From d6e8ba4d45324aff8f1cd3fd814b7d4a96da1f52 Mon Sep 17 00:00:00 2001
From: Jianjun Liao <36503113+Leavrth@users.noreply.github.com>
Date: Mon, 13 Nov 2023 14:50:15 +0800
Subject: [PATCH] br: add more retryable error for br_import (#48529)

close pingcap/tidb#48528
---
 br/pkg/utils/backoff.go | 1 +
 br/pkg/utils/retry.go   | 1 +
 br/tests/br_full/run.sh | 2 +-
 3 files changed, 3 insertions(+), 1 deletion(-)

diff --git a/br/pkg/utils/backoff.go b/br/pkg/utils/backoff.go
index 93e8c6afa52ff..1a9f157807f3e 100644
--- a/br/pkg/utils/backoff.go
+++ b/br/pkg/utils/backoff.go
@@ -131,6 +131,7 @@ func NewDownloadSSTBackoffer() Backoffer {
 }
 
 func (bo *importerBackoffer) NextBackoff(err error) time.Duration {
+	log.Warn("retry to import ssts", zap.Int("attempt", bo.attempt), zap.Error(err))
 	if MessageIsRetryableStorageError(err.Error()) {
 		bo.delayTime = 2 * bo.delayTime
 		bo.attempt--
diff --git a/br/pkg/utils/retry.go b/br/pkg/utils/retry.go
index b4ab0437cf651..8756f2ca33943 100644
--- a/br/pkg/utils/retry.go
+++ b/br/pkg/utils/retry.go
@@ -29,6 +29,7 @@ var retryableServerError = []string{
 	"internalerror",
 	"not read from or written to within the timeout period",
 	"<code>requesttimeout</code>",
+	"end of file before message length reached",
 }
 
 // RetryableFunc presents a retryable operation.
diff --git a/br/tests/br_full/run.sh b/br/tests/br_full/run.sh
index dc785a75423f8..9853f3ea87b88 100755
--- a/br/tests/br_full/run.sh
+++ b/br/tests/br_full/run.sh
@@ -52,7 +52,7 @@ test_log="${TEST_DIR}/${DB}_test.log"
 error_str="not read from or written to within the timeout period"
 unset BR_LOG_TO_TERM
 
-export GO_FAILPOINTS="github.com/pingcap/tidb/br/pkg/backup/backup-storage-error=1*return(\"connection refused\")->1*return(\"InternalError\");github.com/pingcap/tidb/br/pkg/backup/backup-timeout-error=1*return(\"<Code>RequestTimeout</Code>\")->1*return(\"not read from or written to within the timeout period\")"
+export GO_FAILPOINTS="github.com/pingcap/tidb/br/pkg/backup/backup-storage-error=1*return(\"connection refused\")->1*return(\"InternalError\");github.com/pingcap/tidb/br/pkg/backup/backup-timeout-error=1*return(\"<Code>RequestTimeout</Code>\")->1*return(\"not read from or written to within the timeout period\")->1*return(\"end of file before message length reached\")"
 run_br --pd $PD_ADDR backup full -s "local://$TEST_DIR/$DB-lz4" --concurrency 4 --compression lz4 --log-file $test_log
 export GO_FAILPOINTS=""
 size_lz4=$(du -d 0 $TEST_DIR/$DB-lz4 | awk '{print $1}')