Alerting: Add enablement flag for recording rules (#92032)

* Add enablement flag

* Disable if toggle not enabled

conf/defaults.ini

@@ -1438,6 +1438,9 @@ max_age =
 max_annotations_to_keep =
 
 [recording_rules]
+# Enable recording rules. You must provide write credentials below.
+enabled = false
+
 # Target URL (including write path) for recording rules.
 url =
 

conf/sample.ini

@@ -1432,6 +1432,9 @@ max_annotations_to_keep =
 
 #################################### Recording Rules #####################
 [recording_rules]
+# Enable recording rules. You must provide write credentials below.
+enabled = false
+
 # Target URL (including write path) for recording rules.
 url =
 

pkg/services/ngalert/ngalert.go

@@ -346,6 +346,10 @@ func (ng *AlertNG) init() error {
 	evalFactory := eval.NewEvaluatorFactory(ng.Cfg.UnifiedAlerting, ng.DataSourceCache, ng.ExpressionService)
 	conditionValidator := eval.NewConditionValidator(ng.DataSourceCache, ng.ExpressionService, ng.pluginsStore)
 
+	if !ng.FeatureToggles.IsEnabled(initCtx, featuremgmt.FlagGrafanaManagedRecordingRules) {
+		// Force-disable the feature if the feature toggle is not on - sets us up for feature toggle removal.
+		ng.Cfg.UnifiedAlerting.RecordingRules.Enabled = false
+	}
 	recordingWriter, err := createRecordingWriter(ng.FeatureToggles, ng.Cfg.UnifiedAlerting.RecordingRules, ng.httpClientProvider, clk, ng.Metrics.GetRemoteWriterMetrics())
 	if err != nil {
 		return fmt.Errorf("failed to initialize recording writer: %w", err)
@@ -362,7 +366,7 @@ func (ng *AlertNG) init() error {
 		AppURL:               appUrl,
 		EvaluatorFactory:     evalFactory,
 		RuleStore:            ng.store,
-		FeatureToggles:       ng.FeatureToggles,
+		RecordingRulesCfg:    ng.Cfg.UnifiedAlerting.RecordingRules,
 		Metrics:              ng.Metrics.GetSchedulerMetrics(),
 		AlertSender:          alertsRouter,
 		Tracer:               ng.tracer,
@@ -667,7 +671,7 @@ func createRemoteAlertmanager(cfg remote.AlertmanagerConfig, kvstore kvstore.KVS
 func createRecordingWriter(featureToggles featuremgmt.FeatureToggles, settings setting.RecordingRuleSettings, httpClientProvider httpclient.Provider, clock clock.Clock, m *metrics.RemoteWriter) (schedule.RecordingWriter, error) {
 	logger := log.New("ngalert.writer")
 
-	if featureToggles.IsEnabledGlobally(featuremgmt.FlagGrafanaManagedRecordingRules) {
+	if settings.Enabled {
 		return writer.NewPrometheusWriter(settings, httpClientProvider, clock, logger, m)
 	}
 

pkg/services/ngalert/schedule/alert_rule.go

@@ -16,7 +16,6 @@ import (
 	"github.com/grafana/grafana/pkg/infra/log"
 	"github.com/grafana/grafana/pkg/infra/tracing"
 	"github.com/grafana/grafana/pkg/services/datasources"
-	"github.com/grafana/grafana/pkg/services/featuremgmt"
 	"github.com/grafana/grafana/pkg/services/ngalert/api/tooling/definitions"
 	"github.com/grafana/grafana/pkg/services/ngalert/eval"
 	"github.com/grafana/grafana/pkg/services/ngalert/metrics"
@@ -24,6 +23,7 @@ import (
 	"github.com/grafana/grafana/pkg/services/ngalert/state"
 	"github.com/grafana/grafana/pkg/services/org"
 	"github.com/grafana/grafana/pkg/services/user"
+	"github.com/grafana/grafana/pkg/setting"
 	"github.com/grafana/grafana/pkg/util"
 )
 
@@ -57,7 +57,7 @@ func newRuleFactory(
 	evalFactory eval.EvaluatorFactory,
 	ruleProvider ruleProvider,
 	clock clock.Clock,
-	featureToggles featuremgmt.FeatureToggles,
+	rrCfg setting.RecordingRuleSettings,
 	met *metrics.Scheduler,
 	logger log.Logger,
 	tracer tracing.Tracer,
@@ -73,7 +73,7 @@ func newRuleFactory(
 				maxAttempts,
 				clock,
 				evalFactory,
-				featureToggles,
+				rrCfg,
 				logger,
 				met,
 				tracer,

pkg/services/ngalert/schedule/alert_rule_test.go

@@ -827,7 +827,7 @@ func TestRuleRoutine(t *testing.T) {
 }
 
 func ruleFactoryFromScheduler(sch *schedule) ruleFactory {
-	return newRuleFactory(sch.appURL, sch.disableGrafanaFolder, sch.maxAttempts, sch.alertsSender, sch.stateManager, sch.evaluatorFactory, &sch.schedulableAlertRules, sch.clock, sch.featureToggles, sch.metrics, sch.log, sch.tracer, sch.recordingWriter, sch.evalAppliedFunc, sch.stopAppliedFunc)
+	return newRuleFactory(sch.appURL, sch.disableGrafanaFolder, sch.maxAttempts, sch.alertsSender, sch.stateManager, sch.evaluatorFactory, &sch.schedulableAlertRules, sch.clock, sch.rrCfg, sch.metrics, sch.log, sch.tracer, sch.recordingWriter, sch.evalAppliedFunc, sch.stopAppliedFunc)
 }
 
 func stateForRule(rule *models.AlertRule, ts time.Time, evalState eval.State) *state.State {

pkg/services/ngalert/schedule/recording_rule.go

@@ -15,10 +15,10 @@ import (
 
 	"github.com/grafana/grafana/pkg/infra/log"
 	"github.com/grafana/grafana/pkg/infra/tracing"
-	"github.com/grafana/grafana/pkg/services/featuremgmt"
 	"github.com/grafana/grafana/pkg/services/ngalert/eval"
 	"github.com/grafana/grafana/pkg/services/ngalert/metrics"
 	ngmodels "github.com/grafana/grafana/pkg/services/ngalert/models"
+	"github.com/grafana/grafana/pkg/setting"
 	"github.com/grafana/grafana/pkg/util"
 )
 
@@ -42,10 +42,10 @@ type recordingRule struct {
 
 	maxAttempts int64
 
-	clock          clock.Clock
-	evalFactory    eval.EvaluatorFactory
-	featureToggles featuremgmt.FeatureToggles
-	writer         RecordingWriter
+	clock       clock.Clock
+	evalFactory eval.EvaluatorFactory
+	cfg         setting.RecordingRuleSettings
+	writer      RecordingWriter
 
 	// Event hooks that are only used in tests.
 	evalAppliedHook evalAppliedFunc
@@ -56,7 +56,7 @@ type recordingRule struct {
 	tracer  tracing.Tracer
 }
 
-func newRecordingRule(parent context.Context, key ngmodels.AlertRuleKey, maxAttempts int64, clock clock.Clock, evalFactory eval.EvaluatorFactory, ft featuremgmt.FeatureToggles, logger log.Logger, metrics *metrics.Scheduler, tracer tracing.Tracer, writer RecordingWriter, evalAppliedHook evalAppliedFunc, stopAppliedHook stopAppliedFunc) *recordingRule {
+func newRecordingRule(parent context.Context, key ngmodels.AlertRuleKey, maxAttempts int64, clock clock.Clock, evalFactory eval.EvaluatorFactory, cfg setting.RecordingRuleSettings, logger log.Logger, metrics *metrics.Scheduler, tracer tracing.Tracer, writer RecordingWriter, evalAppliedHook evalAppliedFunc, stopAppliedHook stopAppliedFunc) *recordingRule {
 	ctx, stop := util.WithCancelCause(ngmodels.WithRuleKey(parent, key))
 	return &recordingRule{
 		key:                 key,
@@ -69,7 +69,7 @@ func newRecordingRule(parent context.Context, key ngmodels.AlertRuleKey, maxAtte
 		evaluationDuration:  atomic.NewDuration(0),
 		clock:               clock,
 		evalFactory:         evalFactory,
-		featureToggles:      ft,
+		cfg:                 cfg,
 		maxAttempts:         maxAttempts,
 		evalAppliedHook:     evalAppliedHook,
 		stopAppliedHook:     stopAppliedHook,
@@ -132,8 +132,8 @@ func (r *recordingRule) Run() error {
 				r.logger.Debug("Evaluation channel has been closed. Exiting")
 				return nil
 			}
-			if !r.featureToggles.IsEnabled(ctx, featuremgmt.FlagGrafanaManagedRecordingRules) {
-				r.logger.Warn("Recording rule scheduled but toggle is not enabled. Skipping")
+			if !r.cfg.Enabled {
+				r.logger.Warn("Recording rule scheduled but subsystem is not enabled. Skipping")
 				return nil
 			}
 			// TODO: Skipping the "evalRunning" guard that the alert rule routine does, because it seems to be dead code and impossible to hit.

pkg/services/ngalert/schedule/recording_rule_test.go

@@ -19,10 +19,10 @@ import (
 	"github.com/grafana/grafana-plugin-sdk-go/backend/httpclient"
 	"github.com/grafana/grafana/pkg/expr"
 	"github.com/grafana/grafana/pkg/infra/log"
-	"github.com/grafana/grafana/pkg/services/featuremgmt"
 	"github.com/grafana/grafana/pkg/services/ngalert/metrics"
 	models "github.com/grafana/grafana/pkg/services/ngalert/models"
 	"github.com/grafana/grafana/pkg/services/ngalert/writer"
+	"github.com/grafana/grafana/pkg/setting"
 	"github.com/grafana/grafana/pkg/util"
 )
 
@@ -153,8 +153,10 @@ func TestRecordingRule(t *testing.T) {
 }
 
 func blankRecordingRuleForTests(ctx context.Context) *recordingRule {
-	ft := featuremgmt.WithFeatures(featuremgmt.FlagGrafanaManagedRecordingRules)
-	return newRecordingRule(context.Background(), models.AlertRuleKey{}, 0, nil, nil, ft, log.NewNopLogger(), nil, nil, writer.FakeWriter{}, nil, nil)
+	st := setting.RecordingRuleSettings{
+		Enabled: true,
+	}
+	return newRecordingRule(context.Background(), models.AlertRuleKey{}, 0, nil, nil, st, log.NewNopLogger(), nil, nil, writer.FakeWriter{}, nil, nil)
 }
 
 func TestRecordingRule_Integration(t *testing.T) {

pkg/services/ngalert/schedule/schedule.go

@@ -15,12 +15,12 @@ import (
 
 	"github.com/grafana/grafana/pkg/infra/log"
 	"github.com/grafana/grafana/pkg/infra/tracing"
-	"github.com/grafana/grafana/pkg/services/featuremgmt"
 	"github.com/grafana/grafana/pkg/services/ngalert/api/tooling/definitions"
 	"github.com/grafana/grafana/pkg/services/ngalert/eval"
 	"github.com/grafana/grafana/pkg/services/ngalert/metrics"
 	ngmodels "github.com/grafana/grafana/pkg/services/ngalert/models"
 	"github.com/grafana/grafana/pkg/services/ngalert/state"
+	"github.com/grafana/grafana/pkg/setting"
 	"github.com/grafana/grafana/pkg/util/ticker"
 )
 
@@ -84,7 +84,7 @@ type schedule struct {
 	appURL               *url.URL
 	disableGrafanaFolder bool
 	jitterEvaluations    JitterStrategy
-	featureToggles       featuremgmt.FeatureToggles
+	rrCfg                setting.RecordingRuleSettings
 
 	metrics *metrics.Scheduler
 	// lastUpdatedMetricsForOrgsAndGroups contains AlertRuleGroupKeyWithFolderFullpaths that
@@ -113,7 +113,7 @@ type SchedulerCfg struct {
 	C                    clock.Clock
 	MinRuleInterval      time.Duration
 	DisableGrafanaFolder bool
-	FeatureToggles       featuremgmt.FeatureToggles
+	RecordingRulesCfg    setting.RecordingRuleSettings
 	AppURL               *url.URL
 	JitterEvaluations    JitterStrategy
 	EvaluatorFactory     eval.EvaluatorFactory
@@ -146,7 +146,7 @@ func NewScheduler(cfg SchedulerCfg, stateManager *state.Manager) *schedule {
 		appURL:                             cfg.AppURL,
 		disableGrafanaFolder:               cfg.DisableGrafanaFolder,
 		jitterEvaluations:                  cfg.JitterEvaluations,
-		featureToggles:                     cfg.FeatureToggles,
+		rrCfg:                              cfg.RecordingRulesCfg,
 		stateManager:                       stateManager,
 		minRuleInterval:                    cfg.MinRuleInterval,
 		schedulableAlertRules:              alertRulesRegistry{rules: make(map[ngmodels.AlertRuleKey]*ngmodels.AlertRule)},
@@ -266,7 +266,7 @@ func (sch *schedule) processTick(ctx context.Context, dispatcherGroup *errgroup.
 		sch.evaluatorFactory,
 		&sch.schedulableAlertRules,
 		sch.clock,
-		sch.featureToggles,
+		sch.rrCfg,
 		sch.metrics,
 		sch.log,
 		sch.tracer,

pkg/services/ngalert/schedule/schedule_unit_test.go

@@ -65,18 +65,21 @@ func TestProcessTicks(t *testing.T) {
 
 	cacheServ := &datasources.FakeCacheService{}
 	evaluator := eval.NewEvaluatorFactory(setting.UnifiedAlertingSettings{}, cacheServ, expr.ProvideService(&setting.Cfg{ExpressionsEnabled: true}, nil, nil, featuremgmt.WithFeatures(), nil, tracing.InitializeTracerForTest()))
+	rrSet := setting.RecordingRuleSettings{
+		Enabled: true,
+	}
 
 	schedCfg := SchedulerCfg{
-		BaseInterval:     cfg.BaseInterval,
-		C:                mockedClock,
-		AppURL:           appUrl,
-		EvaluatorFactory: evaluator,
-		RuleStore:        ruleStore,
-		Metrics:          testMetrics.GetSchedulerMetrics(),
-		AlertSender:      notifier,
-		FeatureToggles:   featuremgmt.WithFeatures(featuremgmt.FlagGrafanaManagedRecordingRules),
-		Tracer:           testTracer,
-		Log:              log.New("ngalert.scheduler"),
+		BaseInterval:      cfg.BaseInterval,
+		C:                 mockedClock,
+		AppURL:            appUrl,
+		EvaluatorFactory:  evaluator,
+		RuleStore:         ruleStore,
+		Metrics:           testMetrics.GetSchedulerMetrics(),
+		AlertSender:       notifier,
+		RecordingRulesCfg: rrSet,
+		Tracer:            testTracer,
+		Log:               log.New("ngalert.scheduler"),
 	}
 	managerCfg := state.ManagerCfg{
 		Metrics:       testMetrics.GetStateMetrics(),
@@ -854,23 +857,26 @@ func setupScheduler(t *testing.T, rs *fakeRulesStore, is *state.FakeInstanceStor
 	cfg := setting.UnifiedAlertingSettings{
 		BaseInterval: time.Second,
 		MaxAttempts:  1,
+		RecordingRules: setting.RecordingRuleSettings{
+			Enabled: true,
+		},
 	}
 
 	fakeRecordingWriter := writer.FakeWriter{}
 
 	schedCfg := SchedulerCfg{
-		BaseInterval:     cfg.BaseInterval,
-		MaxAttempts:      cfg.MaxAttempts,
-		C:                mockedClock,
-		AppURL:           appUrl,
-		EvaluatorFactory: evaluator,
-		RuleStore:        rs,
-		FeatureToggles:   featuremgmt.WithFeatures(featuremgmt.FlagGrafanaManagedRecordingRules),
-		Metrics:          m.GetSchedulerMetrics(),
-		AlertSender:      senderMock,
-		Tracer:           testTracer,
-		Log:              log.New("ngalert.scheduler"),
-		RecordingWriter:  fakeRecordingWriter,
+		BaseInterval:      cfg.BaseInterval,
+		MaxAttempts:       cfg.MaxAttempts,
+		C:                 mockedClock,
+		AppURL:            appUrl,
+		EvaluatorFactory:  evaluator,
+		RuleStore:         rs,
+		RecordingRulesCfg: cfg.RecordingRules,
+		Metrics:           m.GetSchedulerMetrics(),
+		AlertSender:       senderMock,
+		Tracer:            testTracer,
+		Log:               log.New("ngalert.scheduler"),
+		RecordingWriter:   fakeRecordingWriter,
 	}
 	managerCfg := state.ManagerCfg{
 		Metrics:                 m.GetStateMetrics(),

pkg/setting/setting_unified_alerting.go

@@ -124,6 +124,7 @@ type UnifiedAlertingSettings struct {
 }
 
 type RecordingRuleSettings struct {
+	Enabled           bool
 	URL               string
 	BasicAuthUsername string
 	BasicAuthPassword string
@@ -421,6 +422,7 @@ func (cfg *Cfg) ReadUnifiedAlertingSettings(iniFile *ini.File) error {
 
 	rr := iniFile.Section("recording_rules")
 	uaCfgRecordingRules := RecordingRuleSettings{
+		Enabled:           rr.Key("enabled").MustBool(false),
 		URL:               rr.Key("url").MustString(""),
 		BasicAuthUsername: rr.Key("basic_auth_username").MustString(""),
 		BasicAuthPassword: rr.Key("basic_auth_password").MustString(""),