add a timeout in case a system-container refuses to quit

Signed-off-by: Sven Dowideit <SvenDowideit@home.org.au>

cmd/power/power.go

@@ -2,17 +2,20 @@ package power
 
 import (
 	"errors"
+	"fmt"
 	"os"
 	"path/filepath"
 	"strconv"
 	"strings"
 	"syscall"
+	"time"
 
 	"golang.org/x/net/context"
 
 	"github.com/docker/engine-api/types"
 	"github.com/docker/engine-api/types/container"
 	"github.com/docker/engine-api/types/filters"
+	"github.com/rancher/os/config"
 	"github.com/rancher/os/log"
 
 	"github.com/rancher/os/docker"
@@ -76,21 +79,35 @@ func runDocker(name string) error {
 		return err
 	}
 
-	go func() {
-		client.ContainerAttach(context.Background(), types.ContainerAttachOptions{
-			ContainerID: powerContainer.ID,
-			Stream:      true,
-			Stderr:      true,
-			Stdout:      true,
-		})
-	}()
-
 	err = client.ContainerStart(context.Background(), powerContainer.ID)
 	if err != nil {
 		return err
 	}
 
-	_, err = client.ContainerWait(context.Background(), powerContainer.ID)
+	reader, err := client.ContainerLogs(context.Background(), types.ContainerLogsOptions{
+		ContainerID: powerContainer.ID,
+		ShowStderr:  true,
+		ShowStdout:  true,
+		Follow:      true,
+	})
+	if err != nil {
+		log.Fatal(err)
+	}
+
+	for {
+		p := make([]byte, 4096)
+		n, err := reader.Read(p)
+		if err != nil {
+			log.Error(err)
+			if n == 0 {
+				reader.Close()
+				break
+			}
+		}
+		if n > 0 {
+			fmt.Print(string(p))
+		}
+	}
 
 	if err != nil {
 		log.Fatal(err)
@@ -126,13 +143,39 @@ func Halt() {
 }
 
 func reboot(code uint) {
+	cfg := config.LoadConfig()
+	timeoutValue := cfg.Rancher.ShutdownTimeout
+	if timeoutValue == 0 {
+		timeoutValue = 60
+	}
+	if timeoutValue < 5 {
+		timeoutValue = 5
+	}
+	log.Infof("Setting %s timeout to %d (rancher.shutdown_timeout set to %d)", os.Args[0], timeoutValue, cfg.Rancher.ShutdownTimeout)
+
+	go func() {
+		timeout := time.After(time.Duration(timeoutValue) * time.Second)
+		tick := time.Tick(100 * time.Millisecond)
+		// Keep trying until we're timed out or got a result or got an error
+		for {
+			select {
+			// Got a timeout! fail with a timeout error
+			case <-timeout:
+				log.Errorf("Container shutdown taking too long, forcing %s.", os.Args[0])
+				syscall.Sync()
+				syscall.Reboot(int(code))
+			case <-tick:
+				fmt.Printf(".")
+			}
+		}
+	}()
+
 	err := shutDownContainers()
 	if err != nil {
 		log.Error(err)
 	}
 
 	syscall.Sync()
-
 	err = syscall.Reboot(int(code))
 	if err != nil {
 		log.Fatal(err)
@@ -206,13 +249,20 @@ func shutDownContainers() error {
 		}
 	}
 
+	// lets see what containers are still running and only wait on those
+	containers, err = client.ContainerList(context.Background(), opts)
+	if err != nil {
+		return err
+	}
+
 	var waitErrorStrings []string
 
-	for _, container := range containers {
+	for idx, container := range containers {
 		if container.ID == currentContainerID {
 			continue
 		}
 		if container.Names[0] == "/console" {
+			consoleContainerIdx = idx
 			continue
 		}
 		log.Infof("Waiting %s : %s", container.Names[0], container.ID[:12])

cmd/power/shutdown.go

@@ -31,9 +31,6 @@ func Main() {
 	}
 	app.HideHelp = true
 
-	log.Infof("%s, %s", app.Usage, app.Version)
-	fmt.Printf("%s, %s", app.Usage, app.Version)
-
 	app.Run(os.Args)
 }
 

config/schema.go

@@ -52,7 +52,8 @@ var schema = `{
         "defaults": {"$ref": "#/definitions/defaults_config"},
         "resize_device": {"type": "string"},
         "sysctl": {"type": "object"},
-        "restart_services": {"type": "array"}
+        "restart_services": {"type": "array"},
+        "shutdown_timeout": {"type": "integer"}
       }
     },
 

config/types.go

@@ -130,6 +130,7 @@ type RancherConfig struct {
 	ResizeDevice        string                                    `yaml:"resize_device,omitempty"`
 	Sysctl              map[string]string                         `yaml:"sysctl,omitempty"`
 	RestartServices     []string                                  `yaml:"restart_services,omitempty"`
+	ShutdownTimeout     int                                       `yaml:"shutdown_timeout,omitempty"`
 }
 
 type UpgradeConfig struct {

os-config.tpl.yml

@@ -1,4 +1,5 @@
 rancher:
+  shutdown_timeout: 60
   environment:
     VERSION: {{.VERSION}}
     SUFFIX: {{.SUFFIX}}

scripts/installer/kexec/Dockerfile.dapper

@@ -10,7 +10,7 @@ RUN echo "Acquire::http { Proxy \"$APTPROXY\"; };" >> /etc/apt/apt.conf.d/01prox
     && apt-get install -yq build-essential autoconf libtool gawk alien fakeroot \
         zlib1g-dev uuid-dev libattr1-dev libblkid-dev libselinux-dev libudev-dev libdevmapper-dev \
         module-init-tools \
-        parted lsscsi ksh curl git
+        parted lsscsi ksh curl git wget
 
 WORKDIR /source
 
@@ -20,8 +20,9 @@ WORKDIR /source
 #    && tar zxvf /source/build-linux-4.9.15-rancher-x86.tar.gz
 
 # https://www.kernel.org/pub/linux/utils/kernel/kexec/
-ENV VERSION 2.0.14
-ADD https://www.kernel.org/pub/linux/utils/kernel/kexec/kexec-tools-$VERSION.tar.gz .
+ENV VERSION 2.0.15
+RUN wget https://www.kernel.org/pub/linux/utils/kernel/kexec/kexec-tools-$VERSION.tar.gz \
+    && tar zxvf kexec-tools-$VERSION.tar.gz
 
 RUN zcat kexec-tools-$VERSION.tar.gz | tar xvf - \
   && cd kexec-tools-$VERSION \

scripts/schema.json

@@ -50,7 +50,8 @@
         "defaults": {"$ref": "#/definitions/defaults_config"},
         "resize_device": {"type": "string"},
         "sysctl": {"type": "object"},
-        "restart_services": {"type": "array"}
+        "restart_services": {"type": "array"},
+	"shutdown_timeout": {"type": "integer"}
       }
     },
 

tests/cmdline_test.go

@@ -19,7 +19,7 @@ func (s *QemuSuite) TestElideCmdLine(c *C) {
 	s.RunQemuWith(c, runArgs...)
 
 	s.CheckOutput(c, "nope\n", Equals, "hostname")
-	cmdline := s.CheckOutput(c, "", Not(Equals), "cat /proc/cmdline",)
+	cmdline := s.CheckOutput(c, "", Not(Equals), "cat /proc/cmdline")
 	if strings.Contains(cmdline, extra) {
 		c.Errorf("/proc/cmdline (%s) contains info that should be elided (%s)", cmdline, extra)
 	}