From 3ac1232a2e3fbfc0465473e5d81cde41847c4252 Mon Sep 17 00:00:00 2001
From: jiangpengfei <jiangpengfei9@huawei.com>
Date: Wed, 19 Aug 2020 11:47:37 +0800
Subject: [PATCH 11/16] agent: fix agent reap agent process blocked problem

reason: add container waitProcess() timeout when
container process status is D/T.

Signed-off-by: jiangpengfei <jiangpengfei9@huawei.com>
---
 grpc.go | 43 +++++++++++++++++++++++++++++++++----------
 1 file changed, 33 insertions(+), 10 deletions(-)

diff --git a/grpc.go b/grpc.go
index de2cae7..3dd088e 100644
--- a/grpc.go
+++ b/grpc.go
@@ -49,6 +49,11 @@ const (
 	libcontainerPath = "/run/libcontainer"
 )
 
+// keep waitProcessTimeout value same as value in kata-runtime wait WaitProcessRequest response
+const (
+	waitProcessTimeOut = 10
+)
+
 var (
 	sysfsCPUOnlinePath          = "/sys/devices/system/cpu"
 	sysfsMemOnlinePath          = "/sys/devices/system/memory"
@@ -996,17 +1001,35 @@ func (a *agentGRPC) WaitProcess(ctx context.Context, req *pb.WaitProcessRequest)
 		ctr.deleteProcess(proc.id)
 	})
 
-	// Using helper function wait() to deal with the subreaper.
-	libContProcess := (*reaperLibcontainerProcess)(&(proc.process))
-	exitCode, err := a.sandbox.subreaper.wait(proc.exitCodeCh, libContProcess)
-	if err != nil {
-		return &pb.WaitProcessResponse{}, err
+	done := make(chan error)
+	var exitCode int = 0
+	go func() {
+		// Using helper function wait() to deal with the subreaper.
+		libContProcess := (*reaperLibcontainerProcess)(&(proc.process))
+		var err error
+		exitCode, err = a.sandbox.subreaper.wait(proc.exitCodeCh, libContProcess)
+		if err != nil {
+			done <- err
+			close(done)
+			return
+		}
+		// refill the exitCodeCh with the exitcode which can be read out
+		// by another WaitProcess(). Since this channel isn't be closed,
+		// here the refill will always success and it will be free by GC
+		// once the process exits.
+		proc.exitCodeCh <- exitCode
+
+		close(done)
+	}()
+
+	select {
+	case err := <-done:
+		if err != nil {
+			return &pb.WaitProcessResponse{}, err
+		}
+	case <-time.After(time.Duration(waitProcessTimeOut) * time.Second):
+		return &pb.WaitProcessResponse{}, grpcStatus.Errorf(codes.DeadlineExceeded, "agent wait reap container process timeout reached after %ds", waitProcessTimeOut)
 	}
-	//refill the exitCodeCh with the exitcode which can be read out
-	//by another WaitProcess(). Since this channel isn't be closed,
-	//here the refill will always success and it will be free by GC
-	//once the process exits.
-	proc.exitCodeCh <- exitCode
 
 	return &pb.WaitProcessResponse{
 		Status: int32(exitCode),
-- 
2.14.3 (Apple Git-98)