checker: unify checker name and checker type

metrics/alertmanager/pd.rules.yml

@@ -172,6 +172,18 @@ groups:
       value: '{{ $value }}'
       summary: PD_no_store_for_making_replica
 
+  - alert: PD_no_store_for_making_replica_v2
+    expr: increase(pd_checker_event_count{type="replica-checker", name="no_target_store"}[1m]) > 0
+    for: 1m
+    labels:
+      env: ENV_LABELS_ENV
+      level: warning
+      expr: increase(pd_checker_event_count{type="replica-checker", name="no_target_store"}[1m]) > 0
+    annotations:
+      description: 'cluster: ENV_LABELS_ENV, type: {{ $labels.type }}, instance: {{ $labels.instance }}, values: {{ $value }}'
+      value: '{{ $value }}'
+      summary: PD_no_store_for_making_replica
+
   - alert: PD_node_restart
     expr: changes(process_start_time_seconds{job="pd"}[5m]) > 0
     for: 1m

metrics/grafana/pd.json

@@ -7980,6 +7980,14 @@
               "legendFormat": "{{name}}",
               "refId": "A",
               "step": 10
+            },
+            {
+              "expr": "sum(rate(pd_checker_event_count{k8s_cluster=\"$k8s_cluster\", tidb_cluster=\"$tidb_cluster\", type=\"rule-checker\"}[1m])) by (name)",
+              "format": "time_series",
+              "intervalFactor": 2,
+              "legendFormat": "{{name}}",
+              "refId": "B",
+              "step": 10
             }
           ],
           "thresholds": [],
@@ -8072,6 +8080,14 @@
               "legendFormat": "{{name}}",
               "refId": "A",
               "step": 10
+            },
+            {
+              "expr": "sum(rate(pd_checker_event_count{k8s_cluster=\"$k8s_cluster\", tidb_cluster=\"$tidb_cluster\", type=\"replica-checker\"}[1m])) by (name)",
+              "format": "time_series",
+              "intervalFactor": 2,
+              "legendFormat": "{{name}}",
+              "refId": "B",
+              "step": 10
             }
           ],
           "thresholds": [],
@@ -8165,6 +8181,14 @@
               "legendFormat": "{{name}}",
               "refId": "A",
               "step": 10
+            },
+            {
+              "expr": "sum(rate(pd_checker_event_count{k8s_cluster=\"$k8s_cluster\", tidb_cluster=\"$tidb_cluster\", type=\"merge-checker\"}[1m])) by (name)",
+              "format": "time_series",
+              "intervalFactor": 2,
+              "legendFormat": "{{name}}",
+              "refId": "B",
+              "step": 10
             }
           ],
           "thresholds": [],

pkg/mcs/scheduling/server/apis/v1/api.go

@@ -33,6 +33,7 @@ import (
 	scheserver "github.com/tikv/pd/pkg/mcs/scheduling/server"
 	mcsutils "github.com/tikv/pd/pkg/mcs/utils"
 	"github.com/tikv/pd/pkg/response"
+	"github.com/tikv/pd/pkg/schedule/config"
 	sche "github.com/tikv/pd/pkg/schedule/core"
 	"github.com/tikv/pd/pkg/schedule/handler"
 	"github.com/tikv/pd/pkg/schedule/operator"
@@ -44,6 +45,7 @@ import (
 	"github.com/tikv/pd/pkg/utils/logutil"
 	"github.com/tikv/pd/pkg/utils/typeutil"
 	"github.com/unrolled/render"
+	"go.uber.org/zap"
 )
 
 // APIPathPrefix is the prefix of the API path.
@@ -573,11 +575,19 @@ func getSchedulerConfigByName(c *gin.Context) {
 	}
 	handlers := sc.GetSchedulerHandlers()
 	name := c.Param("name")
-	if _, ok := handlers[name]; !ok {
+	schedulerName, err := config.ConvertSchedulerStr2Name(name)
+	if err != nil {
+		log.Error("failed to convert scheduler name",
+			zap.String("scheduler", name),
+			errs.ZapError(err))
+		c.String(http.StatusNotFound, errs.ErrSchedulerNotFound.GenWithStackByArgs().Error())
+		return
+	}
+	if _, ok := handlers[schedulerName]; !ok {
 		c.String(http.StatusNotFound, errs.ErrSchedulerNotFound.GenWithStackByArgs().Error())
 		return
 	}
-	isDisabled, err := sc.IsSchedulerDisabled(name)
+	isDisabled, err := sc.IsSchedulerDisabled(schedulerName)
 	if err != nil {
 		c.String(http.StatusInternalServerError, err.Error())
 		return
@@ -587,7 +597,7 @@ func getSchedulerConfigByName(c *gin.Context) {
 		return
 	}
 	c.Request.URL.Path = "/list"
-	handlers[name].ServeHTTP(c.Writer, c.Request)
+	handlers[schedulerName].ServeHTTP(c.Writer, c.Request)
 }
 
 // @Tags     schedulers
@@ -599,7 +609,15 @@ func getSchedulerConfigByName(c *gin.Context) {
 func getDiagnosticResult(c *gin.Context) {
 	handler := c.MustGet(handlerKey).(*handler.Handler)
 	name := c.Param("name")
-	result, err := handler.GetDiagnosticResult(name)
+	schedulerName, err := config.ConvertSchedulerStr2Name(name)
+	if err != nil {
+		log.Error("failed to convert scheduler name",
+			zap.String("scheduler", name),
+			errs.ZapError(err))
+		c.String(http.StatusNotFound, errs.ErrSchedulerNotFound.GenWithStackByArgs().Error())
+		return
+	}
+	result, err := handler.GetDiagnosticResult(schedulerName)
 	if err != nil {
 		c.String(http.StatusInternalServerError, err.Error())
 		return

pkg/mcs/scheduling/server/cluster.go

@@ -308,54 +308,60 @@ func (c *Cluster) updateScheduler() {
 		)
 		// Create the newly added schedulers.
 		for _, scheduler := range latestSchedulersConfig {
+			name, err := sc.ConvertSchedulerStr2Name(scheduler.Type)
+			if err != nil {
+				log.Error("failed to convert scheduler name",
+					zap.String("scheduler", scheduler.Type),
+					errs.ZapError(err))
+				continue
+			}
 			s, err := schedulers.CreateScheduler(
-				scheduler.Type,
+				name,
 				c.coordinator.GetOperatorController(),
 				c.storage,
-				schedulers.ConfigSliceDecoder(scheduler.Type, scheduler.Args),
+				schedulers.ConfigSliceDecoder(name, scheduler.Args),
 				schedulersController.RemoveScheduler,
 			)
 			if err != nil {
 				log.Error("failed to create scheduler",
-					zap.String("scheduler-type", scheduler.Type),
+					zap.String("scheduler", scheduler.Type),
 					zap.Strings("scheduler-args", scheduler.Args),
 					errs.ZapError(err))
 				continue
 			}
-			name := s.GetName()
 			if existed, _ := schedulersController.IsSchedulerExisted(name); existed {
 				log.Info("scheduler has already existed, skip adding it",
-					zap.String("scheduler-name", name),
+					zap.Stringer("scheduler", name),
 					zap.Strings("scheduler-args", scheduler.Args))
 				continue
 			}
 			if err := schedulersController.AddScheduler(s, scheduler.Args...); err != nil {
 				log.Error("failed to add scheduler",
-					zap.String("scheduler-name", name),
+					zap.Stringer("scheduler", name),
 					zap.Strings("scheduler-args", scheduler.Args),
 					errs.ZapError(err))
 				continue
 			}
 			log.Info("add scheduler successfully",
-				zap.String("scheduler-name", name),
+				zap.Stringer("scheduler-name", name),
 				zap.Strings("scheduler-args", scheduler.Args))
 		}
 		// Remove the deleted schedulers.
 		for _, name := range schedulersController.GetSchedulerNames() {
 			scheduler := schedulersController.GetScheduler(name)
 			if slice.AnyOf(latestSchedulersConfig, func(i int) bool {
-				return latestSchedulersConfig[i].Type == scheduler.GetType()
+				return latestSchedulersConfig[i].Type == scheduler.Name()
 			}) {
 				continue
 			}
 			if err := schedulersController.RemoveScheduler(name); err != nil {
 				log.Error("failed to remove scheduler",
-					zap.String("scheduler-name", name),
+					zap.Stringer("scheduler-name", name),
 					errs.ZapError(err))
 				continue
 			}
 			log.Info("remove scheduler successfully",
-				zap.String("scheduler-name", name))
+				zap.Stringer("scheduler-name", name))
 		}
 	}
 }

pkg/mcs/scheduling/server/config/watcher.go

@@ -24,6 +24,7 @@ import (
 
 	"github.com/coreos/go-semver/semver"
 	"github.com/pingcap/log"
+	"github.com/tikv/pd/pkg/errs"
 	sc "github.com/tikv/pd/pkg/schedule/config"
 	"github.com/tikv/pd/pkg/schedule/schedulers"
 	"github.com/tikv/pd/pkg/storage"
@@ -199,8 +200,15 @@ func (cw *Watcher) initializeSchedulerConfigWatcher() error {
 			return err
 		}
 		// Ensure the scheduler config could be updated as soon as possible.
-		if sc := cw.getSchedulersController(); sc != nil {
-			return sc.ReloadSchedulerConfig(name)
+		if schedulersController := cw.getSchedulersController(); schedulersController != nil {
+			schedulerName, err := sc.ConvertSchedulerStr2Name(name)
+			if err != nil {
+				log.Error("failed to convert scheduler name",
+					zap.String("scheduler", name),
+					errs.ZapError(err))
+				return errs.ErrSchedulerNotFound.GenWithStackByArgs()
+			}
+			return schedulersController.ReloadSchedulerConfig(schedulerName)
 		}
 		return nil
 	}

pkg/mock/mockconfig/mockconfig.go

@@ -23,7 +23,7 @@ import (
 func NewTestOptions() *config.PersistOptions {
 	// register default schedulers in case config check fail.
 	for _, d := range sc.DefaultSchedulers {
-		sc.RegisterScheduler(d.Type)
+		sc.RegisterScheduler(sc.CheckerSchedulerName(d.Type))
 	}
 	c := config.NewConfig()
 	c.Adjust(nil, false)

pkg/schedule/checker/checker_controller.go

@@ -104,7 +104,7 @@ func (c *Controller) CheckRegion(region *core.RegionInfo) []*operator.Operator {
 				if opController.OperatorCount(operator.OpReplica) < c.conf.GetReplicaScheduleLimit() {
 					return []*operator.Operator{op}
 				}
-				operator.OperatorLimitCounter.WithLabelValues(c.ruleChecker.GetType(), operator.OpReplica.String()).Inc()
+				operator.OperatorLimitCounter.WithLabelValues(c.ruleChecker.Name(), operator.OpReplica.String()).Inc()
 				c.regionWaitingList.Put(region.GetID(), nil)
 			}
 		}
@@ -116,7 +116,7 @@ func (c *Controller) CheckRegion(region *core.RegionInfo) []*operator.Operator {
 			if opController.OperatorCount(operator.OpReplica) < c.conf.GetReplicaScheduleLimit() {
 				return []*operator.Operator{op}
 			}
-			operator.OperatorLimitCounter.WithLabelValues(c.replicaChecker.GetType(), operator.OpReplica.String()).Inc()
+			operator.OperatorLimitCounter.WithLabelValues(c.replicaChecker.Name(), operator.OpReplica.String()).Inc()
 			c.regionWaitingList.Put(region.GetID(), nil)
 		}
 	}
@@ -133,7 +133,7 @@ func (c *Controller) CheckRegion(region *core.RegionInfo) []*operator.Operator {
 	if c.mergeChecker != nil {
 		allowed := opController.OperatorCount(operator.OpMerge) < c.conf.GetMergeScheduleLimit()
 		if !allowed {
-			operator.OperatorLimitCounter.WithLabelValues(c.mergeChecker.GetType(), operator.OpMerge.String()).Inc()
+			operator.OperatorLimitCounter.WithLabelValues(c.mergeChecker.Name(), operator.OpMerge.String()).Inc()
 		} else if ops := c.mergeChecker.Check(region); ops != nil {
 			// It makes sure that two operators can be added successfully altogether.
 			return ops

pkg/schedule/checker/joint_state_checker.go

@@ -19,6 +19,7 @@ import (
 	"github.com/tikv/pd/pkg/core"
 	"github.com/tikv/pd/pkg/core/constant"
 	"github.com/tikv/pd/pkg/errs"
+	"github.com/tikv/pd/pkg/schedule/config"
 	sche "github.com/tikv/pd/pkg/schedule/core"
 	"github.com/tikv/pd/pkg/schedule/operator"
 )
@@ -29,15 +30,13 @@ type JointStateChecker struct {
 	cluster sche.CheckerCluster
 }
 
-const jointStateCheckerName = "joint_state_checker"
-
 var (
 	// WithLabelValues is a heavy operation, define variable to avoid call it every time.
-	jointCheckCounter                 = checkerCounter.WithLabelValues(jointStateCheckerName, "check")
-	jointCheckerPausedCounter         = checkerCounter.WithLabelValues(jointStateCheckerName, "paused")
-	jointCheckerFailedCounter         = checkerCounter.WithLabelValues(jointStateCheckerName, "create-operator-fail")
-	jointCheckerNewOpCounter          = checkerCounter.WithLabelValues(jointStateCheckerName, "new-operator")
-	jointCheckerTransferLeaderCounter = checkerCounter.WithLabelValues(jointStateCheckerName, "transfer-leader")
+	jointCheckCounter                 = counterWithEvent(config.JointStateCheckerName, "check")
+	jointCheckerPausedCounter         = counterWithEvent(config.JointStateCheckerName, "paused")
+	jointCheckerFailedCounter         = counterWithEvent(config.JointStateCheckerName, "create-operator-fail")
+	jointCheckerNewOpCounter          = counterWithEvent(config.JointStateCheckerName, "new-operator")
+	jointCheckerTransferLeaderCounter = counterWithEvent(config.JointStateCheckerName, "transfer-leader")
 )
 
 // NewJointStateChecker creates a joint state checker.
@@ -47,6 +46,11 @@ func NewJointStateChecker(cluster sche.CheckerCluster) *JointStateChecker {
 	}
 }
 
+// Name returns the checker name.
+func (*JointStateChecker) Name() string {
+	return config.JointStateCheckerName.String()
+}
+
 // Check verifies a region's role, creating an Operator if need.
 func (c *JointStateChecker) Check(region *core.RegionInfo) *operator.Operator {
 	jointCheckCounter.Inc()

pkg/schedule/checker/learner_checker.go

@@ -18,6 +18,7 @@ import (
 	"github.com/pingcap/log"
 	"github.com/tikv/pd/pkg/core"
 	"github.com/tikv/pd/pkg/errs"
+	"github.com/tikv/pd/pkg/schedule/config"
 	sche "github.com/tikv/pd/pkg/schedule/core"
 	"github.com/tikv/pd/pkg/schedule/operator"
 )
@@ -30,7 +31,7 @@ type LearnerChecker struct {
 
 var (
 	// WithLabelValues is a heavy operation, define variable to avoid call it every time.
-	learnerCheckerPausedCounter = checkerCounter.WithLabelValues("learner_checker", "paused")
+	learnerCheckerPausedCounter = counterWithEvent(config.LearnerCheckerName, "paused")
 )
 
 // NewLearnerChecker creates a learner checker.
@@ -40,6 +41,11 @@ func NewLearnerChecker(cluster sche.CheckerCluster) *LearnerChecker {
 	}
 }
 
+// Name returns the checker name.
+func (*LearnerChecker) Name() string {
+	return config.LearnerCheckerName.String()
+}
+
 // Check verifies a region's role, creating an Operator if need.
 func (l *LearnerChecker) Check(region *core.RegionInfo) *operator.Operator {
 	if l.IsPaused() {