schedule: adjust store balance rate meaning

conf/config.toml

@@ -64,8 +64,8 @@ max-snapshot-count = 3
 max-pending-peer-count = 16
 max-store-down-time = "30m"
 leader-schedule-limit = 4
-region-schedule-limit = 1024
-replica-schedule-limit = 1024
+region-schedule-limit = 64
+replica-schedule-limit = 64
 merge-schedule-limit = 8
 tolerant-size-ratio = 0
 # Enable two-way merge, set it to true may help improving merge speed.

server/api/store.go

@@ -26,6 +26,7 @@ import (
 	"github.com/pingcap/pd/pkg/typeutil"
 	"github.com/pingcap/pd/server"
 	"github.com/pingcap/pd/server/core"
+	"github.com/pingcap/pd/server/schedule"
 	"github.com/pkg/errors"
 	"github.com/unrolled/render"
 )
@@ -328,7 +329,7 @@ func (h *storeHandler) SetLimit(w http.ResponseWriter, r *http.Request) {
 		return
 	}
 
-	if err := h.SetStoreLimit(storeID, rate); err != nil {
+	if err := h.SetStoreLimit(storeID, rate/schedule.StoreBalanceBaseTime); err != nil {
 		h.rd.JSON(w, http.StatusInternalServerError, err.Error())
 		return
 	}
@@ -381,7 +382,7 @@ func (h *storesHandler) SetAllLimit(w http.ResponseWriter, r *http.Request) {
 		return
 	}
 
-	if err := h.SetAllStoresLimit(rate); err != nil {
+	if err := h.SetAllStoresLimit(rate / schedule.StoreBalanceBaseTime); err != nil {
 		h.rd.JSON(w, http.StatusInternalServerError, err.Error())
 		return
 	}

server/api/trend_test.go

@@ -29,7 +29,7 @@ var _ = Suite(&testTrendSuite{})
 type testTrendSuite struct{}
 
 func (s *testTrendSuite) TestTrend(c *C) {
-	svr, cleanup := mustNewServer(c)
+	svr, cleanup := mustNewServer(c, func(cfg *server.Config) { cfg.Schedule.StoreBalanceRate = 60 })
 	defer cleanup()
 	mustWaitLeader(c, []*server.Server{svr})
 
@@ -49,7 +49,7 @@ func (s *testTrendSuite) TestTrend(c *C) {
 	// Create 3 operators that transfers leader, moves follower, moves leader.
 	c.Assert(svr.GetHandler().AddTransferLeaderOperator(4, 2), IsNil)
 	c.Assert(svr.GetHandler().AddTransferPeerOperator(5, 2, 3), IsNil)
-	time.Sleep(2 * time.Second)
+	time.Sleep(1 * time.Second)
 	c.Assert(svr.GetHandler().AddTransferPeerOperator(6, 1, 3), IsNil)
 
 	// Complete the operators.

server/config.go

@@ -568,11 +568,11 @@ const (
 	defaultPatrolRegionInterval   = 100 * time.Millisecond
 	defaultMaxStoreDownTime       = 30 * time.Minute
 	defaultLeaderScheduleLimit    = 8
-	defaultRegionScheduleLimit    = 1024
-	defaultReplicaScheduleLimit   = 1024
+	defaultRegionScheduleLimit    = 64
+	defaultReplicaScheduleLimit   = 64
 	defaultMergeScheduleLimit     = 8
 	defaultHotRegionScheduleLimit = 2
-	defaultStoreBalanceRate       = 1
+	defaultStoreBalanceRate       = 15
 	defaultTolerantSizeRatio      = 0
 	defaultLowSpaceRatio          = 0.8
 	defaultHighSpaceRatio         = 0.6

server/coordinator_test.go

@@ -38,6 +38,7 @@ func newTestOperator(regionID uint64, regionEpoch *metapb.RegionEpoch, kind sche
 func newTestScheduleConfig() (*ScheduleConfig, *scheduleOption, error) {
 	cfg := NewConfig()
 	cfg.Schedule.TolerantSizeRatio = 5
+	cfg.Schedule.StoreBalanceRate = 60
 	if err := cfg.Adjust(nil); err != nil {
 		return nil, nil, err
 	}

server/schedule/mockcluster.go

@@ -534,7 +534,7 @@ const (
 	defaultReplicaScheduleLimit        = 8
 	defaultMergeScheduleLimit          = 8
 	defaultHotRegionScheduleLimit      = 2
-	defaultStoreBalanceRate            = 1
+	defaultStoreBalanceRate            = 60
 	defaultTolerantSizeRatio           = 2.5
 	defaultLowSpaceRatio               = 0.8
 	defaultHighSpaceRatio              = 0.6

server/schedule/operator_controller.go

@@ -44,6 +44,8 @@ var (
 	fastNotifyInterval = 2 * time.Second
 	// PushOperatorTickInterval is the interval try to push the operator.
 	PushOperatorTickInterval = 500 * time.Millisecond
+	// StoreBalanceBaseTime represents the base time of balance rate.
+	StoreBalanceBaseTime float64 = 60
 )
 
 // HeartbeatStreams is an interface of async region heartbeat.
@@ -660,7 +662,7 @@ func (oc *OperatorController) newStoreLimit(storeID uint64, rate float64) {
 // getOrCreateStoreLimit is used to get or create the limit of a store.
 func (oc *OperatorController) getOrCreateStoreLimit(storeID uint64) *ratelimit.Bucket {
 	if oc.storesLimit[storeID] == nil {
-		rate := oc.cluster.GetStoreBalanceRate()
+		rate := oc.cluster.GetStoreBalanceRate() / StoreBalanceBaseTime
 		oc.newStoreLimit(storeID, rate)
 		oc.cluster.AttachOverloadStatus(storeID, func() bool {
 			oc.RLock()

tests/cmd/pdctl_test.go

@@ -911,7 +911,7 @@ func (s *cmdTestSuite) TestOperator(c *C) {
 	cluster, err := tests.NewTestCluster(3,
 		func(conf *server.Config) { conf.Replication.MaxReplicas = 2 },
 		func(conf *server.Config) { conf.Schedule.MaxStoreDownTime.Duration = time.Since(t) },
-		func(conf *server.Config) { conf.Schedule.StoreBalanceRate = 4 },
+		func(conf *server.Config) { conf.Schedule.StoreBalanceRate = 240 },
 	)
 	c.Assert(err, IsNil)
 	err = cluster.RunInitialServers()