set MaxScheduleCost to 0 by default

Signed-off-by: Ryan Leung <rleungx@gmail.com>

server/config.go

@@ -560,13 +560,13 @@ const (
 	defaultSplitMergeInterval     = 1 * time.Hour
 	defaultPatrolRegionInterval   = 100 * time.Millisecond
 	defaultMaxStoreDownTime       = 30 * time.Minute
-	defaultLeaderScheduleLimit    = 4
-	defaultRegionScheduleLimit    = 4
-	defaultReplicaScheduleLimit   = 8
+	defaultLeaderScheduleLimit    = 1000
+	defaultRegionScheduleLimit    = 1000
+	defaultReplicaScheduleLimit   = 1000
 	defaultMergeScheduleLimit     = 8
 	defaultHotRegionScheduleLimit = 2
-	defaultMaxScheduleCost        = 200
-	defaultStoreMaxScheduleCost   = 20
+	defaultMaxScheduleCost        = 0
+	defaultStoreMaxScheduleCost   = 50
 	defaultTolerantSizeRatio      = 5
 	defaultLowSpaceRatio          = 0.8
 	defaultHighSpaceRatio         = 0.6

server/coordinator.go

@@ -133,7 +133,8 @@ func (c *coordinator) checkRegion(region *core.RegionInfo) bool {
 	// If PD has restarted, it need to check learners added before and promote them.
 	// Don't check isRaftLearnerEnabled cause it maybe disable learner feature but there are still some learners to promote.
 	opController := c.opController
-	if c.opController.GetScheduleCost() >= c.cluster.GetMaxScheduleCost() {
+	maxScheduleCost := c.cluster.GetMaxScheduleCost()
+	if maxScheduleCost != 0 && c.opController.GetScheduleCost() >= maxScheduleCost {
 		return false
 	}
 	for _, p := range region.GetLearners() {
@@ -472,7 +473,8 @@ func (s *scheduleController) GetInterval() time.Duration {
 
 // AllowSchedule returns if a scheduler is allowed to schedule.
 func (s *scheduleController) AllowSchedule() bool {
-	if s.opController.GetScheduleCost() >= s.cluster.GetMaxScheduleCost() {
+	maxScheduleCost := s.cluster.GetMaxScheduleCost()
+	if maxScheduleCost != 0 && s.opController.GetScheduleCost() >= maxScheduleCost {
 		return false
 	}
 	return s.Scheduler.IsScheduleAllowed(s.cluster)

server/schedule/mockcluster.go

@@ -495,8 +495,8 @@ const (
 	defaultReplicaScheduleLimit        = 8
 	defaultMergeScheduleLimit          = 8
 	defaultHotRegionScheduleLimit      = 2
-	defaultMaxScheduleCost             = 200
-	defaultStoreMaxScheduleCost        = 20
+	defaultMaxScheduleCost             = 0
+	defaultStoreMaxScheduleCost        = 50
 	defaultTolerantSizeRatio           = 2.5
 	defaultLowSpaceRatio               = 0.8
 	defaultHighSpaceRatio              = 0.6

server/schedule/operator.go

@@ -36,7 +36,7 @@ const (
 	// longer than it, the operator will be considered timeout.
 	RegionOperatorWaitTime = 10 * time.Minute
 	// RegionWeight reflects the influence which is caused by a region related step in an operator.
-	RegionWeight = 2
+	RegionWeight = 10
 	// LeaderWeight reflects the influence which is caused by a leader related step in an operator.
 	LeaderWeight = 1
 )

server/schedule/operator_test.go

@@ -123,7 +123,7 @@ func (s *testOperatorSuite) TestInfluence(c *C) {
 		LeaderCount: 0,
 		RegionSize:  10,
 		RegionCount: 1,
-		StepCost:    2,
+		StepCost:    10,
 	})
 
 	TransferLeader{FromStore: 1, ToStore: 2}.Influence(opInfluence, region)
@@ -139,7 +139,7 @@ func (s *testOperatorSuite) TestInfluence(c *C) {
 		LeaderCount: 1,
 		RegionSize:  10,
 		RegionCount: 1,
-		StepCost:    3,
+		StepCost:    11,
 	})
 
 	RemovePeer{FromStore: 1}.Influence(opInfluence, region)
@@ -148,14 +148,14 @@ func (s *testOperatorSuite) TestInfluence(c *C) {
 		LeaderCount: -1,
 		RegionSize:  -10,
 		RegionCount: -1,
-		StepCost:    3,
+		StepCost:    11,
 	})
 	c.Assert(*storeOpInfluence[2], DeepEquals, StoreInfluence{
 		LeaderSize:  10,
 		LeaderCount: 1,
 		RegionSize:  10,
 		RegionCount: 1,
-		StepCost:    3,
+		StepCost:    11,
 	})
 
 	MergeRegion{IsPassive: false}.Influence(opInfluence, region)
@@ -164,14 +164,14 @@ func (s *testOperatorSuite) TestInfluence(c *C) {
 		LeaderCount: -1,
 		RegionSize:  -10,
 		RegionCount: -1,
-		StepCost:    3,
+		StepCost:    11,
 	})
 	c.Assert(*storeOpInfluence[2], DeepEquals, StoreInfluence{
 		LeaderSize:  10,
 		LeaderCount: 1,
 		RegionSize:  10,
 		RegionCount: 1,
-		StepCost:    3,
+		StepCost:    11,
 	})
 
 	MergeRegion{IsPassive: true}.Influence(opInfluence, region)
@@ -180,14 +180,14 @@ func (s *testOperatorSuite) TestInfluence(c *C) {
 		LeaderCount: -2,
 		RegionSize:  -10,
 		RegionCount: -2,
-		StepCost:    6,
+		StepCost:    22,
 	})
 	c.Assert(*storeOpInfluence[2], DeepEquals, StoreInfluence{
 		LeaderSize:  10,
 		LeaderCount: 1,
 		RegionSize:  10,
 		RegionCount: 0,
-		StepCost:    5,
+		StepCost:    21,
 	})
 }
 

tools/pd-simulator/simulator/raft.go

@@ -185,6 +185,8 @@ func (r *RaftEngine) NeedSplit(size, rows int64) bool {
 }
 
 func (r *RaftEngine) recordRegionChange(region *core.RegionInfo) {
+	r.Lock()
+	defer r.Unlock()
 	n := region.GetLeader().GetStoreId()
 	r.regionChange[n] = append(r.regionChange[n], region.GetID())
 }

tools/pd-simulator/simulator/task.go

@@ -17,13 +17,10 @@ import (
 	"bytes"
 	"fmt"
 
-	"go.uber.org/zap"
-
 	"github.com/pingcap/kvproto/pkg/eraftpb"
 	"github.com/pingcap/kvproto/pkg/metapb"
 	"github.com/pingcap/kvproto/pkg/pdpb"
 	"github.com/pingcap/pd/server/core"
-	"github.com/pingcap/pd/tools/pd-simulator/simulator/simutil"
 )
 
 // Task running in node.
@@ -234,7 +231,6 @@ func (a *addPeer) Step(r *RaftEngine) {
 	snapshotSize := region.GetApproximateSize()
 	sendNode := r.conn.Nodes[region.GetLeader().GetStoreId()]
 	if sendNode == nil {
-		simutil.Logger.Error("failed to sent snapshot, node has been deleted", zap.Uint64("node-id", sendNode.Id))
 		a.finished = true
 		return
 	}
@@ -245,7 +241,6 @@ func (a *addPeer) Step(r *RaftEngine) {
 
 	recvNode := r.conn.Nodes[a.peer.GetStoreId()]
 	if recvNode == nil {
-		simutil.Logger.Error("failed to receive snapshot: node has been deleted", zap.Uint64("node-id", recvNode.Id))
 		a.finished = true
 		return
 	}