schedule: not limit remove peer of the down store (#4097)

server/cluster/coordinator_test.go

@@ -987,6 +987,47 @@ func (s *testOperatorControllerSuite) TestStoreOverloadedWithReplace(c *C) {
 	c.Assert(lb.Schedule(tc), NotNil)
 }
 
+func (s *testOperatorControllerSuite) TestDownStoreLimit(c *C) {
+	tc, co, cleanup := prepare(nil, nil, nil, c)
+	defer cleanup()
+	oc := co.opController
+	rc := co.checkers.GetReplicaChecker()
+
+	tc.addRegionStore(1, 100)
+	tc.addRegionStore(2, 100)
+	tc.addRegionStore(3, 100)
+	tc.addLeaderRegion(1, 1, 2, 3)
+
+	region := tc.GetRegion(1)
+	tc.setStoreDown(1)
+	tc.SetStoreLimit(1, storelimit.RemovePeer, 1)
+
+	region = region.Clone(core.WithDownPeers([]*pdpb.PeerStats{
+		{
+			Peer:        region.GetStorePeer(1),
+			DownSeconds: 24 * 60 * 60,
+		},
+	}), core.SetApproximateSize(1))
+	tc.putRegion(region)
+	for i := uint64(1); i < 20; i++ {
+		tc.addRegionStore(i+3, 100)
+		op := rc.Check(region)
+		c.Assert(op, NotNil)
+		c.Assert(oc.AddOperator(op), IsTrue)
+		oc.RemoveOperator(op)
+	}
+
+	region = region.Clone(core.SetApproximateSize(100))
+	tc.putRegion(region)
+	for i := uint64(20); i < 25; i++ {
+		tc.addRegionStore(i+3, 100)
+		op := rc.Check(region)
+		c.Assert(op, NotNil)
+		c.Assert(oc.AddOperator(op), IsTrue)
+		oc.RemoveOperator(op)
+	}
+}
+
 var _ = Suite(&testScheduleControllerSuite{})
 
 type testScheduleControllerSuite struct {

server/schedule/checker/rule_checker.go

@@ -53,7 +53,6 @@ func (c *RuleChecker) GetType() string {
 // fix it.
 func (c *RuleChecker) Check(region *core.RegionInfo) *operator.Operator {
 	checkerCounter.WithLabelValues("rule_checker", "check").Inc()
-
 	fit := c.cluster.FitRegion(region)
 	if len(fit.RuleFits) == 0 {
 		checkerCounter.WithLabelValues("rule_checker", "fix-range").Inc()

server/schedule/checker_controller.go

@@ -93,3 +93,8 @@ func (c *CheckerController) CheckRegion(region *core.RegionInfo) (bool, []*opera
 func (c *CheckerController) GetMergeChecker() *checker.MergeChecker {
 	return c.mergeChecker
 }
+
+// GetReplicaChecker returns the replica checker.
+func (c *CheckerController) GetReplicaChecker() *checker.ReplicaChecker {
+	return c.replicaChecker
+}

server/schedule/operator/builder.go

@@ -341,7 +341,13 @@ func (b *Builder) execAddPeer(p *metapb.Peer) {
 }
 
 func (b *Builder) execRemovePeer(p *metapb.Peer) {
-	b.steps = append(b.steps, RemovePeer{FromStore: p.GetStoreId()})
+	removeStoreID := p.GetStoreId()
+	var isDownStore bool
+	store := b.cluster.GetStore(removeStoreID)
+	if store != nil {
+		isDownStore = store.DownTime() > b.cluster.GetMaxStoreDownTime()
+	}
+	b.steps = append(b.steps, RemovePeer{FromStore: removeStoreID, IsDownStore: isDownStore})
 	b.currentPeers.Delete(p.GetStoreId())
 	b.toRemove.Delete(p.GetStoreId())
 }

server/schedule/operator/influence.go

@@ -80,7 +80,7 @@ func (s *StoreInfluence) addStepCost(limitType storelimit.Type, cost int64) {
 func (s *StoreInfluence) AdjustStepCost(limitType storelimit.Type, regionSize int64) {
 	if regionSize > storelimit.SmallRegionThreshold {
 		s.addStepCost(limitType, storelimit.RegionInfluence[limitType])
-	} else if regionSize <= storelimit.SmallRegionThreshold && regionSize > core.EmptyRegionApproximateSize {
+	} else if regionSize > core.EmptyRegionApproximateSize {
 		s.addStepCost(limitType, storelimit.SmallRegionInfluence[limitType])
 	}
 }

server/schedule/operator/step.go

@@ -220,7 +220,8 @@ func (pl PromoteLearner) Influence(opInfluence OpInfluence, region *core.RegionI
 
 // RemovePeer is an OpStep that removes a region peer.
 type RemovePeer struct {
-	FromStore uint64
+	FromStore   uint64
+	IsDownStore bool
 }
 
 // ConfVerChanged returns true if the conf version has been changed by this step
@@ -252,6 +253,10 @@ func (rp RemovePeer) Influence(opInfluence OpInfluence, region *core.RegionInfo)
 	regionSize := region.GetApproximateSize()
 	from.RegionSize -= regionSize
 	from.RegionCount--
+
+	if rp.IsDownStore && regionSize > storelimit.SmallRegionThreshold {
+		regionSize = storelimit.SmallRegionThreshold
+	}
 	from.AdjustStepCost(storelimit.RemovePeer, regionSize)
 }