ddl: use a global owner manager instance for DDL, to avoid owner change (#57179)

ref #57185

Author: D3Hunter

br/pkg/conn/BUILD.bazel

@@ -15,6 +15,7 @@ go_library(
         "//br/pkg/utils",
         "//br/pkg/version",
         "//pkg/config",
+        "//pkg/ddl",
         "//pkg/domain",
         "//pkg/kv",
         "@com_github_docker_go_units//:go-units",

br/pkg/conn/conn.go

@@ -29,6 +29,7 @@ import (
 	"github.com/pingcap/tidb/br/pkg/utils"
 	"github.com/pingcap/tidb/br/pkg/version"
 	"github.com/pingcap/tidb/pkg/config"
+	"github.com/pingcap/tidb/pkg/ddl"
 	"github.com/pingcap/tidb/pkg/domain"
 	"github.com/pingcap/tidb/pkg/kv"
 	"github.com/tikv/client-go/v2/oracle"
@@ -194,6 +195,9 @@ func NewMgr(
 		return nil, errors.Trace(err)
 	}
 
+	if config.GetGlobalConfig().Store != "tikv" {
+		config.GetGlobalConfig().Store = "tikv"
+	}
 	// Disable GC because TiDB enables GC already.
 	path := fmt.Sprintf(
 		"tikv://%s?disableGC=true&keyspaceName=%s",
@@ -292,6 +296,7 @@ func (mgr *Mgr) Close() {
 		if mgr.dom != nil {
 			mgr.dom.Close()
 		}
+		ddl.CloseOwnerManager()
 		tikv.StoreShuttingDown(1)
 		_ = mgr.storage.Close()
 	}

br/pkg/gluetidb/glue.go

@@ -107,6 +107,9 @@ func (g Glue) startDomainAsNeeded(store kv.Storage) error {
 	if existDom != nil {
 		return nil
 	}
+	if err := ddl.StartOwnerManager(context.Background(), store); err != nil {
+		return errors.Trace(err)
+	}
 	dom, err := session.GetDomain(store)
 	if err != nil {
 		return err

br/pkg/task/stream.go

@@ -920,7 +920,11 @@ func RunStreamAdvancer(c context.Context, g glue.Glue, cmdName string, cfg *Stre
 	env := streamhelper.CliEnv(mgr.StoreManager, mgr.GetStore(), etcdCLI)
 	advancer := streamhelper.NewCheckpointAdvancer(env)
 	advancer.UpdateConfig(cfg.AdvancerCfg)
-	advancerd := daemon.New(advancer, streamhelper.OwnerManagerForLogBackup(ctx, etcdCLI), cfg.AdvancerCfg.TickDuration)
+	ownerMgr := streamhelper.OwnerManagerForLogBackup(ctx, etcdCLI)
+	defer func() {
+		ownerMgr.Close()
+	}()
+	advancerd := daemon.New(advancer, ownerMgr, cfg.AdvancerCfg.TickDuration)
 	loop, err := advancerd.Begin(ctx)
 	if err != nil {
 		return err

cmd/benchdb/BUILD.bazel

@@ -6,6 +6,8 @@ go_library(
     importpath = "github.com/pingcap/tidb/cmd/benchdb",
     visibility = ["//visibility:private"],
     deps = [
+        "//pkg/config",
+        "//pkg/ddl",
         "//pkg/parser/terror",
         "//pkg/session",
         "//pkg/session/types",

cmd/benchdb/main.go

@@ -24,6 +24,8 @@ import (
 	"time"
 
 	"github.com/pingcap/log"
+	"github.com/pingcap/tidb/pkg/config"
+	"github.com/pingcap/tidb/pkg/ddl"
 	"github.com/pingcap/tidb/pkg/parser/terror"
 	"github.com/pingcap/tidb/pkg/session"
 	sessiontypes "github.com/pingcap/tidb/pkg/session/types"
@@ -96,11 +98,16 @@ func newBenchDB() *benchDB {
 	// Create TiKV store and disable GC as we will trigger GC manually.
 	store, err := store.New("tikv://" + *addr + "?disableGC=true")
 	terror.MustNil(err)
+	// maybe close below components, but it's for test anyway.
+	ctx := context.Background()
+	config.GetGlobalConfig().Store = "tikv"
+	err = ddl.StartOwnerManager(ctx, store)
+	terror.MustNil(err)
 	_, err = session.BootstrapSession(store)
 	terror.MustNil(err)
 	se, err := session.CreateSession(store)
 	terror.MustNil(err)
-	_, err = se.ExecuteInternal(context.Background(), "use test")
+	_, err = se.ExecuteInternal(ctx, "use test")
 	terror.MustNil(err)
 
 	return &benchDB{

cmd/ddltest/BUILD.bazel

@@ -14,7 +14,9 @@ go_test(
     race = "on",
     shard_count = 6,
     deps = [
+        "//dumpling/context",
         "//pkg/config",
+        "//pkg/ddl",
         "//pkg/domain",
         "//pkg/kv",
         "//pkg/parser/model",

cmd/ddltest/ddl_test.go

@@ -33,7 +33,9 @@ import (
 	_ "github.com/go-sql-driver/mysql"
 	"github.com/pingcap/errors"
 	"github.com/pingcap/log"
+	"github.com/pingcap/tidb/dumpling/context"
 	"github.com/pingcap/tidb/pkg/config"
+	"github.com/pingcap/tidb/pkg/ddl"
 	"github.com/pingcap/tidb/pkg/domain"
 	"github.com/pingcap/tidb/pkg/kv"
 	"github.com/pingcap/tidb/pkg/parser/model"
@@ -95,12 +97,13 @@ func createDDLSuite(t *testing.T) (s *ddlSuite) {
 
 	s.quit = make(chan struct{})
 
+	config.GetGlobalConfig().Store = "tikv"
 	s.store, err = store.New(fmt.Sprintf("tikv://%s%s", *etcd, *tikvPath))
 	require.NoError(t, err)
 
 	// Make sure the schema lease of this session is equal to other TiDB servers'.
 	session.SetSchemaLease(time.Duration(*lease) * time.Second)
-
+	require.NoError(t, ddl.StartOwnerManager(context.Background(), s.store))
 	s.dom, err = session.BootstrapSession(s.store)
 	require.NoError(t, err)
 
@@ -118,6 +121,7 @@ func createDDLSuite(t *testing.T) (s *ddlSuite) {
 	err = domain.GetDomain(s.ctx).DDL().Stop()
 	require.NoError(t, err)
 	config.GetGlobalConfig().Instance.TiDBEnableDDL.Store(false)
+	ddl.CloseOwnerManager()
 	session.ResetStoreForWithTiKVTest(s.store)
 	s.dom.Close()
 	require.NoError(t, s.store.Close())

cmd/tidb-server/main.go

@@ -317,7 +317,7 @@ func main() {
 	keyspaceName := keyspace.GetKeyspaceNameBySettings()
 	executor.Start()
 	resourcemanager.InstanceResourceManager.Start()
-	storage, dom := createStoreAndDomain(keyspaceName)
+	storage, dom := createStoreDDLOwnerMgrAndDomain(keyspaceName)
 	svr := createServer(storage, dom)
 
 	exited := make(chan struct{})
@@ -397,7 +397,7 @@ func registerStores() {
 	terror.MustNil(err)
 }
 
-func createStoreAndDomain(keyspaceName string) (kv.Storage, *domain.Domain) {
+func createStoreDDLOwnerMgrAndDomain(keyspaceName string) (kv.Storage, *domain.Domain) {
 	cfg := config.GetGlobalConfig()
 	var fullPath string
 	if keyspaceName == "" {
@@ -411,6 +411,8 @@ func createStoreAndDomain(keyspaceName string) (kv.Storage, *domain.Domain) {
 	copr.GlobalMPPFailedStoreProber.Run()
 	mppcoordmanager.InstanceMPPCoordinatorManager.Run()
 	// Bootstrap a session to load information schema.
+	err = ddl.StartOwnerManager(context.Background(), storage)
+	terror.MustNil(err)
 	dom, err := session.BootstrapSession(storage)
 	terror.MustNil(err)
 	return storage, dom
@@ -859,7 +861,7 @@ func createServer(storage kv.Storage, dom *domain.Domain) *server.Server {
 	svr, err := server.NewServer(cfg, driver)
 	// Both domain and storage have started, so we have to clean them before exiting.
 	if err != nil {
-		closeDomainAndStorage(storage, dom)
+		closeDDLOwnerMgrDomainAndStorage(storage, dom)
 		log.Fatal("failed to create the server", zap.Error(err), zap.Stack("stack"))
 	}
 	svr.SetDomain(dom)
@@ -893,9 +895,10 @@ func setupTracing() {
 	opentracing.SetGlobalTracer(tracer)
 }
 
-func closeDomainAndStorage(storage kv.Storage, dom *domain.Domain) {
+func closeDDLOwnerMgrDomainAndStorage(storage kv.Storage, dom *domain.Domain) {
 	tikv.StoreShuttingDown(1)
 	dom.Close()
+	ddl.CloseOwnerManager()
 	copr.GlobalMPPFailedStoreProber.Stop()
 	mppcoordmanager.InstanceMPPCoordinatorManager.Stop()
 	err := storage.Close()
@@ -918,7 +921,7 @@ func cleanup(svr *server.Server, storage kv.Storage, dom *domain.Domain) {
 	// See https://github.com/pingcap/tidb/issues/40038 for details.
 	svr.KillSysProcesses()
 	plugin.Shutdown(context.Background())
-	closeDomainAndStorage(storage, dom)
+	closeDDLOwnerMgrDomainAndStorage(storage, dom)
 	disk.CleanUp()
 	closeStmtSummary()
 	topsql.Close()

pkg/autoid_service/autoid.go

@@ -379,8 +379,8 @@ func MockForTest(store kv.Storage) autoid.AutoIDAllocClient {
 
 // Close closes the Service and clean up resource.
 func (s *Service) Close() {
-	if s.leaderShip != nil && s.leaderShip.IsOwner() {
-		s.leaderShip.Cancel()
+	if s.leaderShip != nil {
+		s.leaderShip.Close()
 	}
 }