feat(unified-storage): add qos support for the resource server (#105939)

pkg/server/distributor_test.go

@@ -352,7 +352,18 @@ func createBaselineServer(t *testing.T, dbType, dbConnStr string, testNamespaces
 	require.NoError(t, err)
 	searchOpts, err := search.NewSearchOptions(features, cfg, tracer, docBuilders, nil)
 	require.NoError(t, err)
-	server, err := sql.NewResourceServer(nil, cfg, tracer, nil, nil, searchOpts, nil, nil, features)
+	server, err := sql.NewResourceServer(sql.ServerOptions{
+		DB:             nil,
+		Cfg:            cfg,
+		Tracer:         tracer,
+		Reg:            nil,
+		AccessClient:   nil,
+		SearchOptions:  searchOpts,
+		StorageMetrics: nil,
+		IndexMetrics:   nil,
+		Features:       features,
+		QOSQueue:       nil,
+	})
 	require.NoError(t, err)
 
 	testUserA := &identity.StaticRequester{

pkg/server/module_server.go

@@ -53,7 +53,16 @@ func NewModule(opts Options,
 	return s, nil
 }
 
-func newModuleServer(opts Options, apiOpts api.ServerOptions, features featuremgmt.FeatureToggles, cfg *setting.Cfg, storageMetrics *resource.StorageMetrics, indexMetrics *resource.BleveIndexMetrics, reg prometheus.Registerer, promGatherer prometheus.Gatherer, license licensing.Licensing) (*ModuleServer, error) {
+func newModuleServer(opts Options,
+	apiOpts api.ServerOptions,
+	features featuremgmt.FeatureToggles,
+	cfg *setting.Cfg,
+	storageMetrics *resource.StorageMetrics,
+	indexMetrics *resource.BleveIndexMetrics,
+	reg prometheus.Registerer,
+	promGatherer prometheus.Gatherer,
+	license licensing.Licensing,
+) (*ModuleServer, error) {
 	rootCtx, shutdownFn := context.WithCancel(context.Background())
 
 	s := &ModuleServer{

pkg/setting/setting.go

@@ -567,6 +567,9 @@ type Cfg struct {
 	IndexRebuildInterval                       time.Duration
 	IndexCacheTTL                              time.Duration
 	EnableSharding                             bool
+	QOSEnabled                                 bool
+	QOSNumberWorker                            int
+	QOSMaxSizePerTenant                        int
 	MemberlistBindAddr                         string
 	MemberlistAdvertiseAddr                    string
 	MemberlistAdvertisePort                    int

pkg/setting/setting_unified_storage.go

@@ -49,13 +49,16 @@ func (cfg *Cfg) setUnifiedStorageConfig() {
 	}
 	cfg.UnifiedStorage = storageConfig
 
-	// Set indexer config for unified storaae
+	// Set indexer config for unified storage
 	section := cfg.Raw.Section("unified_storage")
 	cfg.MaxPageSizeBytes = section.Key("max_page_size_bytes").MustInt(0)
 	cfg.IndexPath = section.Key("index_path").String()
 	cfg.IndexWorkers = section.Key("index_workers").MustInt(10)
 	cfg.IndexMaxBatchSize = section.Key("index_max_batch_size").MustInt(100)
 	cfg.EnableSharding = section.Key("enable_sharding").MustBool(false)
+	cfg.QOSEnabled = section.Key("qos_enabled").MustBool(false)
+	cfg.QOSNumberWorker = section.Key("qos_num_worker").MustInt(16)
+	cfg.QOSMaxSizePerTenant = section.Key("qos_max_size_per_tenant").MustInt(1000)
 	cfg.MemberlistBindAddr = section.Key("memberlist_bind_addr").String()
 	cfg.MemberlistAdvertiseAddr = section.Key("memberlist_advertise_addr").String()
 	cfg.MemberlistAdvertisePort = section.Key("memberlist_advertise_port").MustInt(7946)

pkg/storage/unified/client.go

@@ -20,6 +20,7 @@ import (
 	"github.com/grafana/dskit/flagext"
 	"github.com/grafana/dskit/grpcclient"
 	"github.com/grafana/dskit/middleware"
+	"github.com/grafana/dskit/services"
 
 	infraDB "github.com/grafana/grafana/pkg/infra/db"
 	"github.com/grafana/grafana/pkg/infra/tracing"
@@ -31,6 +32,7 @@ import (
 	"github.com/grafana/grafana/pkg/storage/unified/resource"
 	"github.com/grafana/grafana/pkg/storage/unified/search"
 	"github.com/grafana/grafana/pkg/storage/unified/sql"
+	"github.com/grafana/grafana/pkg/util/scheduler"
 )
 
 type Options struct {
@@ -49,7 +51,10 @@ type clientMetrics struct {
 }
 
 // This adds a UnifiedStorage client into the wire dependency tree
-func ProvideUnifiedStorageClient(opts *Options, storageMetrics *resource.StorageMetrics, indexMetrics *resource.BleveIndexMetrics) (resource.ResourceClient, error) {
+func ProvideUnifiedStorageClient(opts *Options,
+	storageMetrics *resource.StorageMetrics,
+	indexMetrics *resource.BleveIndexMetrics,
+) (resource.ResourceClient, error) {
 	// See: apiserver.applyAPIServerConfig(cfg, features, o)
 	apiserverCfg := opts.Cfg.SectionWithEnvOverrides("grafana-apiserver")
 	client, err := newClient(options.StorageOptions{
@@ -83,6 +88,7 @@ func newClient(opts options.StorageOptions,
 	indexMetrics *resource.BleveIndexMetrics,
 ) (resource.ResourceClient, error) {
 	ctx := context.Background()
+
 	switch opts.StorageType {
 	case options.StorageTypeFile:
 		if opts.DataPath == "" {
@@ -146,13 +152,50 @@ func newClient(opts options.StorageOptions,
 		}
 		return client, nil
 
-	// Use the local SQL
 	default:
 		searchOptions, err := search.NewSearchOptions(features, cfg, tracer, docs, indexMetrics)
 		if err != nil {
 			return nil, err
 		}
-		server, err := sql.NewResourceServer(db, cfg, tracer, reg, authzc, searchOptions, storageMetrics, indexMetrics, features)
+
+		serverOptions := sql.ServerOptions{
+			DB:             db,
+			Cfg:            cfg,
+			Tracer:         tracer,
+			Reg:            reg,
+			AccessClient:   authzc,
+			SearchOptions:  searchOptions,
+			StorageMetrics: storageMetrics,
+			IndexMetrics:   indexMetrics,
+			Features:       features,
+		}
+
+		if cfg.QOSEnabled {
+			qosReg := prometheus.WrapRegistererWithPrefix("resource_server_qos_", reg)
+			queue := scheduler.NewQueue(&scheduler.QueueOptions{
+				MaxSizePerTenant: cfg.QOSMaxSizePerTenant,
+				Registerer:       qosReg,
+				Logger:           cfg.Logger,
+			})
+			if err := services.StartAndAwaitRunning(ctx, queue); err != nil {
+				return nil, fmt.Errorf("failed to start queue: %w", err)
+			}
+			scheduler, err := scheduler.NewScheduler(queue, &scheduler.Config{
+				NumWorkers: cfg.QOSNumberWorker,
+				Logger:     cfg.Logger,
+			})
+			if err != nil {
+				return nil, fmt.Errorf("failed to create scheduler: %w", err)
+			}
+
+			err = services.StartAndAwaitRunning(ctx, scheduler)
+			if err != nil {
+				return nil, fmt.Errorf("failed to start scheduler: %w", err)
+			}
+			serverOptions.QOSQueue = queue
+		}
+
+		server, err := sql.NewResourceServer(serverOptions)
 		if err != nil {
 			return nil, err
 		}

pkg/storage/unified/resource/errors.go

@@ -12,6 +12,7 @@ import (
 	grpcstatus "google.golang.org/grpc/status"
 
 	"github.com/grafana/grafana/pkg/storage/unified/resourcepb"
+	"github.com/grafana/grafana/pkg/util/scheduler"
 )
 
 // Package-level errors.
@@ -50,6 +51,14 @@ func NewNotFoundError(key *resourcepb.ResourceKey) *resourcepb.ErrorResult {
 	}
 }
 
+func NewTooManyRequestsError(msg string) *resourcepb.ErrorResult {
+	return &resourcepb.ErrorResult{
+		Message: msg,
+		Code:    http.StatusTooManyRequests,
+		Reason:  string(metav1.StatusReasonTooManyRequests),
+	}
+}
+
 // Convert golang errors to status result errors that can be returned to a client
 func AsErrorResult(err error) *resourcepb.ErrorResult {
 	if err == nil {
@@ -125,3 +134,10 @@ func GetError(res *resourcepb.ErrorResult) error {
 	}
 	return status
 }
+
+func HandleQueueError[T any](err error, makeResp func(*resourcepb.ErrorResult) *T) (*T, error) {
+	if errors.Is(err, scheduler.ErrTenantQueueFull) {
+		return makeResp(NewTooManyRequestsError("tenant queue is full, please try again later")), nil
+	}
+	return makeResp(AsErrorResult(err)), nil
+}

pkg/storage/unified/resource/server.go

@@ -19,9 +19,20 @@ import (
 	"k8s.io/apimachinery/pkg/apis/meta/v1/unstructured"
 
 	claims "github.com/grafana/authlib/types"
+	"github.com/grafana/dskit/backoff"
 
 	"github.com/grafana/grafana/pkg/apimachinery/utils"
 	"github.com/grafana/grafana/pkg/storage/unified/resourcepb"
+	"github.com/grafana/grafana/pkg/util/scheduler"
+)
+
+const (
+	// DefaultMaxBackoff is the default maximum backoff duration for enqueue operations.
+	DefaultMaxBackoff = 1 * time.Second
+	// DefaultMinBackoff is the default minimum backoff duration for enqueue operations.
+	DefaultMinBackoff = 100 * time.Millisecond
+	// DefaultMaxRetries is the default maximum number of retries for enqueue operations.
+	DefaultMaxRetries = 3
 )
 
 // ResourceServer implements all gRPC services
@@ -134,6 +145,10 @@ type BlobSupport interface {
 	// TODO? List+Delete?  This is for admin access
 }
 
+type QOSEnqueuer interface {
+	Enqueue(ctx context.Context, tenantID string, runnable func(ctx context.Context)) error
+}
+
 type BlobConfig struct {
 	// The CDK configuration URL
 	URL string
@@ -203,7 +218,11 @@ type ResourceServerOptions struct {
 
 	IndexMetrics *BleveIndexMetrics
 
+	// MaxPageSizeBytes is the maximum size of a page in bytes.
 	MaxPageSizeBytes int
+
+	// QOSQueue is the quality of service queue used to enqueue
+	QOSQueue QOSEnqueuer
 }
 
 func NewResourceServer(opts ResourceServerOptions) (ResourceServer, error) {
@@ -222,6 +241,7 @@ func NewResourceServer(opts ResourceServerOptions) (ResourceServer, error) {
 	if opts.Diagnostics == nil {
 		opts.Diagnostics = &noopService{}
 	}
+
 	if opts.Now == nil {
 		opts.Now = func() int64 {
 			return time.Now().UnixMilli()
@@ -233,6 +253,10 @@ func NewResourceServer(opts ResourceServerOptions) (ResourceServer, error) {
 		opts.MaxPageSizeBytes = 1024 * 1024 * 2
 	}
 
+	if opts.QOSQueue == nil {
+		opts.QOSQueue = scheduler.NewNoopQueue()
+	}
+
 	// Initialize the blob storage
 	blobstore := opts.Blob.Backend
 	if blobstore == nil {
@@ -275,6 +299,8 @@ func NewResourceServer(opts ResourceServerOptions) (ResourceServer, error) {
 		storageMetrics:   opts.storageMetrics,
 		indexMetrics:     opts.IndexMetrics,
 		maxPageSizeBytes: opts.MaxPageSizeBytes,
+		reg:              opts.Reg,
+		queue:            opts.QOSQueue,
 	}
 
 	if opts.Search.Resources != nil {
@@ -321,6 +347,8 @@ type server struct {
 	initErr error
 
 	maxPageSizeBytes int
+	reg              prometheus.Registerer
+	queue            QOSEnqueuer
 }
 
 // Init implements ResourceServer.
@@ -570,6 +598,25 @@ func (s *server) Create(ctx context.Context, req *resourcepb.CreateRequest) (*re
 		return rsp, nil
 	}
 
+	var (
+		res *resourcepb.CreateResponse
+		err error
+	)
+	runErr := s.runInQueue(ctx, req.Key.Namespace, func(ctx context.Context) {
+		res, err = s.create(ctx, user, req)
+	})
+	if runErr != nil {
+		return HandleQueueError(runErr, func(e *resourcepb.ErrorResult) *resourcepb.CreateResponse {
+			return &resourcepb.CreateResponse{Error: e}
+		})
+	}
+
+	return res, err
+}
+
+func (s *server) create(ctx context.Context, user claims.AuthInfo, req *resourcepb.CreateRequest) (*resourcepb.CreateResponse, error) {
+	rsp := &resourcepb.CreateResponse{}
+
 	event, e := s.newEvent(ctx, user, req.Key, req.Value, nil)
 	if e != nil {
 		rsp.Error = e
@@ -605,6 +652,24 @@ func (s *server) Update(ctx context.Context, req *resourcepb.UpdateRequest) (*re
 		return rsp, nil
 	}
 
+	var (
+		res *resourcepb.UpdateResponse
+		err error
+	)
+	runErr := s.runInQueue(ctx, req.Key.Namespace, func(ctx context.Context) {
+		res, err = s.update(ctx, user, req)
+	})
+	if runErr != nil {
+		return HandleQueueError(runErr, func(e *resourcepb.ErrorResult) *resourcepb.UpdateResponse {
+			return &resourcepb.UpdateResponse{Error: e}
+		})
+	}
+
+	return res, err
+}
+
+func (s *server) update(ctx context.Context, user claims.AuthInfo, req *resourcepb.UpdateRequest) (*resourcepb.UpdateResponse, error) {
+	rsp := &resourcepb.UpdateResponse{}
 	latest := s.backend.ReadResource(ctx, &resourcepb.ReadRequest{
 		Key: req.Key,
 	})
@@ -654,6 +719,25 @@ func (s *server) Delete(ctx context.Context, req *resourcepb.DeleteRequest) (*re
 		return rsp, nil
 	}
 
+	var (
+		res *resourcepb.DeleteResponse
+		err error
+	)
+
+	runErr := s.runInQueue(ctx, req.Key.Namespace, func(ctx context.Context) {
+		res, err = s.delete(ctx, user, req)
+	})
+	if runErr != nil {
+		return HandleQueueError(runErr, func(e *resourcepb.ErrorResult) *resourcepb.DeleteResponse {
+			return &resourcepb.DeleteResponse{Error: e}
+		})
+	}
+
+	return res, err
+}
+
+func (s *server) delete(ctx context.Context, user claims.AuthInfo, req *resourcepb.DeleteRequest) (*resourcepb.DeleteResponse, error) {
+	rsp := &resourcepb.DeleteResponse{}
 	latest := s.backend.ReadResource(ctx, &resourcepb.ReadRequest{
 		Key: req.Key,
 	})
@@ -744,6 +828,23 @@ func (s *server) Read(ctx context.Context, req *resourcepb.ReadRequest) (*resour
 		return &resourcepb.ReadResponse{Error: NewBadRequestError("missing resource")}, nil
 	}
 
+	var (
+		res *resourcepb.ReadResponse
+		err error
+	)
+	runErr := s.runInQueue(ctx, req.Key.Namespace, func(ctx context.Context) {
+		res, err = s.read(ctx, user, req)
+	})
+	if runErr != nil {
+		return HandleQueueError(runErr, func(e *resourcepb.ErrorResult) *resourcepb.ReadResponse {
+			return &resourcepb.ReadResponse{Error: e}
+		})
+	}
+
+	return res, err
+}
+
+func (s *server) read(ctx context.Context, user claims.AuthInfo, req *resourcepb.ReadRequest) (*resourcepb.ReadResponse, error) {
 	rsp := s.backend.ReadResource(ctx, req)
 	if rsp.Error != nil && rsp.Error.Code == http.StatusNotFound {
 		return &resourcepb.ReadResponse{Error: rsp.Error}, nil
@@ -1237,3 +1338,41 @@ func (s *server) GetBlob(ctx context.Context, req *resourcepb.GetBlobRequest) (*
 	}
 	return rsp, nil
 }
+
+func (s *server) runInQueue(ctx context.Context, tenantID string, runnable func(ctx context.Context)) error {
+	boff := backoff.New(ctx, backoff.Config{
+		MinBackoff: DefaultMinBackoff,
+		MaxBackoff: DefaultMaxBackoff,
+		MaxRetries: DefaultMaxRetries,
+	})
+
+	var (
+		wg  sync.WaitGroup
+		err error
+	)
+	wg.Add(1)
+	wrapped := func(ctx context.Context) {
+		runnable(ctx)
+		wg.Done()
+	}
+	for boff.Ongoing() {
+		err = s.queue.Enqueue(ctx, tenantID, wrapped)
+		if err == nil {
+			break
+		}
+		s.log.Warn("failed to enqueue runnable, retrying",
+			"maxRetries", DefaultMaxRetries,
+			"tenantID", tenantID,
+			"error", err)
+		boff.Wait()
+	}
+	if err != nil {
+		s.log.Error("failed to enqueue runnable",
+			"maxRetries", DefaultMaxRetries,
+			"tenantID", tenantID,
+			"error", err)
+		return fmt.Errorf("failed to enqueue runnable for tenant %s: %w", tenantID, err)
+	}
+	wg.Wait()
+	return nil
+}

pkg/storage/unified/sql/server.go

@@ -1,6 +1,7 @@
 package sql
 
 import (
+	"context"
 	"os"
 	"strings"
 
@@ -8,6 +9,7 @@ import (
 	"go.opentelemetry.io/otel/trace"
 
 	"github.com/grafana/authlib/types"
+	"github.com/grafana/dskit/services"
 
 	infraDB "github.com/grafana/grafana/pkg/infra/db"
 	"github.com/grafana/grafana/pkg/services/featuremgmt"
@@ -17,70 +19,85 @@ import (
 	"github.com/grafana/grafana/pkg/storage/unified/sql/db/dbimpl"
 )
 
+type QOSEnqueueDequeuer interface {
+	services.Service
+	Enqueue(ctx context.Context, tenantID string, runnable func(ctx context.Context)) error
+	Dequeue(ctx context.Context) (func(ctx context.Context), error)
+}
+
+// ServerOptions contains the options for creating a new ResourceServer
+type ServerOptions struct {
+	DB             infraDB.DB
+	Cfg            *setting.Cfg
+	Tracer         trace.Tracer
+	Reg            prometheus.Registerer
+	AccessClient   types.AccessClient
+	SearchOptions  resource.SearchOptions
+	StorageMetrics *resource.StorageMetrics
+	IndexMetrics   *resource.BleveIndexMetrics
+	Features       featuremgmt.FeatureToggles
+	QOSQueue       QOSEnqueueDequeuer
+}
+
 // Creates a new ResourceServer
-func NewResourceServer(db infraDB.DB, cfg *setting.Cfg,
-	tracer trace.Tracer, reg prometheus.Registerer, ac types.AccessClient,
-	searchOptions resource.SearchOptions, storageMetrics *resource.StorageMetrics,
-	indexMetrics *resource.BleveIndexMetrics, features featuremgmt.FeatureToggles) (resource.ResourceServer, error) {
-	apiserverCfg := cfg.SectionWithEnvOverrides("grafana-apiserver")
-	opts := resource.ResourceServerOptions{
-		Tracer: tracer,
+func NewResourceServer(
+	opts ServerOptions,
+) (resource.ResourceServer, error) {
+	apiserverCfg := opts.Cfg.SectionWithEnvOverrides("grafana-apiserver")
+	serverOptions := resource.ResourceServerOptions{
+		Tracer: opts.Tracer,
 		Blob: resource.BlobConfig{
 			URL: apiserverCfg.Key("blob_url").MustString(""),
 		},
-		Reg: reg,
+		Reg: opts.Reg,
 	}
-	if ac != nil {
-		opts.AccessClient = resource.NewAuthzLimitedClient(ac, resource.AuthzOptions{Tracer: tracer, Registry: reg})
+	if opts.AccessClient != nil {
+		serverOptions.AccessClient = resource.NewAuthzLimitedClient(opts.AccessClient, resource.AuthzOptions{Tracer: opts.Tracer, Registry: opts.Reg})
 	}
 	// Support local file blob
-	if strings.HasPrefix(opts.Blob.URL, "./data/") {
-		dir := strings.Replace(opts.Blob.URL, "./data", cfg.DataPath, 1)
+	if strings.HasPrefix(serverOptions.Blob.URL, "./data/") {
+		dir := strings.Replace(serverOptions.Blob.URL, "./data", opts.Cfg.DataPath, 1)
 		err := os.MkdirAll(dir, 0700)
 		if err != nil {
 			return nil, err
 		}
-		opts.Blob.URL = "file:///" + dir
+		serverOptions.Blob.URL = "file:///" + dir
 	}
 
 	// This is mostly for testing, being able to influence when we paginate
 	// based on the page size during tests.
-	unifiedStorageCfg := cfg.SectionWithEnvOverrides("unified_storage")
+	unifiedStorageCfg := opts.Cfg.SectionWithEnvOverrides("unified_storage")
 	maxPageSizeBytes := unifiedStorageCfg.Key("max_page_size_bytes")
-	opts.MaxPageSizeBytes = maxPageSizeBytes.MustInt(0)
+	serverOptions.MaxPageSizeBytes = maxPageSizeBytes.MustInt(0)
 
-	eDB, err := dbimpl.ProvideResourceDB(db, cfg, tracer)
+	eDB, err := dbimpl.ProvideResourceDB(opts.DB, opts.Cfg, opts.Tracer)
 	if err != nil {
 		return nil, err
 	}
 
-	isHA := isHighAvailabilityEnabled(cfg.SectionWithEnvOverrides("database"),
-		cfg.SectionWithEnvOverrides("resource_api"))
-	withPruner := features.IsEnabledGlobally(featuremgmt.FlagUnifiedStorageHistoryPruner)
+	isHA := isHighAvailabilityEnabled(opts.Cfg.SectionWithEnvOverrides("database"),
+		opts.Cfg.SectionWithEnvOverrides("resource_api"))
+	withPruner := opts.Features.IsEnabledGlobally(featuremgmt.FlagUnifiedStorageHistoryPruner)
 
 	store, err := NewBackend(BackendOptions{
 		DBProvider:     eDB,
-		Tracer:         tracer,
-		Reg:            reg,
+		Tracer:         opts.Tracer,
+		Reg:            opts.Reg,
 		IsHA:           isHA,
 		withPruner:     withPruner,
-		storageMetrics: storageMetrics,
+		storageMetrics: opts.StorageMetrics,
 	})
 	if err != nil {
 		return nil, err
 	}
-	opts.Backend = store
-	opts.Diagnostics = store
-	opts.Lifecycle = store
-	opts.Search = searchOptions
-	opts.IndexMetrics = indexMetrics
-
-	rs, err := resource.NewResourceServer(opts)
-	if err != nil {
-		return nil, err
-	}
+	serverOptions.Backend = store
+	serverOptions.Diagnostics = store
+	serverOptions.Lifecycle = store
+	serverOptions.Search = opts.SearchOptions
+	serverOptions.IndexMetrics = opts.IndexMetrics
+	serverOptions.QOSQueue = opts.QOSQueue
 
-	return rs, nil
+	return resource.NewResourceServer(serverOptions)
 }
 
 // isHighAvailabilityEnabled determines if high availability mode should

pkg/storage/unified/sql/service.go

@@ -34,6 +34,7 @@ import (
 	"github.com/grafana/grafana/pkg/storage/unified/resource/grpc"
 	"github.com/grafana/grafana/pkg/storage/unified/resourcepb"
 	"github.com/grafana/grafana/pkg/storage/unified/search"
+	"github.com/grafana/grafana/pkg/util/scheduler"
 )
 
 var (
@@ -50,6 +51,11 @@ type UnifiedStorageGrpcService interface {
 type service struct {
 	*services.BasicService
 
+	// Subservices manager
+	subservices        *services.Manager
+	subservicesWatcher *services.FailureWatcher
+	hasSubservices     bool
+
 	cfg       *setting.Cfg
 	features  featuremgmt.FeatureToggles
 	db        infraDB.DB
@@ -71,6 +77,9 @@ type service struct {
 
 	storageRing *ring.Ring
 	lifecycler  *ring.BasicLifecycler
+
+	queue     QOSEnqueueDequeuer
+	scheduler *scheduler.Scheduler
 }
 
 func ProvideUnifiedStorageGrpcService(
@@ -85,6 +94,7 @@ func ProvideUnifiedStorageGrpcService(
 	storageRing *ring.Ring,
 	memberlistKVConfig kv.Config,
 ) (UnifiedStorageGrpcService, error) {
+	var err error
 	tracer := otel.Tracer("unified-storage")
 
 	// FIXME: This is a temporary solution while we are migrating to the new authn interceptor
@@ -95,20 +105,22 @@ func ProvideUnifiedStorageGrpcService(
 	})
 
 	s := &service{
-		cfg:            cfg,
-		features:       features,
-		stopCh:         make(chan struct{}),
-		authenticator:  authn,
-		tracing:        tracer,
-		db:             db,
-		log:            log,
-		reg:            reg,
-		docBuilders:    docBuilders,
-		storageMetrics: storageMetrics,
-		indexMetrics:   indexMetrics,
-		storageRing:    storageRing,
+		cfg:                cfg,
+		features:           features,
+		stopCh:             make(chan struct{}),
+		authenticator:      authn,
+		tracing:            tracer,
+		db:                 db,
+		log:                log,
+		reg:                reg,
+		docBuilders:        docBuilders,
+		storageMetrics:     storageMetrics,
+		indexMetrics:       indexMetrics,
+		storageRing:        storageRing,
+		subservicesWatcher: services.NewFailureWatcher(),
 	}
 
+	subservices := []services.Service{}
 	if cfg.EnableSharding {
 		ringStore, err := kv.NewClient(
 			memberlistKVConfig,
@@ -143,15 +155,50 @@ func ProvideUnifiedStorageGrpcService(
 		if err != nil {
 			return nil, fmt.Errorf("failed to initialize storage-ring lifecycler: %s", err)
 		}
+		subservices = append(subservices, s.lifecycler)
+	}
+
+	if cfg.QOSEnabled {
+		qosReg := prometheus.WrapRegistererWithPrefix("resource_server_qos_", reg)
+		queue := scheduler.NewQueue(&scheduler.QueueOptions{
+			MaxSizePerTenant: cfg.QOSMaxSizePerTenant,
+			Registerer:       qosReg,
+		})
+		scheduler, err := scheduler.NewScheduler(queue, &scheduler.Config{
+			NumWorkers: cfg.QOSNumberWorker,
+			Logger:     log,
+		})
+		if err != nil {
+			return nil, fmt.Errorf("failed to create qos scheduler: %s", err)
+		}
+
+		s.queue = queue
+		s.scheduler = scheduler
+		subservices = append(subservices, s.queue, s.scheduler)
+	}
+
+	if len(subservices) > 0 {
+		s.hasSubservices = true
+		s.subservices, err = services.NewManager(subservices...)
+		if err != nil {
+			return nil, fmt.Errorf("failed to create subservices manager: %w", err)
+		}
 	}
 
 	// This will be used when running as a dskit service
-	s.BasicService = services.NewBasicService(s.start, s.running, s.stopping).WithName(modules.StorageServer)
+	s.BasicService = services.NewBasicService(s.starting, s.running, s.stopping).WithName(modules.StorageServer)
 
 	return s, nil
 }
 
-func (s *service) start(ctx context.Context) error {
+func (s *service) starting(ctx context.Context) error {
+	if s.hasSubservices {
+		s.subservicesWatcher.WatchManager(s.subservices)
+		if err := services.StartManagerAndAwaitHealthy(ctx, s.subservices); err != nil {
+			return fmt.Errorf("failed to start subservices: %w", err)
+		}
+	}
+
 	authzClient, err := authz.ProvideStandaloneAuthZClient(s.cfg, s.features, s.tracing)
 	if err != nil {
 		return err
@@ -162,7 +209,19 @@ func (s *service) start(ctx context.Context) error {
 		return err
 	}
 
-	server, err := NewResourceServer(s.db, s.cfg, s.tracing, s.reg, authzClient, searchOptions, s.storageMetrics, s.indexMetrics, s.features)
+	serverOptions := ServerOptions{
+		DB:             s.db,
+		Cfg:            s.cfg,
+		Tracer:         s.tracing,
+		Reg:            s.reg,
+		AccessClient:   authzClient,
+		SearchOptions:  searchOptions,
+		StorageMetrics: s.storageMetrics,
+		IndexMetrics:   s.indexMetrics,
+		Features:       s.features,
+		QOSQueue:       s.queue,
+	}
+	server, err := NewResourceServer(serverOptions)
 	if err != nil {
 		return err
 	}
@@ -192,11 +251,6 @@ func (s *service) start(ctx context.Context) error {
 	}
 
 	if s.cfg.EnableSharding {
-		err = s.lifecycler.StartAsync(ctx)
-		if err != nil {
-			return fmt.Errorf("failed to start the lifecycler: %s", err)
-		}
-
 		s.log.Info("waiting until resource server is JOINING in the ring")
 		lfcCtx, cancel := context.WithTimeout(context.Background(), time.Second*10)
 		defer cancel()
@@ -231,15 +285,27 @@ func (s *service) GetAddress() string {
 func (s *service) running(ctx context.Context) error {
 	select {
 	case err := <-s.stoppedCh:
-		if err != nil {
+		if err != nil && !errors.Is(err, context.Canceled) {
 			return err
 		}
+	case err := <-s.subservicesWatcher.Chan():
+		return fmt.Errorf("subservice failure: %w", err)
 	case <-ctx.Done():
 		close(s.stopCh)
 	}
 	return nil
 }
 
+func (s *service) stopping(_ error) error {
+	if s.hasSubservices {
+		err := services.StopManagerAndAwaitStopped(context.Background(), s.subservices)
+		if err != nil {
+			return fmt.Errorf("failed to stop subservices: %w", err)
+		}
+	}
+	return nil
+}
+
 type authenticatorWithFallback struct {
 	authenticator func(ctx context.Context) (context.Context, error)
 	fallback      func(ctx context.Context) (context.Context, error)
@@ -309,14 +375,6 @@ func NewAuthenticatorWithFallback(cfg *setting.Cfg, reg prometheus.Registerer, t
 	}
 }
 
-func (s *service) stopping(err error) error {
-	if err != nil && !errors.Is(err, context.Canceled) {
-		s.log.Error("stopping unified storage grpc service", "error", err)
-		return err
-	}
-	return nil
-}
-
 func toLifecyclerConfig(cfg *setting.Cfg, logger log.Logger) (ring.BasicLifecyclerConfig, error) {
 	instanceAddr, err := ring.GetInstanceAddr(cfg.MemberlistBindAddr, netutil.PrivateNetworkInterfacesWithFallback([]string{"eth0", "en0"}, logger), logger, true)
 	if err != nil {

pkg/util/scheduler/queue.go

@@ -9,6 +9,8 @@ import (
 	"github.com/grafana/dskit/services"
 	"github.com/prometheus/client_golang/prometheus"
 	"github.com/prometheus/client_golang/prometheus/promauto"
+
+	"github.com/grafana/grafana/pkg/infra/log"
 )
 
 const (
@@ -82,6 +84,8 @@ func NewNoopQueue() *NoopQueue {
 type Queue struct {
 	services.Service
 
+	logger log.Logger
+
 	enqueueChan           chan enqueueRequest
 	dequeueChan           chan dequeueRequest
 	lenChan               chan lenRequest
@@ -108,6 +112,7 @@ type Queue struct {
 type QueueOptions struct {
 	MaxSizePerTenant int
 	Registerer       prometheus.Registerer
+	Logger           log.Logger
 }
 
 // NewQueue creates a new Queue and starts its dispatcher goroutine.
@@ -116,7 +121,13 @@ func NewQueue(opts *QueueOptions) *Queue {
 		opts.MaxSizePerTenant = DefaultMaxSizePerTenant
 	}
 
+	if opts.Logger == nil {
+		opts.Logger = log.NewNopLogger()
+	}
+
 	q := &Queue{
+		logger: opts.Logger,
+
 		enqueueChan:           make(chan enqueueRequest),
 		dequeueChan:           make(chan dequeueRequest),
 		lenChan:               make(chan lenRequest),
@@ -226,6 +237,8 @@ func (q *Queue) handleLenRequest(req lenRequest) {
 func (q *Queue) dispatcherLoop(ctx context.Context) error {
 	defer close(q.dispatcherStoppedChan)
 
+	q.logger.Info("queue running", "maxSizePerTenant", q.maxSizePerTenant)
+
 	for {
 		q.scheduleRoundRobin()
 
@@ -275,7 +288,6 @@ func (q *Queue) Enqueue(ctx context.Context, tenantID string, runnable func(ctx
 	select {
 	case q.enqueueChan <- req:
 		err = <-respChan
-		q.enqueueDuration.Observe(time.Since(start).Seconds())
 	case <-q.dispatcherStoppedChan:
 		q.discardedRequests.WithLabelValues(tenantID, "dispatcher_stopped").Inc()
 		err = ErrQueueClosed
@@ -283,6 +295,7 @@ func (q *Queue) Enqueue(ctx context.Context, tenantID string, runnable func(ctx
 		q.discardedRequests.WithLabelValues(tenantID, "context_canceled").Inc()
 		err = ctx.Err()
 	}
+	q.enqueueDuration.Observe(time.Since(start).Seconds())
 
 	return err
 }
@@ -352,6 +365,8 @@ func (q *Queue) ActiveTenantsLen() int {
 }
 
 func (q *Queue) stopping(_ error) error {
+	q.logger.Info("queue stopping")
+
 	q.queueLength.Reset()
 	q.discardedRequests.Reset()
 	for _, tq := range q.tenantQueues {
@@ -359,5 +374,7 @@ func (q *Queue) stopping(_ error) error {
 	}
 	q.activeTenants.Init()
 	q.pendingDequeueRequests.Init()
+
+	q.logger.Info("queue stopped")
 	return nil
 }

pkg/util/scheduler/queue_test.go

@@ -11,6 +11,7 @@ import (
 	"time"
 
 	"github.com/grafana/dskit/services"
+	"github.com/grafana/grafana/pkg/infra/log"
 	"github.com/prometheus/client_golang/prometheus"
 	"github.com/stretchr/testify/require"
 )
@@ -25,6 +26,9 @@ func QueueOptionsWithDefaults(opts *QueueOptions) *QueueOptions {
 	if opts.Registerer == nil {
 		opts.Registerer = prometheus.NewRegistry()
 	}
+	if opts.Logger == nil {
+		opts.Logger = log.New("qos.test")
+	}
 	return opts
 }
 

pkg/util/scheduler/scheduler_test.go

@@ -2,6 +2,7 @@ package scheduler
 
 import (
 	"context"
+	"fmt"
 	"sync"
 	"sync/atomic"
 	"testing"
@@ -130,16 +131,16 @@ func TestScheduler(t *testing.T) {
 	t.Run("ProcessItems", func(t *testing.T) {
 		t.Parallel()
 
-		q := NewQueue(QueueOptionsWithDefaults(nil))
+		q := NewQueue(QueueOptionsWithDefaults(&QueueOptions{MaxSizePerTenant: 1000}))
 		require.NoError(t, services.StartAndAwaitRunning(context.Background(), q))
 
-		const itemCount = 10
+		const itemCount = 1000
 		var processed sync.Map
 		var wg sync.WaitGroup
 		wg.Add(itemCount)
 
 		scheduler, err := NewScheduler(q, &Config{
-			NumWorkers: 2,
+			NumWorkers: 10,
 			MaxBackoff: 100 * time.Millisecond,
 			Logger:     log.New("qos.test"),
 		})
@@ -148,8 +149,11 @@ func TestScheduler(t *testing.T) {
 
 		for i := 0; i < itemCount; i++ {
 			itemID := i
-			require.NoError(t, q.Enqueue(context.Background(), "tenant-1", func(_ context.Context) {
+			tenantIndex := itemID % 10
+			tenantID := fmt.Sprintf("tenant-%d", tenantIndex)
+			require.NoError(t, q.Enqueue(context.Background(), tenantID, func(_ context.Context) {
 				processed.Store(itemID, true)
+				time.Sleep(10 * time.Millisecond)
 				wg.Done()
 			}))
 		}