cherry-pick-7.5: br: restore checksum shouldn't rely on backup checksum (#56712) (#58225)

Author: Tristan1900

br/cmd/br/backup.go

@@ -21,10 +21,11 @@ import (
 
 func runBackupCommand(command *cobra.Command, cmdName string) error {
 	cfg := task.BackupConfig{Config: task.Config{LogProgress: HasLogFile()}}
-	if err := cfg.ParseFromFlags(command.Flags()); err != nil {
+	if err := cfg.ParseFromFlags(command.Flags(), false); err != nil {
 		command.SilenceUsage = false
 		return errors.Trace(err)
 	}
+	overrideDefaultBackupConfigIfNeeded(&cfg, command)
 
 	if err := metricsutil.RegisterMetricsForBR(cfg.PD, cfg.KeyspaceName); err != nil {
 		return errors.Trace(err)
@@ -206,3 +207,10 @@ func newTxnBackupCommand() *cobra.Command {
 	task.DefineTxnBackupFlags(command)
 	return command
 }
+
+func overrideDefaultBackupConfigIfNeeded(config *task.BackupConfig, cmd *cobra.Command) {
+	// override only if flag not set by user
+	if !cmd.Flags().Changed(task.FlagChecksum) {
+		config.Checksum = false
+	}
+}

br/cmd/br/cmd.go

@@ -79,8 +79,8 @@ func timestampLogFileName() string {
 	return filepath.Join(os.TempDir(), time.Now().Format("br.log.2006-01-02T15.04.05Z0700"))
 }
 
-// AddFlags adds flags to the given cmd.
-func AddFlags(cmd *cobra.Command) {
+// DefineCommonFlags defines the common flags for all BR cmd operation.
+func DefineCommonFlags(cmd *cobra.Command) {
 	cmd.Version = build.Info()
 	cmd.Flags().BoolP(flagVersion, flagVersionShort, false, "Display version information about BR")
 	cmd.SetVersionTemplate("{{printf \"%s\" .Version}}\n")
@@ -97,6 +97,8 @@ func AddFlags(cmd *cobra.Command) {
 		"Set whether to redact sensitive info in log")
 	cmd.PersistentFlags().String(FlagStatusAddr, "",
 		"Set the HTTP listening address for the status report service. Set to empty string to disable")
+
+	// defines BR task common flags, this is shared by cmd and sql(brie)
 	task.DefineCommonFlags(cmd.PersistentFlags())
 
 	cmd.PersistentFlags().StringP(FlagSlowLogFile, "", "",

br/cmd/br/main.go

@@ -20,7 +20,7 @@ func main() {
 		TraverseChildren: true,
 		SilenceUsage:     true,
 	}
-	AddFlags(rootCmd)
+	DefineCommonFlags(rootCmd)
 	SetDefaultContext(ctx)
 	rootCmd.AddCommand(
 		NewDebugCommand(),

br/cmd/br/restore.go

@@ -23,7 +23,7 @@ import (
 
 func runRestoreCommand(command *cobra.Command, cmdName string) error {
 	cfg := task.RestoreConfig{Config: task.Config{LogProgress: HasLogFile()}}
-	if err := cfg.ParseFromFlags(command.Flags()); err != nil {
+	if err := cfg.ParseFromFlags(command.Flags(), false); err != nil {
 		command.SilenceUsage = false
 		return errors.Trace(err)
 	}

br/pkg/backup/schema.go

@@ -104,7 +104,7 @@ func (ss *Schemas) BackupSchemas(
 		}
 
 		var checksum *checkpoint.ChecksumItem
-		var exists bool = false
+		var exists = false
 		if ss.checkpointChecksum != nil && schema.tableInfo != nil {
 			checksum, exists = ss.checkpointChecksum[schema.tableInfo.ID]
 		}
@@ -143,7 +143,7 @@ func (ss *Schemas) BackupSchemas(
 							zap.Uint64("Crc64Xor", schema.crc64xor),
 							zap.Uint64("TotalKvs", schema.totalKvs),
 							zap.Uint64("TotalBytes", schema.totalBytes),
-							zap.Duration("calculate-take", calculateCost))
+							zap.Duration("TimeTaken", calculateCost))
 					}
 				}
 				if statsHandle != nil {

br/pkg/metautil/metafile.go

@@ -156,11 +156,6 @@ type Table struct {
 	Stats           *util.JSONTable
 }
 
-// NoChecksum checks whether the table has a calculated checksum.
-func (tbl *Table) NoChecksum() bool {
-	return tbl.Crc64Xor == 0 && tbl.TotalKvs == 0 && tbl.TotalBytes == 0
-}
-
 // MetaReader wraps a reader to read both old and new version of backupmeta.
 type MetaReader struct {
 	storage    storage.ExternalStorage
@@ -225,14 +220,38 @@ func (reader *MetaReader) readDataFiles(ctx context.Context, output func(*backup
 }
 
 // ArchiveSize return the size of Archive data
-func (*MetaReader) ArchiveSize(_ context.Context, files []*backuppb.File) uint64 {
+func ArchiveSize(files []*backuppb.File) uint64 {
 	total := uint64(0)
 	for _, file := range files {
 		total += file.Size_
 	}
 	return total
 }
 
+type ChecksumStats struct {
+	Crc64Xor   uint64
+	TotalKvs   uint64
+	TotalBytes uint64
+}
+
+func (stats ChecksumStats) ChecksumExists() bool {
+	if stats.Crc64Xor == 0 && stats.TotalKvs == 0 && stats.TotalBytes == 0 {
+		return false
+	}
+	return true
+}
+
+// CalculateChecksumStatsOnFiles returns the ChecksumStats for the given files
+func CalculateChecksumStatsOnFiles(files []*backuppb.File) ChecksumStats {
+	var stats ChecksumStats
+	for _, file := range files {
+		stats.Crc64Xor ^= file.Crc64Xor
+		stats.TotalKvs += file.TotalKvs
+		stats.TotalBytes += file.TotalBytes
+	}
+	return stats
+}
+
 // ReadDDLs reads the ddls from the backupmeta.
 // This function is compatible with the old backupmeta.
 func (reader *MetaReader) ReadDDLs(ctx context.Context) ([]byte, error) {

br/pkg/restore/client.go

@@ -1696,7 +1696,7 @@ func (rc *Client) GoValidateChecksum(
 			elapsed := time.Since(start)
 			summary.CollectSuccessUnit("table checksum", 1, elapsed)
 		}()
-		err := rc.execChecksum(c, tbl, kvClient, concurrency)
+		err := rc.execAndValidateChecksum(c, tbl, kvClient, concurrency)
 		if err != nil {
 			return errors.Trace(err)
 		}
@@ -1708,7 +1708,7 @@ func (rc *Client) GoValidateChecksum(
 	return outCh
 }
 
-func (rc *Client) execChecksum(
+func (rc *Client) execAndValidateChecksum(
 	ctx context.Context,
 	tbl *CreatedTable,
 	kvClient kv.Client,
@@ -1719,13 +1719,14 @@ func (rc *Client) execChecksum(
 		zap.String("table", tbl.OldTable.Info.Name.O),
 	)
 
-	if tbl.OldTable.NoChecksum() {
+	expectedChecksumStats := metautil.CalculateChecksumStatsOnFiles(tbl.OldTable.Files)
+	if !expectedChecksumStats.ChecksumExists() {
 		logger.Warn("table has no checksum, skipping checksum")
 		return nil
 	}
 
 	if span := opentracing.SpanFromContext(ctx); span != nil && span.Tracer() != nil {
-		span1 := span.Tracer().StartSpan("Client.execChecksum", opentracing.ChildOf(span.Context()))
+		span1 := span.Tracer().StartSpan("Client.execAndValidateChecksum", opentracing.ChildOf(span.Context()))
 		defer span1.Finish()
 		ctx = opentracing.ContextWithSpan(ctx, span1)
 	}
@@ -1765,21 +1766,24 @@ func (rc *Client) execChecksum(
 			}
 		}
 	}
-	table := tbl.OldTable
-	if item.Crc64xor != table.Crc64Xor ||
-		item.TotalKvs != table.TotalKvs ||
-		item.TotalBytes != table.TotalBytes {
+	checksumMatch := item.Crc64xor == expectedChecksumStats.Crc64Xor &&
+		item.TotalKvs == expectedChecksumStats.TotalKvs &&
+		item.TotalBytes == expectedChecksumStats.TotalBytes
+	failpoint.Inject("full-restore-validate-checksum", func(_ failpoint.Value) {
+		checksumMatch = false
+	})
+	if !checksumMatch {
 		logger.Error("failed in validate checksum",
-			zap.Uint64("origin tidb crc64", table.Crc64Xor),
+			zap.Uint64("expected tidb crc64", expectedChecksumStats.Crc64Xor),
 			zap.Uint64("calculated crc64", item.Crc64xor),
-			zap.Uint64("origin tidb total kvs", table.TotalKvs),
+			zap.Uint64("expected tidb total kvs", expectedChecksumStats.TotalKvs),
 			zap.Uint64("calculated total kvs", item.TotalKvs),
-			zap.Uint64("origin tidb total bytes", table.TotalBytes),
+			zap.Uint64("expected tidb total bytes", expectedChecksumStats.TotalBytes),
 			zap.Uint64("calculated total bytes", item.TotalBytes),
 		)
 		return errors.Annotate(berrors.ErrRestoreChecksumMismatch, "failed to validate checksum")
 	}
-	logger.Info("success in validate checksum")
+	logger.Info("success in validating checksum")
 	return nil
 }
 

br/pkg/task/backup.go

@@ -42,7 +42,6 @@ import (
 	"github.com/spf13/pflag"
 	"github.com/tikv/client-go/v2/oracle"
 	kvutil "github.com/tikv/client-go/v2/util"
-	"go.uber.org/multierr"
 	"go.uber.org/zap"
 )
 
@@ -160,7 +159,7 @@ func DefineBackupFlags(flags *pflag.FlagSet) {
 }
 
 // ParseFromFlags parses the backup-related flags from the flag set.
-func (cfg *BackupConfig) ParseFromFlags(flags *pflag.FlagSet) error {
+func (cfg *BackupConfig) ParseFromFlags(flags *pflag.FlagSet, skipCommonConfig bool) error {
 	timeAgo, err := flags.GetDuration(flagBackupTimeago)
 	if err != nil {
 		return errors.Trace(err)
@@ -213,9 +212,13 @@ func (cfg *BackupConfig) ParseFromFlags(flags *pflag.FlagSet) error {
 	}
 	cfg.CompressionConfig = *compressionCfg
 
-	if err = cfg.Config.ParseFromFlags(flags); err != nil {
-		return errors.Trace(err)
+	// parse common flags if needed
+	if !skipCommonConfig {
+		if err = cfg.Config.ParseFromFlags(flags); err != nil {
+			return errors.Trace(err)
+		}
 	}
+
 	cfg.RemoveSchedulers, err = flags.GetBool(flagRemoveSchedulers)
 	if err != nil {
 		return errors.Trace(err)
@@ -790,18 +793,15 @@ func ParseTSString(ts string, tzCheck bool) (uint64, error) {
 	return oracle.GoTimeToTS(t1), nil
 }
 
-func DefaultBackupConfig() BackupConfig {
+func DefaultBackupConfig(commonConfig Config) BackupConfig {
 	fs := pflag.NewFlagSet("dummy", pflag.ContinueOnError)
-	DefineCommonFlags(fs)
 	DefineBackupFlags(fs)
 	cfg := BackupConfig{}
-	err := multierr.Combine(
-		cfg.ParseFromFlags(fs),
-		cfg.Config.ParseFromFlags(fs),
-	)
+	err := cfg.ParseFromFlags(fs, true)
 	if err != nil {
-		log.Panic("infallible operation failed.", zap.Error(err))
+		log.Panic("failed to parse backup flags to config", zap.Error(err))
 	}
+	cfg.Config = commonConfig
 	return cfg
 }
 

br/pkg/task/common.go

@@ -64,7 +64,7 @@ const (
 	flagRateLimit           = "ratelimit"
 	flagRateLimitUnit       = "ratelimit-unit"
 	flagConcurrency         = "concurrency"
-	flagChecksum            = "checksum"
+	FlagChecksum            = "checksum"
 	flagFilter              = "filter"
 	flagCaseSensitive       = "case-sensitive"
 	flagRemoveTiFlash       = "remove-tiflash"
@@ -273,7 +273,7 @@ func DefineCommonFlags(flags *pflag.FlagSet) {
 	flags.Uint(flagChecksumConcurrency, variable.DefChecksumTableConcurrency, "The concurrency of checksumming in one table")
 
 	flags.Uint64(flagRateLimit, unlimited, "The rate limit of the task, MB/s per node")
-	flags.Bool(flagChecksum, true, "Run checksum at end of task")
+	flags.Bool(FlagChecksum, true, "Run checksum at end of task")
 	flags.Bool(flagRemoveTiFlash, true,
 		"Remove TiFlash replicas before backup or restore, for unsupported versions of TiFlash")
 
@@ -318,7 +318,7 @@ func DefineCommonFlags(flags *pflag.FlagSet) {
 
 // HiddenFlagsForStream temporary hidden flags that stream cmd not support.
 func HiddenFlagsForStream(flags *pflag.FlagSet) {
-	_ = flags.MarkHidden(flagChecksum)
+	_ = flags.MarkHidden(FlagChecksum)
 	_ = flags.MarkHidden(flagChecksumConcurrency)
 	_ = flags.MarkHidden(flagRateLimit)
 	_ = flags.MarkHidden(flagRateLimitUnit)
@@ -505,7 +505,7 @@ func (cfg *Config) ParseFromFlags(flags *pflag.FlagSet) error {
 		return errors.Trace(err)
 	}
 
-	if cfg.Checksum, err = flags.GetBool(flagChecksum); err != nil {
+	if cfg.Checksum, err = flags.GetBool(FlagChecksum); err != nil {
 		return errors.Trace(err)
 	}
 	if cfg.ChecksumConcurrency, err = flags.GetUint(flagChecksumConcurrency); err != nil {
@@ -618,6 +618,11 @@ func (cfg *Config) ParseFromFlags(flags *pflag.FlagSet) error {
 	return cfg.normalizePDURLs()
 }
 
+// OverrideDefaultForBackup override common config for backup tasks
+func (cfg *Config) OverrideDefaultForBackup() {
+	cfg.Checksum = false
+}
+
 // NewMgr creates a new mgr at the given PD address.
 func NewMgr(ctx context.Context,
 	g glue.Glue, pds []string,

br/pkg/task/common_test.go

@@ -232,8 +232,10 @@ func expectedDefaultConfig() Config {
 }
 
 func expectedDefaultBackupConfig() BackupConfig {
+	defaultConfig := expectedDefaultConfig()
+	defaultConfig.Checksum = false
 	return BackupConfig{
-		Config: expectedDefaultConfig(),
+		Config: defaultConfig,
 		GCTTL:  utils.DefaultBRGCSafePointTTL,
 		CompressionConfig: CompressionConfig{
 			CompressionType: backup.CompressionType_ZSTD,
@@ -270,13 +272,16 @@ func TestDefault(t *testing.T) {
 }
 
 func TestDefaultBackup(t *testing.T) {
-	def := DefaultBackupConfig()
+	commonConfig := DefaultConfig()
+	commonConfig.OverrideDefaultForBackup()
+	def := DefaultBackupConfig(commonConfig)
 	defaultConfig := expectedDefaultBackupConfig()
 	require.Equal(t, defaultConfig, def)
 }
 
 func TestDefaultRestore(t *testing.T) {
-	def := DefaultRestoreConfig()
+	commonConfig := DefaultConfig()
+	def := DefaultRestoreConfig(commonConfig)
 	defaultConfig := expectedDefaultRestoreConfig()
 	require.Equal(t, defaultConfig, def)
 }