duration `yaml:"refresh_interval" mapstructure:"refresh_interval" default:"1m"`
+}

core/metaschema/service.go

@@ -2,37 +2,56 @@ package metaschema
 
 import (
 	"context"
+	"fmt"
+	"log/slog"
+	"sync"
+	"time"
 
 	"github.com/raystack/frontier/pkg/utils"
 
 	"github.com/pkg/errors"
 	"github.com/raystack/frontier/pkg/metadata"
+	"github.com/robfig/cron/v3"
 	"github.com/xeipuuv/gojsonschema"
 )
 
 type Service struct {
 	repository      Repository
+	logger          *slog.Logger
+	refreshInterval time.Duration
+
+	mu              sync.RWMutex
 	metaSchemaCache map[string]MetaSchema
+
+	syncJob   *cron.Cron
+	syncJobMu sync.Mutex
 }
 
-func NewService(repository Repository) *Service {
+func NewService(repository Repository, logger *slog.Logger, refreshInterval time.Duration) *Service {
 	return &Service{
 		repository:      repository,
+		logger:          logger,
+		refreshInterval: refreshInterval,
 		metaSchemaCache: make(map[string]MetaSchema),
 	}
 }
 
-func (s Service) Create(ctx context.Context, toCreate MetaSchema) (MetaSchema, error) {
+func (s *Service) Create(ctx context.Context, toCreate MetaSchema) (MetaSchema, error) {
 	mschema, err := s.repository.Create(ctx, toCreate)
 	if err != nil {
 		return MetaSchema{}, err
 	}
+	s.mu.Lock()
 	s.metaSchemaCache[mschema.Name] = mschema
+	s.mu.Unlock()
 	return mschema, nil
 }
 
-func (s Service) Get(ctx context.Context, idOrName string) (MetaSchema, error) {
-	if schema, ok := s.metaSchemaCache[idOrName]; ok {
+func (s *Service) Get(ctx context.Context, idOrName string) (MetaSchema, error) {
+	s.mu.RLock()
+	schema, ok := s.metaSchemaCache[idOrName]
+	s.mu.RUnlock()
+	if ok {
 		return schema, nil
 	}
 
@@ -41,65 +60,60 @@ func (s Service) Get(ctx context.Context, idOrName string) (MetaSchema, error) {
 		if err != nil {
 			return MetaSchema{}, err
 		}
-
 		return schema, nil
 	}
 	return MetaSchema{}, ErrInvalidID
 }
 
-func (s Service) List(ctx context.Context) ([]MetaSchema, error) {
-	if len(s.metaSchemaCache) == 0 {
-		schemas, err := s.repository.List(ctx)
-		if err != nil {
-			return nil, err
-		}
-		for _, schema := range schemas {
-			s.metaSchemaCache[schema.Name] = schema
-		}
-		return schemas, nil
-	}
-
+func (s *Service) List(ctx context.Context) ([]MetaSchema, error) {
+	s.mu.RLock()
+	defer s.mu.RUnlock()
 	schemas := make([]MetaSchema, 0, len(s.metaSchemaCache))
 	for _, schema := range s.metaSchemaCache {
 		schemas = append(schemas, schema)
 	}
 	return schemas, nil
 }
 
-func (s Service) Update(ctx context.Context, id string, toUpdate MetaSchema) (MetaSchema, error) {
+func (s *Service) Update(ctx context.Context, id string, toUpdate MetaSchema) (MetaSchema, error) {
 	if utils.IsValidUUID(id) {
 		schema, err := s.repository.Update(ctx, id, toUpdate)
 		if err != nil {
 			return MetaSchema{}, err
 		}
+		s.mu.Lock()
 		s.metaSchemaCache[schema.Name] = schema
+		s.mu.Unlock()
 		return schema, nil
 	}
 	return MetaSchema{}, ErrInvalidID
 }
 
-func (s Service) Delete(ctx context.Context, id string) error {
+func (s *Service) Delete(ctx context.Context, id string) error {
 	if utils.IsValidUUID(id) {
 		name, err := s.repository.Delete(ctx, id)
 		if err != nil {
 			return err
 		}
-
+		s.mu.Lock()
 		delete(s.metaSchemaCache, name)
+		s.mu.Unlock()
 		return nil
 	}
 	return ErrInvalidID
 }
 
-func (s Service) MigrateDefault(ctx context.Context) error {
+func (s *Service) MigrateDefault(ctx context.Context) error {
 	return s.repository.MigrateDefaults(ctx)
 }
 
-// Validate the metadata against the json-schema. In case metaschema doesn't exists in the cache, it will return nil (no validation)
-func (s Service) Validate(mdata metadata.Metadata, name string) error {
-	var mschema MetaSchema
-	var ok bool
-	if mschema, ok = s.metaSchemaCache[name]; !ok {
+// Validate checks the metadata against the json-schema. When the named
+// metaschema is not in the cache it returns nil (no validation).
+func (s *Service) Validate(mdata metadata.Metadata, name string) error {
+	s.mu.RLock()
+	mschema, ok := s.metaSchemaCache[name]
+	s.mu.RUnlock()
+	if !ok {
 		return nil
 	}
 
@@ -115,3 +129,79 @@ func (s Service) Validate(mdata metadata.Metadata, name string) error {
 	}
 	return nil
 }
+
+// reload replaces the cache with the current set of metaschemas from the
+// database. It holds the write lock across the read and the swap, so a Create,
+// Update, or Delete that runs concurrently applies to the cache after the swap
+// and is never lost to a stale snapshot. On a database error it returns without
+// touching the cache, and it refuses to swap an empty set over a populated
+// cache, so neither a blip nor an unexpected empty read blanks validation. The
+// caller decides what to do with the error: fail startup for the initial prime,
+// log and keep the cache for a scheduled refresh.
+func (s *Service) reload(ctx context.Context) error {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	schemas, err := s.repository.List(ctx)
+	if err != nil {
+		return err
+	}
+	if len(schemas) == 0 && len(s.metaSchemaCache) > 0 {
+		return fmt.Errorf("metaschema list returned no schemas, keeping the current cache of %d", len(s.metaSchemaCache))
+	}
+	fresh := make(map[string]MetaSchema, len(schemas))
+	for _, schema := range schemas {
+		fresh[schema.Name] = schema
+	}
+	s.metaSchemaCache = fresh
+	return nil
+}
+
+// Init primes the cache from the database and, when refreshInterval is greater
+// than zero, starts a background job that reloads it on that interval so a
+// change made on one pod reaches the others.
+func (s *Service) Init(ctx context.Context) error {
+	// The initial prime must succeed. If it fails, startup stops here rather
+	// than serving with an empty cache, which would skip validation silently.
+	if err := s.reload(ctx); err != nil {
+		return fmt.Errorf("prime metaschema cache: %w", err)
+	}
+	s.mu.RLock()
+	count := len(s.metaSchemaCache)
+	s.mu.RUnlock()
+	s.logger.InfoContext(ctx, "metaschemas loaded", "count", count)
+
+	if s.refreshInterval <= 0 {
+		return nil
+	}
+
+	s.syncJobMu.Lock()
+	defer s.syncJobMu.Unlock()
+	if s.syncJob != nil {
+		<-s.syncJob.Stop().Done()
+	}
+	s.syncJob = cron.New(cron.WithChain(
+		cron.SkipIfStillRunning(cron.DefaultLogger),
+		cron.Recover(cron.DefaultLogger),
+	))
+	if _, err := s.syncJob.AddFunc(fmt.Sprintf("@every %s", s.refreshInterval.String()), func() {
+		// A scheduled refresh keeps the last good cache on error, so a database
+		// blip does not blank validation between successful reloads.
+		if err := s.reload(ctx); err != nil {
+			s.logger.WarnContext(ctx, "metaschema cache refresh failed", "err", err)
+		}
+	}); err != nil {
+		return err
+	}
+	s.syncJob.Start()
+	return nil
+}
+
+// Close stops the background refresh job. It is safe to call when none started.
+func (s *Service) Close() error {
+	s.syncJobMu.Lock()
+	defer s.syncJobMu.Unlock()
+	if s.syncJob != nil {
+		<-s.syncJob.Stop().Done()
+	}
+	return nil
+}