lightning: retry for WaitN error caused by store-write-bwlimit (pingc…

…ap#51425)

close pingcap#51383

br/pkg/lightning/common/BUILD.bazel

@@ -138,6 +138,7 @@ go_test(
         "@com_github_tikv_pd_client//http",
         "@org_golang_google_grpc//codes",
         "@org_golang_google_grpc//status",
+        "@org_golang_x_time//rate",
         "@org_uber_go_goleak//:goleak",
         "@org_uber_go_multierr//:multierr",
     ],

br/pkg/lightning/common/retry.go

@@ -40,6 +40,8 @@ var retryableErrorMsgList = []string{
 	// this error happens on when distsql.Checksum calls TiKV
 	// see https://github.com/pingcap/tidb/blob/2c3d4f1ae418881a95686e8b93d4237f2e76eec6/store/copr/coprocessor.go#L941
 	"coprocessor task terminated due to exceeding the deadline",
+	// fix https://github.com/pingcap/tidb/issues/51383
+	"rate: wait",
 }
 
 func isRetryableFromErrorMessage(err error) bool {

br/pkg/lightning/common/retry_test.go

@@ -21,13 +21,15 @@ import (
 	"net"
 	"net/url"
 	"testing"
+	"time"
 
 	"github.com/go-sql-driver/mysql"
 	"github.com/pingcap/errors"
 	tmysql "github.com/pingcap/tidb/pkg/errno"
 	drivererr "github.com/pingcap/tidb/pkg/store/driver/error"
 	"github.com/stretchr/testify/require"
 	"go.uber.org/multierr"
+	"golang.org/x/time/rate"
 	"google.golang.org/grpc/codes"
 	"google.golang.org/grpc/status"
 )
@@ -117,4 +119,13 @@ func TestIsRetryableError(t *testing.T) {
 	require.False(t, IsRetryableError(multierr.Combine(context.Canceled, &net.DNSError{IsTimeout: true})))
 
 	require.True(t, IsRetryableError(errors.New("other error: Coprocessor task terminated due to exceeding the deadline")))
+
+	// error from limiter
+	l := rate.NewLimiter(rate.Limit(1), 1)
+	ctx, cancel := context.WithTimeout(context.Background(), time.Second)
+	defer cancel()
+	// context has 1 second timeout, can't wait for 10 seconds
+	err = l.WaitN(ctx, 10)
+	require.Error(t, err)
+	require.True(t, IsRetryableError(err))
 }