docker:try to reconnect when containerd grpc return unexpected EOF

(cherry picked from commit bb19128a08aa2355d23555925a14a3733d173b64)

VERSION-vendor

@@ -1 +1 @@
-18.09.0.340
+18.09.0.341

docker.spec

@@ -1,6 +1,6 @@
 Name: docker-engine
 Version: 18.09.0
-Release: 340
+Release: 341
 Epoch: 2
 Summary: The open-source application container engine
 Group: Tools/Docker
@@ -227,6 +227,12 @@ fi
 %endif
 
 %changelog
+* Sat Aug 31 2024 zhongjiawei<zhongjiawei1@huawei.com> - 2:18.09.0-341
+- Type:bugfix
+- CVE:NA
+- SUG:NA
+- DESC:try to reconnect when containerd grpc return unexpected EOF
+
 * Fri Aug 02 2024 zhongjiawei<zhongjiawei1@huawei.com> - 2:18.09.0-340
 - Type:bugfix
 - CVE:NA

git-commit

@@ -1 +1 @@
-29173030b7e118013ed3e8a3773492c40928bb9c
+678fb4d2b2fbf91642358d82e5680aec01a15d56

patch/0279-docker-try-to-reconnect-when-containerd-grpc-return-.patch

@@ -0,0 +1,57 @@
+From 68ea83ecea0e38d084c0d15c9e99c0b4494b1f32 Mon Sep 17 00:00:00 2001
+From: zhongjiawei <zhongjiawei1@huawei.com>
+Date: Thu, 22 Aug 2024 20:22:43 +0800
+Subject: [PATCH] docker: try to reconnect when containerd grpc return
+ unexpected EOF
+
+---
+ .../engine/libcontainerd/client_daemon.go     | 26 ++++++++++++++-----
+ 1 file changed, 20 insertions(+), 6 deletions(-)
+
+diff --git a/components/engine/libcontainerd/client_daemon.go b/components/engine/libcontainerd/client_daemon.go
+index 09ce6e1f5..14f420ed8 100755
+--- a/components/engine/libcontainerd/client_daemon.go
++++ b/components/engine/libcontainerd/client_daemon.go
+@@ -38,9 +38,12 @@ import (
+ 	"google.golang.org/grpc/status"
+ )
+ 
+-// InitProcessName is the name given to the first process of a
+-// container
+-const InitProcessName = "init"
++const (
++	// InitProcessName is the name given to the first process of a container
++	InitProcessName = "init"
++	// RetryMax is the max num to connect containerd grpc
++	RetryMax = 10
++)
+ 
+ type container struct {
+ 	mu sync.Mutex
+@@ -167,9 +170,20 @@ func (c *client) Restore(ctx context.Context, id string, attachStdio StdioCallba
+ 		err = wrapError(err)
+ 	}()
+ 
+-	ctr, err := c.client.LoadContainer(ctx, id)
+-	if err != nil {
+-		return false, -1, errors.WithStack(wrapError(err))
++	var ctr containerd.Container
++	var err1 error
++	for retry := 1; retry <= RetryMax; retry++ {
++		ctr, err1 = c.client.LoadContainer(ctx, id)
++		if err1 == nil {
++			break
++		} else if strings.Contains(err1.Error(), "unexpected EOF") {
++			time.Sleep(time.Millisecond * 100)
++			continue
++		}
++		return false, -1, errors.WithStack(wrapError(err1))
++	}
++	if err1 != nil {
++		return false, -1, errors.Wrap(wrapError(err1), "reconnect load contianer failed")
+ 	}
+ 
+ 	attachIO := func(fifos *cio.FIFOSet) (cio.IO, error) {
+-- 
+2.33.0
+

series.conf

@@ -276,4 +276,5 @@ patch/0275-backport-fix-CVE-2024-32473.patch
 patch/0276-docker-Ignore-SIGURG-on-Linux.patch
 patch/0277-backport-fix-CVE-2024-41110.patch
 patch/0278-docker-add-clone3-seccomp-whitelist-for-arm64.patch
+patch/0279-docker-try-to-reconnect-when-containerd-grpc-return-.patch
 #end