GitList

Browse code

Fix critical bug: can't restart a restarting container

When user try to restart a restarting container, docker client report
error: "container is already active", and container will be stopped
instead be restarted which is seriously wrong.

What's more critical is that when user try to start this container
again, it will always fail.

This error can also be reproduced with a `docker stop`+`docker start`.

And this commit will fix the bug.

Signed-off-by: Zhang Wei <zhangwei555@huawei.com>

Zhang Wei authored on 2016/04/07 23:05:29
Showing 5 changed files

daemon/kill.go index e6ccfc7..3967f0f 100644
integration-cli/docker_cli_restart_test.go index 12fe1c4..31c8920 100644
libcontainerd/client_linux.go index 6bed5dc..3eae2e0 100644
libcontainerd/container_linux.go index a2a67af..019e5c4 100644
restartmanager/restartmanager.go index e534b2c..39b8b15 100644

daemon/kill.go

History View file @ a705e16

@@ -3,6 +3,7 @@ package daemon
                      import (
                      	"fmt"
                      	"runtime"
                     +	"strings"
                      	"syscall"
                      	"time"
@@ -81,7 +82,14 @@ func (daemon *Daemon) killWithSignal(container *container.Container, sig int) er
+                     	}
                      	if err := daemon.kill(container, sig); err != nil {
                     -		return fmt.Errorf("Cannot kill container %s: %s", container.ID, err)
                     +		err = fmt.Errorf("Cannot kill container %s: %s", container.ID, err)
                     +		// if container or process not exists, ignore the error
                     +		if strings.Contains(err.Error(), "container not found") ||
                     +			strings.Contains(err.Error(), "no such process") {
                     +			logrus.Warnf("%s", err.Error())
                     +		} else {
                     +			return err
                     +		}
+                     	}
                      	attributes := map[string]string{

integration-cli/docker_cli_restart_test.go

History View file @ a705e16

@@ -218,3 +218,32 @@ func (s *DockerSuite) TestRestartPolicyAfterRestart(c *check.C) {
                      	err = waitInspect(id, "{{.State.Status}}", "running", 30*time.Second)
                      	c.Assert(err, check.IsNil)
+                     }
+                    +
                     +func (s *DockerSuite) TestRestartContainerwithRestartPolicy(c *check.C) {
                     +	out1, _ := dockerCmd(c, "run", "-d", "--restart=on-failure:3", "busybox", "false")
                     +	out2, _ := dockerCmd(c, "run", "-d", "--restart=always", "busybox", "false")
+                    +
                     +	id1 := strings.TrimSpace(string(out1))
                     +	id2 := strings.TrimSpace(string(out2))
                     +	err := waitInspect(id1, "{{ .State.Restarting }} {{ .State.Running }}", "false false", 30*time.Second)
                     +	c.Assert(err, checker.IsNil)
+                    +
                     +	// TODO: fix racey problem during restart:
                     +	// https://jenkins.dockerproject.org/job/Docker-PRs-Win2Lin/24665/console
                     +	// Error response from daemon: Cannot restart container 6655f620d90b390527db23c0a15b3e46d86a58ecec20a5697ab228d860174251: remove /var/run/docker/libcontainerd/6655f620d90b390527db23c0a15b3e46d86a58ecec20a5697ab228d860174251/rootfs: device or resource busy
                     +	if _, _, err := dockerCmdWithError("restart", id1); err != nil {
                     +		// if restart met racey problem, try again
                     +		time.Sleep(500 * time.Millisecond)
                     +		dockerCmd(c, "restart", id1)
                     +	}
                     +	if _, _, err := dockerCmdWithError("restart", id2); err != nil {
                     +		// if restart met racey problem, try again
                     +		time.Sleep(500 * time.Millisecond)
                     +		dockerCmd(c, "restart", id2)
                     +	}
+                    +
                     +	dockerCmd(c, "stop", id1)
                     +	dockerCmd(c, "stop", id2)
                     +	dockerCmd(c, "start", id1)
                     +	dockerCmd(c, "start", id2)
                     +}

libcontainerd/client_linux.go

History View file @ a705e16

@@ -134,7 +134,7 @@ func (clnt *client) Create(containerID string, spec Spec, options ...CreateOptio
                      	defer clnt.unlock(containerID)
                      	if ctr, err := clnt.getContainer(containerID); err == nil {
                     -		if ctr.restarting { // docker doesn't actually call start if restart is on atm, but probably should in the future
                     +		if ctr.restarting {
                      			ctr.restartManager.Cancel()
                      			ctr.clean()
                      		} else {

libcontainerd/container_linux.go

History View file @ a705e16

@@ -124,6 +124,7 @@ func (ctr *container) handleEvent(e *containerd.Event) error {
                      			} else if restart {
                      				st.State = StateRestart
                      				ctr.restarting = true
                     +				ctr.client.deleteContainer(e.Id)
                      				go func() {
                      					err := <-wait
                      					ctr.restarting = false

restartmanager/restartmanager.go

History View file @ a705e16

@@ -51,7 +51,7 @@ func (rm *restartManager) ShouldRestart(exitCode uint32) (bool, chan error, erro
                      	}()
                      	if rm.canceled {
                     -		return false, nil, nil
                     +		return false, nil, fmt.Errorf("restartmanager canceled")
+                     	}
                      	if rm.active {