From 3e51312a61ea9d727d082818e487b76a7739385a Mon Sep 17 00:00:00 2001
From: botieking98 <botieking@gmail.com>
Date: Wed, 27 Oct 2021 14:58:31 +0800
Subject: [PATCH] fix shim reaper wait command execute blocked

wait no timeout will lead to event publish
process hang in some special scenarios.

Signed-off-by: botieking98 <botieking@gmail.com>
---
 cmd/containerd-shim/main_unix.go |  2 +-
 sys/reaper/reaper_unix.go        | 23 +++++++++++++++++++++++
 2 files changed, 24 insertions(+), 1 deletion(-)

diff --git a/cmd/containerd-shim/main_unix.go b/cmd/containerd-shim/main_unix.go
index 0ba452d35..e1d3eb13f 100644
--- a/cmd/containerd-shim/main_unix.go
+++ b/cmd/containerd-shim/main_unix.go
@@ -308,7 +308,7 @@ func (l *remoteEventsPublisher) Publish(ctx context.Context, topic string, event
 	if err != nil {
 		return err
 	}
-	status, err := reaper.Default.Wait(cmd, c)
+	status, err := reaper.Default.WaitTimeout(cmd, c, 30*time.Second)
 	if err != nil {
 		return errors.Wrapf(err, "failed to publish event: %s", b.String())
 	}
diff --git a/sys/reaper/reaper_unix.go b/sys/reaper/reaper_unix.go
index a735e1cbd..54cfe9eed 100644
--- a/sys/reaper/reaper_unix.go
+++ b/sys/reaper/reaper_unix.go
@@ -21,6 +21,7 @@ package reaper
 
 import (
 	"sync"
+	"syscall"
 	"time"
 
 	runc "github.com/containerd/go-runc"
@@ -116,6 +117,28 @@ func (m *Monitor) Wait(c *exec.Cmd, ec chan runc.Exit) (int, error) {
 	return -1, ErrNoSuchProcess
 }
 
+// WaitTimeout is used to skip the blocked command and kill the left process.
+func (m *Monitor) WaitTimeout(c *exec.Cmd, ec chan runc.Exit, timeout time.Duration) (int, error) {
+	sch := make(chan int)
+	ech := make(chan error)
+	go func() {
+		status, err := m.Wait(c, ec)
+		sch <- status
+		if err != nil {
+			ech <- err
+		}
+	}()
+	select {
+	case <-time.After(timeout):
+		syscall.Kill(c.Process.Pid, syscall.SIGKILL)
+		return 0, errors.Errorf("timeout %ds for cmd(pid=%d): %s, %s", timeout/time.Second, c.Process.Pid, c.Path, c.Args)
+	case status := <-sch:
+		return status, nil
+	case err := <-ech:
+		return -1, err
+	}
+}
+
 // Subscribe to process exit changes
 func (m *Monitor) Subscribe() chan runc.Exit {
 	c := make(chan runc.Exit, bufferSize)