validate: manual reset pd config back

cmd/validate.go

@@ -9,6 +9,7 @@ import (
 	"encoding/hex"
 	"encoding/json"
 	"fmt"
+	"strings"
 
 	"github.com/gogo/protobuf/proto"
 	"github.com/pingcap/errors"
@@ -20,6 +21,7 @@ import (
 	"github.com/tikv/pd/pkg/mock/mockid"
 	"go.uber.org/zap"
 
+	"github.com/pingcap/br/pkg/conn"
 	"github.com/pingcap/br/pkg/restore"
 	"github.com/pingcap/br/pkg/rtree"
 	"github.com/pingcap/br/pkg/task"
@@ -45,6 +47,7 @@ func NewValidateCommand() *cobra.Command {
 	meta.AddCommand(newBackupMetaCommand())
 	meta.AddCommand(decodeBackupMetaCommand())
 	meta.AddCommand(encodeBackupMetaCommand())
+	meta.AddCommand(setPDConfigCommand())
 	meta.Hidden = true
 
 	return meta
@@ -323,3 +326,43 @@ func encodeBackupMetaCommand() *cobra.Command {
 	}
 	return encodeBackupMetaCmd
 }
+
+func setPDConfigCommand() *cobra.Command {
+	pdConfigCmd := &cobra.Command{
+		Use:   "reset-pd-config-as-default",
+		Short: "reset pd scheduler and config adjusted by BR to default value",
+		RunE: func(cmd *cobra.Command, args []string) error {
+			ctx, cancel := context.WithCancel(GetDefaultContext())
+			defer cancel()
+
+			var cfg task.Config
+			if err := cfg.ParseFromFlags(cmd.Flags()); err != nil {
+				return err
+			}
+
+			mgr, err := task.NewMgr(ctx, tidbGlue, cfg.PD, cfg.TLS, cfg.CheckRequirements)
+			if err != nil {
+				return err
+			}
+			defer mgr.Close()
+
+			for scheduler := range conn.Schedulers {
+				if strings.HasPrefix(scheduler, "balance") {
+					err := mgr.AddScheduler(ctx, scheduler)
+					if err != nil {
+						return err
+					}
+					log.Info("add pd schedulers succeed",
+						zap.String("schedulers", scheduler))
+				}
+			}
+
+			if err := mgr.UpdatePDScheduleConfig(ctx, conn.DefaultPDCfg); err != nil {
+				return errors.Annotate(err, "fail to update PD merge config")
+			}
+			log.Info("add pd configs succeed", zap.Any("config", conn.DefaultPDCfg))
+			return nil
+		},
+	}
+	return pdConfigCmd
+}

pkg/conn/scheduler_utils.go

@@ -21,7 +21,8 @@ type clusterConfig struct {
 }
 
 var (
-	schedulers = map[string]struct{}{
+	// Schedulers represent region/leader schedulers which can impact on performance.
+	Schedulers = map[string]struct{}{
 		"balance-leader-scheduler":     {},
 		"balance-hot-region-scheduler": {},
 		"balance-region-scheduler":     {},
@@ -30,6 +31,7 @@ var (
 		"shuffle-region-scheduler":     {},
 		"shuffle-hot-region-scheduler": {},
 	}
+
 	pdRegionMergeCfg = []string{
 		"max-merge-region-keys",
 		"max-merge-region-size",
@@ -39,6 +41,15 @@ var (
 		"region-schedule-limit",
 		"max-snapshot-count",
 	}
+
+	// DefaultPDCfg find by https://github.com/tikv/pd/blob/master/conf/config.toml.
+	DefaultPDCfg = map[string]interface{}{
+		"max-merge-region-keys": 200000,
+		"max-merge-region-size": 20,
+		"leader-schedule-limit": 4,
+		"region-schedule-limit": 2048,
+		"max-snapshot-count":    3,
+	}
 )
 
 func addPDLeaderScheduler(ctx context.Context, mgr *Mgr, removedSchedulers []string) error {
@@ -101,7 +112,7 @@ func (mgr *Mgr) RemoveSchedulers(ctx context.Context) (undo utils.UndoFunc, err
 	}
 	needRemoveSchedulers := make([]string, 0, len(existSchedulers))
 	for _, s := range existSchedulers {
-		if _, ok := schedulers[s]; ok {
+		if _, ok := Schedulers[s]; ok {
 			needRemoveSchedulers = append(needRemoveSchedulers, s)
 		}
 	}

pkg/task/backup.go

@@ -175,7 +175,7 @@ func RunBackup(c context.Context, g glue.Glue, cmdName string, cfg *BackupConfig
 	if err != nil {
 		return err
 	}
-	mgr, err := newMgr(ctx, g, cfg.PD, cfg.TLS, cfg.CheckRequirements)
+	mgr, err := NewMgr(ctx, g, cfg.PD, cfg.TLS, cfg.CheckRequirements)
 	if err != nil {
 		return err
 	}

pkg/task/backup_raw.go

@@ -125,7 +125,7 @@ func RunBackupRaw(c context.Context, g glue.Glue, cmdName string, cfg *RawKvConf
 	if err != nil {
 		return err
 	}
-	mgr, err := newMgr(ctx, g, cfg.PD, cfg.TLS, cfg.CheckRequirements)
+	mgr, err := NewMgr(ctx, g, cfg.PD, cfg.TLS, cfg.CheckRequirements)
 	if err != nil {
 		return err
 	}

pkg/task/common.go

@@ -277,8 +277,8 @@ func (cfg *Config) ParseFromFlags(flags *pflag.FlagSet) error {
 	return cfg.TLS.ParseFromFlags(flags)
 }
 
-// newMgr creates a new mgr at the given PD address.
-func newMgr(ctx context.Context,
+// NewMgr creates a new mgr at the given PD address.
+func NewMgr(ctx context.Context,
 	g glue.Glue, pds []string,
 	tlsConfig TLSConfig,
 	checkRequirements bool) (*conn.Mgr, error) {

pkg/task/restore.go

@@ -93,7 +93,7 @@ func RunRestore(c context.Context, g glue.Glue, cmdName string, cfg *RestoreConf
 	ctx, cancel := context.WithCancel(c)
 	defer cancel()
 
-	mgr, err := newMgr(ctx, g, cfg.PD, cfg.TLS, cfg.CheckRequirements)
+	mgr, err := NewMgr(ctx, g, cfg.PD, cfg.TLS, cfg.CheckRequirements)
 	if err != nil {
 		return err
 	}
@@ -369,7 +369,7 @@ func RunRestoreTiflashReplica(c context.Context, g glue.Glue, cmdName string, cf
 	ctx, cancel := context.WithCancel(c)
 	defer cancel()
 
-	mgr, err := newMgr(ctx, g, cfg.PD, cfg.TLS, cfg.CheckRequirements)
+	mgr, err := NewMgr(ctx, g, cfg.PD, cfg.TLS, cfg.CheckRequirements)
 	if err != nil {
 		return err
 	}

pkg/task/restore_log.go

@@ -97,7 +97,7 @@ func RunLogRestore(c context.Context, g glue.Glue, cfg *LogRestoreConfig) error
 	ctx, cancel := context.WithCancel(c)
 	defer cancel()
 
-	mgr, err := newMgr(ctx, g, cfg.PD, cfg.TLS, cfg.CheckRequirements)
+	mgr, err := NewMgr(ctx, g, cfg.PD, cfg.TLS, cfg.CheckRequirements)
 	if err != nil {
 		return err
 	}

pkg/task/restore_raw.go

@@ -50,7 +50,7 @@ func RunRestoreRaw(c context.Context, g glue.Glue, cmdName string, cfg *RestoreR
 	ctx, cancel := context.WithCancel(c)
 	defer cancel()
 
-	mgr, err := newMgr(ctx, g, cfg.PD, cfg.TLS, cfg.CheckRequirements)
+	mgr, err := NewMgr(ctx, g, cfg.PD, cfg.TLS, cfg.CheckRequirements)
 	if err != nil {
 		return err
 	}

tests/br_other/run.sh

@@ -95,8 +95,33 @@ fi
 
 # make sure we won't stuck in non-scheduler state, even we send a SIGTERM to it. 
 # give enough time to BR so it can gracefully stop.
-sleep 10
-! curl http://$PD_ADDR/pd/api/v1/config/schedule | grep '"disable": true' 
+sleep 5
+if curl http://$PD_ADDR/pd/api/v1/config/schedule | jq '[."schedulers-v2"][0][0]' | grep -q '"disable": false'
+then
+  echo "TEST: [$TEST_NAME] failed because scheduler has not been removed"
+  exit 1
+fi
+
+pd_settings=5
+# we need reset pd scheduler/config to default
+# until pd has the solution to temporary set these scheduler/configs.
+run_br validate reset-pd-config-as-default
+
+# max-merge-region-size set to default 20
+curl http://$PD_ADDR/pd/api/v1/config/schedule | jq '."max-merge-region-size"' | grep "20" || ((pd_settings--))
+# max-merge-region-keys set to default 200000
+curl http://$PD_ADDR/pd/api/v1/config/schedule | jq '."max-merge-region-keys"' | grep "200000" || ((pd_settings--))
+# balance-region scheduler enabled
+curl http://$PD_ADDR/pd/api/v1/config/schedule | jq '[."schedulers-v2"][0][0]' | grep '"disable": false' || ((pd_settings--))
+# balance-leader scheduler enabled
+curl http://$PD_ADDR/pd/api/v1/config/schedule | jq '[."schedulers-v2"][0][1]' | grep '"disable": false' || ((pd_settings--))
+# hot region scheduler enabled
+curl http://$PD_ADDR/pd/api/v1/config/schedule | jq '[."schedulers-v2"][0][2]' | grep '"disable": false' || ((pd_settings--))
+
+if [ "$pd_settings" -ne "5" ];then
+    echo "TEST: [$TEST_NAME] test validate reset pd config failed!"
+    exit 1
+fi
 
 run_sql "DROP DATABASE $DB;"