scheduler: add test for creating evict-leader-scheduler twice (#8757) (#8973)

close #8756

Signed-off-by: okJiang <[email protected]>

Co-authored-by: okJiang <[email protected]>

Author: ti-chi-bot

pkg/core/store.go

@@ -779,6 +779,7 @@ func (s *StoresInfo) ResetStores() {
 func (s *StoresInfo) PauseLeaderTransfer(storeID uint64) error {
 	s.Lock()
 	defer s.Unlock()
+	log.Info("pause store leader transfer", zap.Uint64("store-id", storeID))
 	store, ok := s.stores[storeID]
 	if !ok {
 		return errs.ErrStoreNotFound.FastGenByArgs(storeID)
@@ -795,6 +796,7 @@ func (s *StoresInfo) PauseLeaderTransfer(storeID uint64) error {
 func (s *StoresInfo) ResumeLeaderTransfer(storeID uint64) {
 	s.Lock()
 	defer s.Unlock()
+	log.Info("resume store leader transfer", zap.Uint64("store-id", storeID))
 	store, ok := s.stores[storeID]
 	if !ok {
 		log.Warn("try to clean a store's pause state, but it is not found. It may be cleanup",

pkg/schedule/schedulers/grant_leader.go

@@ -45,7 +45,7 @@ type grantLeaderSchedulerConfig struct {
 }
 
 func (conf *grantLeaderSchedulerConfig) buildWithArgs(args []string) error {
-	if len(args) != 1 {
+	if len(args) < 1 {
 		return errs.ErrSchedulerConfig.FastGenByArgs("id")
 	}
 
@@ -271,6 +271,7 @@ func (handler *grantLeaderHandler) updateConfig(w http.ResponseWriter, r *http.R
 
 	err := handler.config.buildWithArgs(args)
 	if err != nil {
+		log.Error("fail to build config", errs.ZapError(err))
 		handler.config.Lock()
 		handler.config.cluster.ResumeLeaderTransfer(id)
 		handler.config.Unlock()
@@ -279,6 +280,7 @@ func (handler *grantLeaderHandler) updateConfig(w http.ResponseWriter, r *http.R
 	}
 	err = handler.config.persist()
 	if err != nil {
+		log.Error("fail to persist config", errs.ZapError(err))
 		_, _ = handler.config.removeStore(id)
 		handler.rd.JSON(w, http.StatusInternalServerError, err.Error())
 		return

pkg/schedule/schedulers/init.go

@@ -134,7 +134,7 @@ func schedulersRegister() {
 	// evict leader
 	RegisterSliceDecoderBuilder(types.EvictLeaderScheduler, func(args []string) ConfigDecoder {
 		return func(v any) error {
-			if len(args) != 1 {
+			if len(args) < 1 {
 				return errs.ErrSchedulerConfig.FastGenByArgs("id")
 			}
 			conf, ok := v.(*evictLeaderSchedulerConfig)
@@ -268,7 +268,7 @@ func schedulersRegister() {
 	// grant leader
 	RegisterSliceDecoderBuilder(types.GrantLeaderScheduler, func(args []string) ConfigDecoder {
 		return func(v any) error {
-			if len(args) != 1 {
+			if len(args) < 1 {
 				return errs.ErrSchedulerConfig.FastGenByArgs("id")
 			}
 

plugin/scheduler_example/evict_leader.go

@@ -50,7 +50,7 @@ const (
 func init() {
 	schedulers.RegisterSliceDecoderBuilder(userEvictLeaderScheduler, func(args []string) schedulers.ConfigDecoder {
 		return func(v any) error {
-			if len(args) != 1 {
+			if len(args) < 1 {
 				return errors.New("should specify the store-id")
 			}
 			conf, ok := v.(*evictLeaderSchedulerConfig)
@@ -101,7 +101,7 @@ type evictLeaderSchedulerConfig struct {
 
 // BuildWithArgs builds the config with the args.
 func (conf *evictLeaderSchedulerConfig) BuildWithArgs(args []string) error {
-	if len(args) != 1 {
+	if len(args) < 1 {
 		return errors.New("should specify the store-id")
 	}
 

tests/integrations/realcluster/real_cluster.go

@@ -37,7 +37,7 @@ type realClusterSuite struct {
 }
 
 var (
-	playgroundLogDir = filepath.Join("tmp", "real_cluster", "playground")
+	playgroundLogDir = "/tmp/real_cluster/playground"
 	tiupBin          = os.Getenv("HOME") + "/.tiup/bin/tiup"
 )
 

tests/integrations/realcluster/scheduler_test.go

@@ -21,12 +21,14 @@ import (
 	"testing"
 	"time"
 
+	"github.com/pingcap/log"
 	"github.com/stretchr/testify/require"
 	"github.com/stretchr/testify/suite"
 	"github.com/tikv/pd/client/http"
 	"github.com/tikv/pd/client/testutil"
 	"github.com/tikv/pd/pkg/schedule/labeler"
 	"github.com/tikv/pd/pkg/schedule/types"
+	"go.uber.org/zap"
 )
 
 type schedulerSuite struct {
@@ -201,3 +203,72 @@ func (s *schedulerSuite) TestRegionLabelDenyScheduler() {
 		return true
 	}, testutil.WithWaitFor(time.Minute))
 }
+
+func (s *schedulerSuite) TestGrantOrEvictLeaderTwice() {
+	re := require.New(s.T())
+	ctx, cancel := context.WithCancel(context.Background())
+	defer cancel()
+
+	pdHTTPCli := http.NewClient("pd-real-cluster-test", getPDEndpoints(s.T()))
+	regions, err := pdHTTPCli.GetRegions(ctx)
+	re.NoError(err)
+	re.NotEmpty(regions.Regions)
+	region1 := regions.Regions[0]
+
+	var i int
+	evictLeader := func() {
+		re.NoError(pdHTTPCli.CreateScheduler(ctx, types.EvictLeaderScheduler.String(), uint64(region1.Leader.StoreID)))
+		// if the second evict leader scheduler cause the pause-leader-filter
+		// disable, the balance-leader-scheduler need some time to transfer
+		// leader. See details in https://github.com/tikv/pd/issues/8756.
+		if i == 1 {
+			time.Sleep(3 * time.Second)
+		}
+		testutil.Eventually(re, func() bool {
+			regions, err := pdHTTPCli.GetRegions(ctx)
+			if err != nil {
+				log.Error("get regions failed", zap.Error(err))
+				return false
+			}
+			for _, region := range regions.Regions {
+				if region.Leader.StoreID == region1.Leader.StoreID {
+					return false
+				}
+			}
+			return true
+		}, testutil.WithWaitFor(time.Minute))
+
+		i++
+	}
+
+	evictLeader()
+	evictLeader()
+	pdHTTPCli.DeleteScheduler(ctx, types.EvictLeaderScheduler.String())
+
+	i = 0
+	grantLeader := func() {
+		re.NoError(pdHTTPCli.CreateScheduler(ctx, types.GrantLeaderScheduler.String(), uint64(region1.Leader.StoreID)))
+		if i == 1 {
+			time.Sleep(3 * time.Second)
+		}
+		testutil.Eventually(re, func() bool {
+			regions, err := pdHTTPCli.GetRegions(ctx)
+			if err != nil {
+				log.Error("get regions failed", zap.Error(err))
+				return false
+			}
+			for _, region := range regions.Regions {
+				if region.Leader.StoreID != region1.Leader.StoreID {
+					return false
+				}
+			}
+			return true
+		}, testutil.WithWaitFor(2*time.Minute))
+
+		i++
+	}
+
+	grantLeader()
+	grantLeader()
+	pdHTTPCli.DeleteScheduler(ctx, types.GrantLeaderScheduler.String())
+}