schedule, server: fix replicated state semantics

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

@@ -1466,7 +1466,7 @@ func splitRegions(c *gin.Context) {
 }
 
 // @Tags     region
-// @Summary  Check if regions in the given key ranges are replicated. Returns 'REPLICATED', 'INPROGRESS', or 'PENDING'. 'PENDING' means that there is at least one region pending for scheduling. Similarly, 'INPROGRESS' means there is at least one region in scheduling.
+// @Summary  Check if regions in the given key ranges are replicated. Returns 'REPLICATED', 'INPROGRESS', or 'PENDING'. 'PENDING' means that PD cannot currently make scheduling progress because the Store topology has no eligible placement. 'INPROGRESS' means that placement is incomplete but PD can continue or retry scheduling, including incomplete Region metadata in the requested range.
 // @Param    startKey  query  string  true  "Regions start key, hex encoded"
 // @Param    endKey    query  string  true  "Regions end key, hex encoded"
 // @Produce  plain

pkg/schedule/checker/checker_controller.go

@@ -266,6 +266,12 @@ func (c *Controller) checkPendingProcessedRegions() {
 	pendingProcessedRegionsGauge.Set(float64(len(ids)))
 	for _, id := range ids {
 		region := c.cluster.GetRegion(id)
+		if region == nil {
+			continue
+		}
+		// Remove the old entry before checking. A checker will add it back if
+		// temporary conditions still prevent scheduling.
+		c.RemovePendingProcessedRegion(id)
 		c.tryAddOperators(region)
 	}
 }

pkg/schedule/checker/replica_strategy.go

@@ -30,13 +30,15 @@ import (
 // ReplicaStrategy collects some utilities to manipulate region peers. It
 // exists to allow replica_checker and rule_checker to reuse common logics.
 type ReplicaStrategy struct {
-	checkerName    string // replica-checker / rule-checker
-	cluster        sche.CheckerCluster
-	locationLabels []string
-	isolationLevel string
-	region         *core.RegionInfo
-	extraFilters   []filter.Filter
-	fastFailover   bool
+	checkerName             string // replica-checker / rule-checker
+	cluster                 sche.CheckerCluster
+	locationLabels          []string
+	isolationLevel          string
+	region                  *core.RegionInfo
+	extraFilters            []filter.Filter
+	fastFailover            bool
+	storeCandidates         []*core.StoreInfo
+	storeCandidatesPrepared bool
 }
 
 // SelectStoreToAdd returns the store to add a replica to a region.
@@ -62,25 +64,29 @@ func (s *ReplicaStrategy) SelectStoreToAdd(coLocationStores []*core.StoreInfo, e
 	if s.fastFailover {
 		level = constant.Urgent
 	}
-	filters := []filter.Filter{
-		filter.NewExcludedFilter(s.checkerName, nil, s.region.GetStoreIDs()),
-		filter.NewStorageThresholdFilter(s.checkerName),
-		filter.NewSpecialUseFilter(s.checkerName),
-		&filter.StoreStateFilter{ActionScope: s.checkerName, MoveRegion: true, AllowTemporaryStates: true, OperatorLevel: level},
+	stores := s.storeCandidates
+	filters := []filter.Filter{filter.NewExcludedFilter(s.checkerName, nil, s.region.GetStoreIDs())}
+	if !s.storeCandidatesPrepared {
+		stores = s.cluster.GetStores()
+		filters = append(filters,
+			filter.NewStorageThresholdFilter(s.checkerName),
+			filter.NewSpecialUseFilter(s.checkerName),
+			&filter.StoreStateFilter{ActionScope: s.checkerName, MoveRegion: true, AllowTemporaryStates: true, OperatorLevel: level},
+		)
 	}
 	if len(s.locationLabels) > 0 && s.isolationLevel != "" {
 		filters = append(filters, filter.NewIsolationFilter(s.checkerName, s.isolationLevel, s.locationLabels, coLocationStores))
 	}
 	if len(extraFilters) > 0 {
 		filters = append(filters, extraFilters...)
 	}
-	if len(s.extraFilters) > 0 {
+	if !s.storeCandidatesPrepared && len(s.extraFilters) > 0 {
 		filters = append(filters, s.extraFilters...)
 	}
 
 	isolationComparer := filter.IsolationComparer(s.locationLabels, coLocationStores)
 	strictStateFilter := &filter.StoreStateFilter{ActionScope: s.checkerName, MoveRegion: true, AllowFastFailover: s.fastFailover, OperatorLevel: level}
-	targetCandidate := filter.NewCandidates(s.cluster.GetStores()).
+	targetCandidate := filter.NewCandidates(stores).
 		FilterTarget(s.cluster.GetCheckerConfig(), nil, nil, filters...).
 		KeepTheTopStores(isolationComparer, false) // greater isolation score is better
 	if targetCandidate.Len() == 0 {
@@ -94,6 +100,24 @@ func (s *ReplicaStrategy) SelectStoreToAdd(coLocationStores []*core.StoreInfo, e
 	return target.GetID(), false
 }
 
+// prepareStoreCandidates applies the Region-independent target filters so the
+// result can be reused by placement-state checks using the same Rule.
+func (s *ReplicaStrategy) prepareStoreCandidates(stores []*core.StoreInfo) {
+	level := constant.High
+	if s.fastFailover {
+		level = constant.Urgent
+	}
+	filters := []filter.Filter{
+		filter.NewStorageThresholdFilter(s.checkerName),
+		filter.NewSpecialUseFilter(s.checkerName),
+		&filter.StoreStateFilter{ActionScope: s.checkerName, MoveRegion: true, AllowTemporaryStates: true, OperatorLevel: level},
+	}
+	filters = append(filters, s.extraFilters...)
+	s.storeCandidates = filter.NewCandidates(stores).
+		FilterTarget(s.cluster.GetCheckerConfig(), nil, nil, filters...).Stores
+	s.storeCandidatesPrepared = true
+}
+
 // SelectStoreToFix returns a store to replace down/offline old peer. The location
 // placement after scheduling is allowed to be worse than original.
 func (s *ReplicaStrategy) SelectStoreToFix(coLocationStores []*core.StoreInfo, old uint64) (uint64, bool) {