builder/amazonebs: Timeouts while waiting for SSH to connect

builder/amazonebs/builder.go

@@ -14,6 +14,7 @@ import (
 	"github.com/mitchellh/multistep"
 	"github.com/mitchellh/packer/packer"
 	"log"
+	"time"
 )
 
 // The unique ID for this builder
@@ -30,9 +31,12 @@ type config struct {
 	InstanceType string `mapstructure:"instance_type"`
 	SSHUsername  string `mapstructure:"ssh_username"`
 	SSHPort      int    `mapstructure:"ssh_port"`
+	SSHTimeout   time.Duration
 
 	// Configuration of the resulting AMI
 	AMIName string `mapstructure:"ami_name"`
+
+	RawSSHTimeout string `mapstructure:"ssh_timeout"`
 }
 
 type Builder struct {
@@ -40,16 +44,20 @@ type Builder struct {
 	runner multistep.Runner
 }
 
-func (b *Builder) Prepare(raw interface{}) (err error) {
-	err = mapstructure.Decode(raw, &b.config)
+func (b *Builder) Prepare(raw interface{}) error {
+	err := mapstructure.Decode(raw, &b.config)
 	if err != nil {
-		return
+		return err
 	}
 
 	if b.config.SSHPort == 0 {
 		b.config.SSHPort = 22
 	}
 
+	if b.config.RawSSHTimeout == "" {
+		b.config.RawSSHTimeout = "1m"
+	}
+
 	// Accumulate any errors
 	errs := make([]error, 0)
 
@@ -79,12 +87,17 @@ func (b *Builder) Prepare(raw interface{}) (err error) {
 		errs = append(errs, errors.New("An ssh_username must be specified"))
 	}
 
+	b.config.SSHTimeout, err = time.ParseDuration(b.config.RawSSHTimeout)
+	if err != nil {
+		errs = append(errs, fmt.Errorf("Failed parsing ssh_timeout: %s", err))
+	}
+
 	if len(errs) > 0 {
 		return &packer.MultiError{errs}
 	}
 
 	log.Printf("Config: %+v", b.config)
-	return
+	return nil
 }
 
 func (b *Builder) Run(ui packer.Ui, hook packer.Hook, cache packer.Cache) packer.Artifact {

builder/amazonebs/builder_test.go

@@ -191,6 +191,25 @@ func TestBuilderPrepare_SSHPort(t *testing.T) {
 	}
 }
 
+func TestBuilderPrepare_SSHTimeout(t *testing.T) {
+	var b Builder
+	config := testConfig()
+
+	// Test with a bad value
+	config["ssh_timeout"] = "this is not good"
+	err := b.Prepare(config)
+	if err == nil {
+		t.Fatal("should have error")
+	}
+
+	// Test with a good one
+	config["ssh_timeout"] = "5s"
+	err = b.Prepare(config)
+	if err != nil {
+		t.Fatalf("should not have error: %s", err)
+	}
+}
+
 func TestBuilderPrepare_SSHUsername(t *testing.T) {
 	var b Builder
 	config := testConfig()

builder/amazonebs/step_connect_ssh.go

@@ -39,17 +39,50 @@ func (s *stepConnectSSH) Run(state map[string]interface{}) multistep.StepAction
 		},
 	}
 
-	// Try to connect for SSH a few times
-	ui.Say("Connecting to the instance via SSH...")
-	for i := 0; i < 5; i++ {
-		time.Sleep(time.Duration(i) * time.Second)
-
-		log.Printf(
-			"Opening TCP conn for SSH to %s:%d (attempt %d)",
-			instance.DNSName, config.SSHPort, i+1)
-		s.conn, err = net.Dial("tcp", fmt.Sprintf("%s:%d", instance.DNSName, config.SSHPort))
-		if err != nil {
-			continue
+	// Start trying to connect to SSH
+	connected := make(chan bool, 1)
+	connectQuit := make(chan bool, 1)
+	defer func() {
+		connectQuit <- true
+	}()
+
+	go func() {
+		var err error
+
+		ui.Say("Connecting to the instance via SSH...")
+		attempts := 0
+		for {
+			select {
+			case <-connectQuit:
+				return
+			default:
+			}
+
+			attempts += 1
+			log.Printf(
+				"Opening TCP conn for SSH to %s:%d (attempt %d)",
+				instance.DNSName, config.SSHPort, attempts)
+			s.conn, err = net.Dial("tcp", fmt.Sprintf("%s:%d", instance.DNSName, config.SSHPort))
+			if err == nil {
+				break
+			}
+		}
+
+		connected <- true
+	}()
+
+	log.Printf("Waiting up to %s for SSH connection", config.SSHTimeout)
+	timeout := time.After(config.SSHTimeout)
+
+	ConnectWaitLoop:
+	for {
+		select {
+		case <-connected:
+			// We connected. Just break the loop.
+			break ConnectWaitLoop
+		case <-timeout:
+			ui.Error("Timeout while waiting to connect to SSH.")
+			return multistep.ActionHalt
 		}
 	}