br: provide a storage target option for BR restore checkpoint data

br/pkg/checkpoint/BUILD.bazel

@@ -7,6 +7,7 @@ go_library(
         "checkpoint.go",
         "external_storage.go",
         "log_restore.go",
+        "manager.go",
         "restore.go",
         "storage.go",
         "ticker.go",
@@ -45,7 +46,7 @@ go_test(
     srcs = ["checkpoint_test.go"],
     flaky = True,
     race = "on",
-    shard_count = 9,
+    shard_count = 13,
     deps = [
         ":checkpoint",
         "//br/pkg/gluetidb",

br/pkg/checkpoint/backup.go

@@ -37,8 +37,8 @@ const (
 	CheckpointLockPathForBackup    = CheckpointBackupDir + "/checkpoint.lock"
 )
 
-func flushPositionForBackup() flushPosition {
-	return flushPosition{
+func flushPathForBackup() flushPath {
+	return flushPath{
 		CheckpointDataDir:     CheckpointDataDirForBackup,
 		CheckpointChecksumDir: CheckpointChecksumDirForBackup,
 		CheckpointLockPath:    CheckpointLockPathForBackup,
@@ -57,7 +57,7 @@ func StartCheckpointBackupRunnerForTest(
 	tick time.Duration,
 	timer GlobalTimer,
 ) (*CheckpointRunner[BackupKeyType, BackupValueType], error) {
-	checkpointStorage, err := newExternalCheckpointStorage(ctx, storage, timer)
+	checkpointStorage, err := newExternalCheckpointStorage(ctx, storage, timer, flushPathForBackup())
 	if err != nil {
 		return nil, errors.Trace(err)
 	}
@@ -74,7 +74,7 @@ func StartCheckpointRunnerForBackup(
 	cipher *backuppb.CipherInfo,
 	timer GlobalTimer,
 ) (*CheckpointRunner[BackupKeyType, BackupValueType], error) {
-	checkpointStorage, err := newExternalCheckpointStorage(ctx, storage, timer)
+	checkpointStorage, err := newExternalCheckpointStorage(ctx, storage, timer, flushPathForBackup())
 	if err != nil {
 		return nil, errors.Trace(err)
 	}

br/pkg/checkpoint/checkpoint.go

@@ -41,7 +41,7 @@ import (
 
 const CheckpointDir = "checkpoints"
 
-type flushPosition struct {
+type flushPath struct {
 	CheckpointDataDir     string
 	CheckpointChecksumDir string
 	CheckpointLockPath    string

br/pkg/checkpoint/external_storage.go

@@ -31,8 +31,44 @@ import (
 	"go.uber.org/zap"
 )
 
+const (
+	CheckpointRestoreDirFormat                = CheckpointDir + "/restore-%s"
+	CheckpointDataDirForRestoreFormat         = CheckpointRestoreDirFormat + "/data"
+	CheckpointChecksumDirForRestoreFormat     = CheckpointRestoreDirFormat + "/checksum"
+	CheckpointMetaPathForRestoreFormat        = CheckpointRestoreDirFormat + "/checkpoint.meta"
+	CheckpointProgressPathForRestoreFormat    = CheckpointRestoreDirFormat + "/progress.meta"
+	CheckpointIngestIndexPathForRestoreFormat = CheckpointRestoreDirFormat + "/ingest_index.meta"
+)
+
+func flushPathForRestore(taskName string) flushPath {
+	return flushPath{
+		CheckpointDataDir:     getCheckpointDataDirByName(taskName),
+		CheckpointChecksumDir: getCheckpointChecksumDirByName(taskName),
+	}
+}
+
+func getCheckpointMetaPathByName(taskName string) string {
+	return fmt.Sprintf(CheckpointMetaPathForRestoreFormat, taskName)
+}
+
+func getCheckpointDataDirByName(taskName string) string {
+	return fmt.Sprintf(CheckpointDataDirForRestoreFormat, taskName)
+}
+
+func getCheckpointChecksumDirByName(taskName string) string {
+	return fmt.Sprintf(CheckpointChecksumDirForRestoreFormat, taskName)
+}
+
+func getCheckpointProgressPathByName(taskName string) string {
+	return fmt.Sprintf(CheckpointProgressPathForRestoreFormat, taskName)
+}
+
+func getCheckpointIngestIndexPathByName(taskName string) string {
+	return fmt.Sprintf(CheckpointIngestIndexPathForRestoreFormat, taskName)
+}
+
 type externalCheckpointStorage struct {
-	flushPosition
+	flushPath
 	storage storage.ExternalStorage
 
 	lockId uint64
@@ -43,11 +79,12 @@ func newExternalCheckpointStorage(
 	ctx context.Context,
 	s storage.ExternalStorage,
 	timer GlobalTimer,
+	flushPath flushPath,
 ) (*externalCheckpointStorage, error) {
 	checkpointStorage := &externalCheckpointStorage{
-		flushPosition: flushPositionForBackup(),
-		storage:       s,
-		timer:         timer,
+		flushPath: flushPath,
+		storage:   s,
+		timer:     timer,
 	}
 	if timer != nil {
 		if err := checkpointStorage.initialLock(ctx); err != nil {

br/pkg/checkpoint/log_restore.go

@@ -21,10 +21,8 @@ import (
 
 	"github.com/pingcap/errors"
 	"github.com/pingcap/tidb/br/pkg/glue"
-	"github.com/pingcap/tidb/pkg/domain"
 	"github.com/pingcap/tidb/pkg/meta/model"
 	"github.com/pingcap/tidb/pkg/parser/ast"
-	"github.com/pingcap/tidb/pkg/util/sqlexec"
 )
 
 type LogRestoreKeyType = string
@@ -98,31 +96,21 @@ func newTableCheckpointStorage(se glue.Session, checkpointDBName string) *tableC
 // only for test
 func StartCheckpointLogRestoreRunnerForTest(
 	ctx context.Context,
-	se glue.Session,
 	tick time.Duration,
+	manager LogMetaManagerT,
 ) (*CheckpointRunner[LogRestoreKeyType, LogRestoreValueType], error) {
-	runner := newCheckpointRunner[LogRestoreKeyType, LogRestoreValueType](
-		newTableCheckpointStorage(se, LogRestoreCheckpointDatabaseName),
-		nil, valueMarshalerForLogRestore)
-
-	runner.startCheckpointMainLoop(ctx, tick, tick, 0, defaultRetryDuration)
-	return runner, nil
+	cfg := DefaultTickDurationConfig()
+	cfg.tickDurationForChecksum = tick
+	cfg.tickDurationForFlush = tick
+	return manager.StartCheckpointRunner(ctx, cfg, valueMarshalerForLogRestore)
 }
 
 // Notice that the session is owned by the checkpoint runner, and it will be also closed by it.
 func StartCheckpointRunnerForLogRestore(
 	ctx context.Context,
-	se glue.Session,
+	manager LogMetaManagerT,
 ) (*CheckpointRunner[LogRestoreKeyType, LogRestoreValueType], error) {
-	runner := newCheckpointRunner[LogRestoreKeyType, LogRestoreValueType](
-		newTableCheckpointStorage(se, LogRestoreCheckpointDatabaseName),
-		nil, valueMarshalerForLogRestore)
-
-	// for restore, no need to set lock
-	runner.startCheckpointMainLoop(
-		ctx,
-		defaultTickDurationForFlush, defaultTickDurationForChecksum, 0, defaultRetryDuration)
-	return runner, nil
+	return manager.StartCheckpointRunner(ctx, DefaultTickDurationConfig(), valueMarshalerForLogRestore)
 }
 
 func AppendRangeForLogRestore(
@@ -145,16 +133,6 @@ func AppendRangeForLogRestore(
 	})
 }
 
-// load the whole checkpoint range data and retrieve the metadata of restored ranges
-// and return the total time cost in the past executions
-func LoadCheckpointDataForLogRestore[K KeyType, V ValueType](
-	ctx context.Context,
-	execCtx sqlexec.RestrictedSQLExecutor,
-	fn func(K, V),
-) (time.Duration, error) {
-	return selectCheckpointData(ctx, execCtx, LogRestoreCheckpointDatabaseName, fn)
-}
-
 type CheckpointMetadataForLogRestore struct {
 	UpstreamClusterID uint64 `json:"upstream-cluster-id"`
 	RestoredTS        uint64 `json:"restored-ts"`
@@ -165,35 +143,6 @@ type CheckpointMetadataForLogRestore struct {
 	TiFlashItems map[int64]model.TiFlashReplicaInfo `json:"tiflash-recorder,omitempty"`
 }
 
-func LoadCheckpointMetadataForLogRestore(
-	ctx context.Context,
-	execCtx sqlexec.RestrictedSQLExecutor,
-) (*CheckpointMetadataForLogRestore, error) {
-	m := &CheckpointMetadataForLogRestore{}
-	err := selectCheckpointMeta(ctx, execCtx, LogRestoreCheckpointDatabaseName, checkpointMetaTableName, m)
-	return m, err
-}
-
-func SaveCheckpointMetadataForLogRestore(
-	ctx context.Context,
-	se glue.Session,
-	meta *CheckpointMetadataForLogRestore,
-) error {
-	err := initCheckpointTable(ctx, se, LogRestoreCheckpointDatabaseName, []string{checkpointDataTableName})
-	if err != nil {
-		return errors.Trace(err)
-	}
-	return insertCheckpointMeta(ctx, se, LogRestoreCheckpointDatabaseName, checkpointMetaTableName, meta)
-}
-
-func ExistsLogRestoreCheckpointMetadata(
-	ctx context.Context,
-	dom *domain.Domain,
-) bool {
-	return dom.InfoSchema().
-		TableExists(ast.NewCIStr(LogRestoreCheckpointDatabaseName), ast.NewCIStr(checkpointMetaTableName))
-}
-
 // RestoreProgress is a progress type for snapshot + log restore.
 //
 // Before the id-maps is persisted into external storage, the snapshot restore and
@@ -229,34 +178,9 @@ type CheckpointProgress struct {
 	Progress RestoreProgress `json:"progress"`
 }
 
-func LoadCheckpointProgress(
-	ctx context.Context,
-	execCtx sqlexec.RestrictedSQLExecutor,
-) (*CheckpointProgress, error) {
-	m := &CheckpointProgress{}
-	err := selectCheckpointMeta(ctx, execCtx, LogRestoreCheckpointDatabaseName, checkpointProgressTableName, m)
-	return m, errors.Trace(err)
-}
-
-func SaveCheckpointProgress(
-	ctx context.Context,
-	se glue.Session,
-	meta *CheckpointProgress,
-) error {
-	return insertCheckpointMeta(ctx, se, LogRestoreCheckpointDatabaseName, checkpointProgressTableName, meta)
-}
-
-func ExistsCheckpointProgress(
-	ctx context.Context,
-	dom *domain.Domain,
-) bool {
-	return dom.InfoSchema().
-		TableExists(ast.NewCIStr(LogRestoreCheckpointDatabaseName), ast.NewCIStr(checkpointProgressTableName))
-}
-
 // CheckpointTaskInfoForLogRestore is tied to a specific cluster.
 // It represents the last restore task executed in this cluster.
-type CheckpointTaskInfoForLogRestore struct {
+type TaskInfoForLogRestore struct {
 	Metadata            *CheckpointMetadataForLogRestore
 	HasSnapshotMetadata bool
 	// the progress for this task
@@ -265,32 +189,44 @@ type CheckpointTaskInfoForLogRestore struct {
 
 func TryToGetCheckpointTaskInfo(
 	ctx context.Context,
-	dom *domain.Domain,
-	execCtx sqlexec.RestrictedSQLExecutor,
-) (*CheckpointTaskInfoForLogRestore, error) {
+	snapshotManager SnapshotMetaManagerT,
+	logManager LogMetaManagerT,
+) (*TaskInfoForLogRestore, error) {
 	var (
 		metadata *CheckpointMetadataForLogRestore
 		progress RestoreProgress
-		err      error
+
+		hasSnapshotMetadata bool = false
 	)
 	// get the progress
-	if ExistsCheckpointProgress(ctx, dom) {
-		checkpointProgress, err := LoadCheckpointProgress(ctx, execCtx)
+	if exists, err := logManager.ExistsCheckpointProgress(ctx); err != nil {
+		return nil, errors.Trace(err)
+	} else if exists {
+		checkpointProgress, err := logManager.LoadCheckpointProgress(ctx)
 		if err != nil {
 			return nil, errors.Trace(err)
 		}
 		progress = checkpointProgress.Progress
 	}
 	// get the checkpoint metadata
-	if ExistsLogRestoreCheckpointMetadata(ctx, dom) {
-		metadata, err = LoadCheckpointMetadataForLogRestore(ctx, execCtx)
+	if exists, err := logManager.ExistsCheckpointMetadata(ctx); err != nil {
+		return nil, errors.Trace(err)
+	} else if exists {
+		metadata, err = logManager.LoadCheckpointMetadata(ctx)
 		if err != nil {
 			return nil, errors.Trace(err)
 		}
 	}
-	hasSnapshotMetadata := ExistsSstRestoreCheckpoint(ctx, dom, SnapshotRestoreCheckpointDatabaseName)
+	// exists the snapshot checkpoint metadata
+	if snapshotManager != nil {
+		existsSnapshotMetadata, err := snapshotManager.ExistsCheckpointMetadata(ctx)
+		if err != nil {
+			return nil, errors.Trace(err)
+		}
+		hasSnapshotMetadata = existsSnapshotMetadata
+	}
 
-	return &CheckpointTaskInfoForLogRestore{
+	return &TaskInfoForLogRestore{
 		Metadata:            metadata,
 		HasSnapshotMetadata: hasSnapshotMetadata,
 		Progress:            progress,
@@ -312,30 +248,3 @@ type CheckpointIngestIndexRepairSQL struct {
 type CheckpointIngestIndexRepairSQLs struct {
 	SQLs []CheckpointIngestIndexRepairSQL
 }
-
-func LoadCheckpointIngestIndexRepairSQLs(
-	ctx context.Context,
-	execCtx sqlexec.RestrictedSQLExecutor,
-) (*CheckpointIngestIndexRepairSQLs, error) {
-	m := &CheckpointIngestIndexRepairSQLs{}
-	err := selectCheckpointMeta(ctx, execCtx, LogRestoreCheckpointDatabaseName, checkpointIngestTableName, m)
-	return m, errors.Trace(err)
-}
-
-func ExistsCheckpointIngestIndexRepairSQLs(ctx context.Context, dom *domain.Domain) bool {
-	return dom.InfoSchema().
-		TableExists(ast.NewCIStr(LogRestoreCheckpointDatabaseName), ast.NewCIStr(checkpointIngestTableName))
-}
-
-func SaveCheckpointIngestIndexRepairSQLs(
-	ctx context.Context,
-	se glue.Session,
-	meta *CheckpointIngestIndexRepairSQLs,
-) error {
-	return insertCheckpointMeta(ctx, se, LogRestoreCheckpointDatabaseName, checkpointIngestTableName, meta)
-}
-
-func RemoveCheckpointDataForLogRestore(ctx context.Context, dom *domain.Domain, se glue.Session) error {
-	return dropCheckpointTables(ctx, dom, se, LogRestoreCheckpointDatabaseName,
-		[]string{checkpointDataTableName, checkpointMetaTableName, checkpointProgressTableName, checkpointIngestTableName})
-}