[3.1] restore: Make all download error as retryable (#298)

pkg/restore/backoff.go

@@ -60,7 +60,7 @@ func newDownloadSSTBackoffer() utils.Backoffer {
 
 func (bo *importerBackoffer) NextBackoff(err error) time.Duration {
 	switch errors.Cause(err) {
-	case errGrpc, errEpochNotMatch, errIngestFailed:
+	case errGrpc, errEpochNotMatch, errDownloadFailed, errIngestFailed:
 		bo.delayTime = 2 * bo.delayTime
 		bo.attempt--
 	case errRangeIsEmpty, errRewriteRuleNotFound:

pkg/restore/backoff_test.go

@@ -8,6 +8,7 @@ import (
 
 	. "github.com/pingcap/check"
 	"github.com/pingcap/tidb/util/testleak"
+	"go.uber.org/multierr"
 
 	"github.com/pingcap/br/pkg/mock"
 	"github.com/pingcap/br/pkg/utils"
@@ -29,8 +30,28 @@ func (s *testBackofferSuite) TearDownSuite(c *C) {
 	testleak.AfterTest(c)()
 }
 
-func (s *testBackofferSuite) TestImporterBackoffer(c *C) {
+func (s *testBackofferSuite) TestBackoffWithSuccess(c *C) {
 	var counter int
+	backoffer := &newImportSSTBackoffer{attemp: 10, delayTime: time.Nanosecond, maxDelayTime: time.Nanosecond}
+	err := utils.WithRetry(context.Background(), func() error {
+		defer func() { counter++ }()
+		switch counter {
+		case 0:
+			return errGRPC
+		case 1:
+			return errEpochNotMatch
+		case 2:
+			return nil
+		}
+		return nil
+	}, backoffer)
+	c.Assert(counter, Equals, 3)
+	c.Assert(err, IsNil)
+}
+
+func (s *testBackofferSuite) TestBackoffWithFatalError(c *C) {
+	var counter int
+	backoffer := &newImportSSTBackoffer{attemp: 10, delayTime: time.Nanosecond, maxDelayTime: time.Nanosecond}
 	err := utils.WithRetry(context.Background(), func() error {
 		defer func() { counter++ }()
 		switch counter {
@@ -39,23 +60,39 @@ func (s *testBackofferSuite) TestImporterBackoffer(c *C) {
 		case 1:
 			return errEpochNotMatch
 		case 2:
+			return errDownloadFailed
+		case 3:
 			return errRangeIsEmpty
 		}
 		return nil
-	}, newImportSSTBackoffer())
-	c.Assert(counter, Equals, 3)
-	c.Assert(err, Equals, errRangeIsEmpty)
-
-	counter = 0
-	backoffer := importerBackoffer{
-		attempt:      10,
-		delayTime:    time.Nanosecond,
-		maxDelayTime: time.Nanosecond,
-	}
-	err = utils.WithRetry(context.Background(), func() error {
+	}, backoffer)
+	c.Assert(counter, Equals, 4)
+	c.Assert(multierr.Errors(err), DeepEquals, []error{
+		errGRPC,
+		errEpochNotMatch,
+		errDownloadFailed,
+		errRangeIsEmpty,
+	})
+}
+
+func (s *testBackofferSuite) TestBackoffWithRetryableError(c *C) {
+	var counter int
+	backoffer := &newImportSSTBackoffer{attemp: 10, delayTime: time.Nanosecond, maxDelayTime: time.Nanosecond}
+	err := utils.WithRetry(context.Background(), func() error {
 		defer func() { counter++ }()
 		return errEpochNotMatch
 	}, &backoffer)
 	c.Assert(counter, Equals, 10)
-	c.Assert(err, Equals, errEpochNotMatch)
+	c.Assert(multierr.Errors(err), DeepEquals, []error{
+		errEpochNotMatch,
+		errEpochNotMatch,
+		errEpochNotMatch,
+		errEpochNotMatch,
+		errEpochNotMatch,
+		errEpochNotMatch,
+		errEpochNotMatch,
+		errEpochNotMatch,
+		errEpochNotMatch,
+		errEpochNotMatch,
+	})
 }

pkg/restore/import.go

@@ -16,6 +16,7 @@ import (
 	"github.com/pingcap/kvproto/pkg/kvrpcpb"
 	"github.com/pingcap/log"
 	"github.com/pingcap/pd/v3/pkg/codec"
+	"go.uber.org/multierr"
 	"go.uber.org/zap"
 	"google.golang.org/grpc"
 	"google.golang.org/grpc/credentials"
@@ -228,6 +229,7 @@ func (importer *FileImporter) Import(
 
 		log.Debug("scan regions", zap.Stringer("file", file), zap.Int("count", len(regionInfos)))
 		// Try to download and ingest the file in every region
+	regionLoop:
 		for _, regionInfo := range regionInfos {
 			info := regionInfo
 			// Try to download file.
@@ -242,9 +244,12 @@ func (importer *FileImporter) Import(
 				return e
 			}, newDownloadSSTBackoffer())
 			if errDownload != nil {
-				if errDownload == errRewriteRuleNotFound || errDownload == errRangeIsEmpty {
-					// Skip this region
-					continue
+				for _, e := range multierr.Errors(errDownload) {
+					switch e {
+					case errRewriteRuleNotFound, errRangeIsEmpty:
+						// Skip this region
+						continue regionLoop
+					}
 				}
 				log.Error("download file failed",
 					zap.Stringer("file", file),

pkg/utils/retry.go

@@ -5,6 +5,8 @@ package utils
 import (
 	"context"
 	"time"
+
+	"go.uber.org/multierr"
 )
 
 // RetryableFunc presents a retryable opreation
@@ -18,25 +20,28 @@ type Backoffer interface {
 	Attempt() int
 }
 
-// WithRetry retrys a given operation with a backoff policy
+// WithRetry retries a given operation with a backoff policy.
+//
+// Returns nil if `retryableFunc` succeeded at least once. Otherwise, returns a
+// multierr containing all errors encountered.
 func WithRetry(
 	ctx context.Context,
 	retryableFunc RetryableFunc,
 	backoffer Backoffer,
 ) error {
-	var lastErr error
+	var allErrors error
 	for backoffer.Attempt() > 0 {
 		err := retryableFunc()
 		if err != nil {
-			lastErr = err
+			allErrors = multierr.Append(allErrors, err)
 			select {
 			case <-ctx.Done():
-				return lastErr
+				return allErrors
 			case <-time.After(backoffer.NextBackoff(err)):
 			}
 		} else {
 			return nil
 		}
 	}
-	return lastErr
+	return allErrors
 }