Do not tolerate ENOENT while reconnecting socket

Einsfier · Einsfier · commit 467d71720ebd · 2021-06-24T15:48:42.000+08:00
Containerd may stuck for hours at startup while reconnecting shim sockets which are unexpected missing.
This fix adds a new ContextDialerFunc to allow you customize error handler during the dialing.

Signed-off-by: Ethan Chen &lt;pov@mahou-shoujo.moe&gt;
diff --git a/pkg/dialer/dialer.go b/pkg/dialer/dialer.go
@@ -45,6 +45,13 @@ func ContextDialer(ctx context.Context, address string) (net.Conn, error) {
 	return contextDialer(ctx, address, isNoent)
 }
 
+// ContextDialerFunc returns a GRPC net.Conn connected to the provided address,
+// but may tolerate certain dial errors and keep retrying until context canceled.
+// If tolerateErr is nil, it will tolerate no error and return what it got immediately while dialing.
+func ContextDialerFunc(ctx context.Context, address string, tolerateErr func(error) bool) (net.Conn, error) {
+	return contextDialer(ctx, address, tolerateErr)
+}
+
 // Dialer returns a GRPC net.Conn connected to the provided address
 // Deprecated: use ContextDialer and grpc.WithContextDialer.
 var Dialer = timeoutDialer
diff --git a/runtime/v2/shim/util_unix.go b/runtime/v2/shim/util_unix.go
@@ -81,7 +81,9 @@ func AnonDialer(address string, timeout time.Duration) (net.Conn, error) {
 
 // AnonReconnectDialer returns a dialer for an existing socket on reconnection
 func AnonReconnectDialer(address string, timeout time.Duration) (net.Conn, error) {
-	return AnonDialer(address, timeout)
+	ctx, cancel := context.WithTimeout(context.TODO(), timeout)
+	defer cancel()
+	return dialer.ContextDialerFunc(ctx, address, nil)
 }
 
 // NewSocket returns a new socket

Original file line number	Diff line number	Diff line change
`@@ -81,7 +81,9 @@ func AnonDialer(address string, timeout time.Duration) (net.Conn, error) {`
`81`	`81`
`82`	`82`	`// AnonReconnectDialer returns a dialer for an existing socket on reconnection`
`83`	`83`	`func AnonReconnectDialer(address string, timeout time.Duration) (net.Conn, error) {`
`84`		`- return AnonDialer(address, timeout)`
	`84`	`+ ctx, cancel := context.WithTimeout(context.TODO(), timeout)`
	`85`	`+ defer cancel()`
	`86`	`+ return dialer.ContextDialerFunc(ctx, address, nil)`
`85`	`87`	`}`
`86`	`88`
`87`	`89`	`// NewSocket returns a new socket`