fix(lifecycle): cleanup lifecycle

- removes unwieldy SkipUpdate return value in favor of errors.Is
- generalizes the code for all four phases
- allows timeout to be defined for all phases
- enables explicit unit in timeout label values (in addition to implicit minutes)

internal/actions/mocks/client.go

@@ -5,6 +5,7 @@ import (
 	"fmt"
 	"time"
 
+	c "github.com/containrrr/watchtower/pkg/container"
 	t "github.com/containrrr/watchtower/pkg/types"
 )
 
@@ -72,16 +73,16 @@ func (client MockClient) GetContainer(_ t.ContainerID) (t.Container, error) {
 }
 
 // ExecuteCommand is a mock method
-func (client MockClient) ExecuteCommand(_ t.ContainerID, command string, _ int) (SkipUpdate bool, err error) {
+func (client MockClient) ExecuteCommand(_ t.ContainerID, command string, _ time.Duration) error {
 	switch command {
 	case "/PreUpdateReturn0.sh":
-		return false, nil
+		return nil
 	case "/PreUpdateReturn1.sh":
-		return false, fmt.Errorf("command exited with code 1")
+		return fmt.Errorf("command exited with code 1")
 	case "/PreUpdateReturn75.sh":
-		return true, nil
+		return c.ErrorLifecycleSkip
 	default:
-		return false, nil
+		return nil
 	}
 }
 

internal/util/time.go

@@ -0,0 +1,15 @@
+package util
+
+import (
+	"strconv"
+	"time"
+)
+
+// ParseDuration parses the input string as a duration, treating a plain number as implicitly using the specified unit
+func ParseDuration(input string, unitlessUnit time.Duration) (time.Duration, error) {
+	if unitless, err := strconv.Atoi(input); err == nil {
+		return unitlessUnit * time.Duration(unitless), nil
+	}
+
+	return time.ParseDuration(input)
+}

pkg/container/client.go

@@ -31,7 +31,7 @@ type Client interface {
 	StartContainer(t.Container) (t.ContainerID, error)
 	RenameContainer(t.Container, string) error
 	IsContainerStale(t.Container, t.UpdateParams) (stale bool, latestImage t.ImageID, err error)
-	ExecuteCommand(containerID t.ContainerID, command string, timeout int) (SkipUpdate bool, err error)
+	ExecuteCommand(containerID t.ContainerID, command string, timeout time.Duration) error
 	RemoveImageByID(t.ImageID) error
 	WarnOnHeadPullFailed(container t.Container) bool
 }
@@ -439,7 +439,7 @@ func (client dockerClient) RemoveImageByID(id t.ImageID) error {
 	return err
 }
 
-func (client dockerClient) ExecuteCommand(containerID t.ContainerID, command string, timeout int) (SkipUpdate bool, err error) {
+func (client dockerClient) ExecuteCommand(containerID t.ContainerID, command string, timeout time.Duration) error {
 	bg := context.Background()
 	clog := log.WithField("containerID", containerID)
 
@@ -452,7 +452,7 @@ func (client dockerClient) ExecuteCommand(containerID t.ContainerID, command str
 
 	exec, err := client.api.ContainerExecCreate(bg, string(containerID), execConfig)
 	if err != nil {
-		return false, err
+		return err
 	}
 
 	response, attachErr := client.api.ContainerExecAttach(bg, exec.ID, types.ExecStartCheck{
@@ -467,7 +467,7 @@ func (client dockerClient) ExecuteCommand(containerID t.ContainerID, command str
 	execStartCheck := types.ExecStartCheck{Detach: false, Tty: true}
 	err = client.api.ContainerExecStart(bg, exec.ID, execStartCheck)
 	if err != nil {
-		return false, err
+		return err
 	}
 
 	var output string
@@ -484,24 +484,16 @@ func (client dockerClient) ExecuteCommand(containerID t.ContainerID, command str
 
 	// Inspect the exec to get the exit code and print a message if the
 	// exit code is not success.
-	skipUpdate, err := client.waitForExecOrTimeout(bg, exec.ID, output, timeout)
+	return client.waitForExecOrTimeout(bg, exec.ID, output, timeout)
-	if err != nil {
-		return true, err
-	}
-
-	return skipUpdate, nil
 }
 
-func (client dockerClient) waitForExecOrTimeout(bg context.Context, ID string, execOutput string, timeout int) (SkipUpdate bool, err error) {
+func (client dockerClient) waitForExecOrTimeout(ctx context.Context, ID string, execOutput string, timeout time.Duration) error {
 	const ExTempFail = 75
-	var ctx context.Context
-	var cancel context.CancelFunc
 
 	if timeout > 0 {
-		ctx, cancel = context.WithTimeout(bg, time.Duration(timeout)*time.Minute)
+		var cancel context.CancelFunc
+		ctx, cancel = context.WithTimeout(ctx, timeout)
 		defer cancel()
-	} else {
-		ctx = bg
 	}
 
 	for {
@@ -516,7 +508,7 @@ func (client dockerClient) waitForExecOrTimeout(bg context.Context, ID string, e
 		}).Debug("Awaiting timeout or completion")
 
 		if err != nil {
-			return false, err
+			return err
 		}
 		if execInspect.Running {
 			time.Sleep(1 * time.Second)
@@ -527,15 +519,15 @@ func (client dockerClient) waitForExecOrTimeout(bg context.Context, ID string, e
 		}
 
 		if execInspect.ExitCode == ExTempFail {
-			return true, nil
+			return ErrorLifecycleSkip
 		}
 
 		if execInspect.ExitCode > 0 {
-			return false, fmt.Errorf("command exited with code %v  %s", execInspect.ExitCode, execOutput)
+			return fmt.Errorf("command exited with code %v", execInspect.ExitCode)
 		}
 		break
 	}
-	return false, nil
+	return nil
 }
 
 func (client dockerClient) waitForStopOrTimeout(c t.Container, waitTime time.Duration) error {

pkg/container/client_test.go

@@ -308,7 +308,7 @@ var _ = Describe("the client", func() {
 					),
 				)
 
-				_, err := client.ExecuteCommand(containerID, cmd, 1)
+				err := client.ExecuteCommand(containerID, cmd, 1)
 				Expect(err).NotTo(HaveOccurred())
 				// Note: Since Execute requires opening up a raw TCP stream to the daemon for the output, this will fail
 				// when using the mock API server. Regardless of the outcome, the log should include the container ID

pkg/container/container.go

@@ -219,44 +219,6 @@ func (c Container) IsWatchtower() bool {
 	return ContainsWatchtowerLabel(c.containerInfo.Config.Labels)
 }
 
-// PreUpdateTimeout checks whether a container has a specific timeout set
-// for how long the pre-update command is allowed to run. This value is expressed
-// either as an integer, in minutes, or as 0 which will allow the command/script
-// to run indefinitely. Users should be cautious with the 0 option, as that
-// could result in watchtower waiting forever.
-func (c Container) PreUpdateTimeout() int {
-	var minutes int
-	var err error
-
-	val := c.getLabelValueOrEmpty(preUpdateTimeoutLabel)
-
-	minutes, err = strconv.Atoi(val)
-	if err != nil || val == "" {
-		return 1
-	}
-
-	return minutes
-}
-
-// PostUpdateTimeout checks whether a container has a specific timeout set
-// for how long the post-update command is allowed to run. This value is expressed
-// either as an integer, in minutes, or as 0 which will allow the command/script
-// to run indefinitely. Users should be cautious with the 0 option, as that
-// could result in watchtower waiting forever.
-func (c Container) PostUpdateTimeout() int {
-	var minutes int
-	var err error
-
-	val := c.getLabelValueOrEmpty(postUpdateTimeoutLabel)
-
-	minutes, err = strconv.Atoi(val)
-	if err != nil || val == "" {
-		return 1
-	}
-
-	return minutes
-}
-
 // StopSignal returns the custom stop signal (if any) that is encoded in the
 // container's metadata. If the container has not specified a custom stop
 // signal, the empty string "" is returned.

pkg/container/errors.go

@@ -6,3 +6,6 @@ var errorNoImageInfo = errors.New("no available image info")
 var errorNoContainerInfo = errors.New("no available container info")
 var errorInvalidConfig = errors.New("container configuration missing or invalid")
 var errorLabelNotFound = errors.New("label was not found in container")
+
+// ErrorLifecycleSkip is returned by a lifecycle hook when the exit code of the command indicated that it ought to be skipped
+var ErrorLifecycleSkip = errors.New("skipping container as the pre-update command returned exit code 75 (EX_TEMPFAIL)")

pkg/container/metadata.go

@@ -1,67 +1,92 @@
 package container
 
-import "strconv"
+import (
+	"errors"
+	"fmt"
+	"strconv"
+	"time"
+
+	"github.com/containrrr/watchtower/internal/util"
+	wt "github.com/containrrr/watchtower/pkg/types"
+
+	"github.com/sirupsen/logrus"
+)
 
 const (
-	watchtowerLabel        = "com.centurylinklabs.watchtower"
+	namespace        = "com.centurylinklabs.watchtower"
-	signalLabel            = "com.centurylinklabs.watchtower.stop-signal"
+	watchtowerLabel  = namespace
-	enableLabel            = "com.centurylinklabs.watchtower.enable"
+	signalLabel      = namespace + ".stop-signal"
-	monitorOnlyLabel       = "com.centurylinklabs.watchtower.monitor-only"
+	enableLabel      = namespace + ".enable"
-	noPullLabel            = "com.centurylinklabs.watchtower.no-pull"
+	monitorOnlyLabel = namespace + ".monitor-only"
-	dependsOnLabel         = "com.centurylinklabs.watchtower.depends-on"
+	noPullLabel      = namespace + ".no-pull"
-	zodiacLabel            = "com.centurylinklabs.zodiac.original-image"
+	dependsOnLabel   = namespace + ".depends-on"
-	scope                  = "com.centurylinklabs.watchtower.scope"
+	zodiacLabel      = "com.centurylinklabs.zodiac.original-image"
-	preCheckLabel          = "com.centurylinklabs.watchtower.lifecycle.pre-check"
+	scope            = namespace + ".scope"
-	postCheckLabel         = "com.centurylinklabs.watchtower.lifecycle.post-check"
-	preUpdateLabel         = "com.centurylinklabs.watchtower.lifecycle.pre-update"
-	postUpdateLabel        = "com.centurylinklabs.watchtower.lifecycle.post-update"
-	preUpdateTimeoutLabel  = "com.centurylinklabs.watchtower.lifecycle.pre-update-timeout"
-	postUpdateTimeoutLabel = "com.centurylinklabs.watchtower.lifecycle.post-update-timeout"
 )
 
-// GetLifecyclePreCheckCommand returns the pre-check command set in the container metadata or an empty string
+// ContainsWatchtowerLabel takes a map of labels and values and tells
-func (c Container) GetLifecyclePreCheckCommand() string {
+// the consumer whether it contains a valid watchtower instance label
-	return c.getLabelValueOrEmpty(preCheckLabel)
+func ContainsWatchtowerLabel(labels map[string]string) bool {
+	val, ok := labels[watchtowerLabel]
+	return ok && val == "true"
 }
 
-// GetLifecyclePostCheckCommand returns the post-check command set in the container metadata or an empty string
+// GetLifecycleCommand returns the lifecycle command set in the container metadata or an empty string
-func (c Container) GetLifecyclePostCheckCommand() string {
+func (c *Container) GetLifecycleCommand(phase wt.LifecyclePhase) string {
-	return c.getLabelValueOrEmpty(postCheckLabel)
+	label := fmt.Sprintf("%v.lifecycle.%v", namespace, phase)
-}
+	value, found := c.getLabelValue(label)
 
-// GetLifecyclePreUpdateCommand returns the pre-update command set in the container metadata or an empty string
+	if !found {
-func (c Container) GetLifecyclePreUpdateCommand() string {
+		return ""
-	return c.getLabelValueOrEmpty(preUpdateLabel)
+	}
-}
 
-// GetLifecyclePostUpdateCommand returns the post-update command set in the container metadata or an empty string
+	return value
-func (c Container) GetLifecyclePostUpdateCommand() string {
-	return c.getLabelValueOrEmpty(postUpdateLabel)
 }
 
-// ContainsWatchtowerLabel takes a map of labels and values and tells
+// GetLifecycleTimeout checks whether a container has a specific timeout set
-// the consumer whether it contains a valid watchtower instance label
+// for how long the lifecycle command is allowed to run. This value is expressed
-func ContainsWatchtowerLabel(labels map[string]string) bool {
+// either as a duration, an integer (minutes implied), or as 0 which will allow the command/script
-	val, ok := labels[watchtowerLabel]
+// to run indefinitely. Users should be cautious with the 0 option, as that
-	return ok && val == "true"
+// could result in watchtower waiting forever.
+func (c *Container) GetLifecycleTimeout(phase wt.LifecyclePhase) time.Duration {
+	label := fmt.Sprintf("%v.lifecycle.%v-timeout", namespace, phase)
+	timeout, err := c.getDurationLabelValue(label, time.Minute)
+
+	if err != nil {
+		timeout = time.Minute
+		if !errors.Is(err, errorLabelNotFound) {
+			logrus.WithError(err).Errorf("could not parse timeout label value for %v lifecycle", phase)
+		}
+	}
+
+	return timeout
 }
 
-func (c Container) getLabelValueOrEmpty(label string) string {
+func (c *Container) getLabelValueOrEmpty(label string) string {
 	if val, ok := c.containerInfo.Config.Labels[label]; ok {
 		return val
 	}
 	return ""
 }
 
-func (c Container) getLabelValue(label string) (string, bool) {
+func (c *Container) getLabelValue(label string) (string, bool) {
 	val, ok := c.containerInfo.Config.Labels[label]
 	return val, ok
 }
 
-func (c Container) getBoolLabelValue(label string) (bool, error) {
+func (c *Container) getBoolLabelValue(label string) (bool, error) {
 	if strVal, ok := c.containerInfo.Config.Labels[label]; ok {
 		value, err := strconv.ParseBool(strVal)
 		return value, err
 	}
 	return false, errorLabelNotFound
 }
+
+func (c *Container) getDurationLabelValue(label string, unitlessUnit time.Duration) (time.Duration, error) {
+	value, found := c.getLabelValue(label)
+	if !found || len(value) < 1 {
+		return 0, errorLabelNotFound
+	}
+
+	return util.ParseDuration(value, unitlessUnit)
+}

pkg/lifecycle/lifecycle.go

@@ -6,101 +6,60 @@ import (
 	log "github.com/sirupsen/logrus"
 )
 
-// ExecutePreChecks tries to run the pre-check lifecycle hook for all containers included by the current filter.
+type ExecCommandFunc func(client container.Client, container types.Container)
-func ExecutePreChecks(client container.Client, params types.UpdateParams) {
-	containers, err := client.ListContainers(params.Filter)
-	if err != nil {
-		return
-	}
-	for _, currentContainer := range containers {
-		ExecutePreCheckCommand(client, currentContainer)
-	}
-}
-
-// ExecutePostChecks tries to run the post-check lifecycle hook for all containers included by the current filter.
-func ExecutePostChecks(client container.Client, params types.UpdateParams) {
-	containers, err := client.ListContainers(params.Filter)
-	if err != nil {
-		return
-	}
-	for _, currentContainer := range containers {
-		ExecutePostCheckCommand(client, currentContainer)
-	}
-}
 
 // ExecutePreCheckCommand tries to run the pre-check lifecycle hook for a single container.
 func ExecutePreCheckCommand(client container.Client, container types.Container) {
-	clog := log.WithField("container", container.Name())
+	err := ExecuteLifeCyclePhaseCommand(types.PreCheck, client, container)
-	command := container.GetLifecyclePreCheckCommand()
-	if len(command) == 0 {
-		clog.Debug("No pre-check command supplied. Skipping")
-		return
-	}
-
-	clog.Debug("Executing pre-check command.")
-	_, err := client.ExecuteCommand(container.ID(), command, 1)
 	if err != nil {
-		clog.Error(err)
+		log.WithField("container", container.Name()).Error(err)
 	}
 }
 
 // ExecutePostCheckCommand tries to run the post-check lifecycle hook for a single container.
 func ExecutePostCheckCommand(client container.Client, container types.Container) {
-	clog := log.WithField("container", container.Name())
+	err := ExecuteLifeCyclePhaseCommand(types.PostCheck, client, container)
-	command := container.GetLifecyclePostCheckCommand()
-	if len(command) == 0 {
-		clog.Debug("No post-check command supplied. Skipping")
-		return
-	}
-
-	clog.Debug("Executing post-check command.")
-	_, err := client.ExecuteCommand(container.ID(), command, 1)
 	if err != nil {
-		clog.Error(err)
+		log.WithField("container", container.Name()).Error(err)
 	}
 }
 
 // ExecutePreUpdateCommand tries to run the pre-update lifecycle hook for a single container.
-func ExecutePreUpdateCommand(client container.Client, container types.Container) (SkipUpdate bool, err error) {
+func ExecutePreUpdateCommand(client container.Client, container types.Container) error {
-	timeout := container.PreUpdateTimeout()
+	return ExecuteLifeCyclePhaseCommand(types.PreUpdate, client, container)
-	command := container.GetLifecyclePreUpdateCommand()
-	clog := log.WithField("container", container.Name())
-
-	if len(command) == 0 {
-		clog.Debug("No pre-update command supplied. Skipping")
-		return false, nil
-	}
-
-	if !container.IsRunning() || container.IsRestarting() {
-		clog.Debug("Container is not running. Skipping pre-update command.")
-		return false, nil
-	}
-
-	clog.Debug("Executing pre-update command.")
-	return client.ExecuteCommand(container.ID(), command, timeout)
 }
 
 // ExecutePostUpdateCommand tries to run the post-update lifecycle hook for a single container.
 func ExecutePostUpdateCommand(client container.Client, newContainerID types.ContainerID) {
 	newContainer, err := client.GetContainer(newContainerID)
-	timeout := newContainer.PostUpdateTimeout()
-
 	if err != nil {
 		log.WithField("containerID", newContainerID.ShortID()).Error(err)
 		return
 	}
-	clog := log.WithField("container", newContainer.Name())
 
-	command := newContainer.GetLifecyclePostUpdateCommand()
+	err = ExecuteLifeCyclePhaseCommand(types.PostUpdate, client, newContainer)
-	if len(command) == 0 {
+	if err != nil {
-		clog.Debug("No post-update command supplied. Skipping")
+		log.WithField("container", newContainer.Name()).Error(err)
-		return
 	}
+}
 
-	clog.Debug("Executing post-update command.")
+// ExecuteLifeCyclePhaseCommand tries to run the corresponding lifecycle hook for a single container.
-	_, err = client.ExecuteCommand(newContainerID, command, timeout)
+func ExecuteLifeCyclePhaseCommand(phase types.LifecyclePhase, client container.Client, container types.Container) error {
 
-	if err != nil {
+	timeout := container.GetLifecycleTimeout(phase)
-		clog.Error(err)
+	command := container.GetLifecycleCommand(phase)
+	clog := log.WithField("container", container.Name())
+
+	if len(command) == 0 {
+		clog.Debugf("No %v command supplied. Skipping", phase)
+		return nil
 	}
+
+	if !container.IsRunning() || container.IsRestarting() {
+		clog.Debugf("Container is not running. Skipping %v command.", phase)
+		return nil
+	}
+
+	clog.Debugf("Executing %v command.", phase)
+	return client.ExecuteCommand(container.ID(), command, timeout)
 }

pkg/types/container.go

@@ -2,6 +2,7 @@ package types
 
 import (
 	"strings"
+	"time"
 
 	"github.com/docker/docker/api/types"
 	dc "github.com/docker/docker/api/types/container"
@@ -60,18 +61,14 @@ type Container interface {
 	StopSignal() string
 	HasImageInfo() bool
 	ImageInfo() *types.ImageInspect
-	GetLifecyclePreCheckCommand() string
+	GetLifecycleCommand(LifecyclePhase) string
-	GetLifecyclePostCheckCommand() string
+	GetLifecycleTimeout(LifecyclePhase) time.Duration
-	GetLifecyclePreUpdateCommand() string
-	GetLifecyclePostUpdateCommand() string
 	VerifyConfiguration() error
 	SetStale(bool)
 	IsStale() bool
 	IsNoPull(UpdateParams) bool
 	SetLinkedToRestarting(bool)
 	IsLinkedToRestarting() bool
-	PreUpdateTimeout() int
-	PostUpdateTimeout() int
 	IsRestarting() bool
 	GetCreateConfig() *dc.Config
 	GetCreateHostConfig() *dc.HostConfig

pkg/types/lifecycle.go

@@ -0,0 +1,27 @@
+package types
+
+import "fmt"
+
+type LifecyclePhase int
+
+const (
+	PreCheck LifecyclePhase = iota
+	PreUpdate
+	PostUpdate
+	PostCheck
+)
+
+func (p LifecyclePhase) String() string {
+	switch p {
+	case PreCheck:
+		return "pre-check"
+	case PreUpdate:
+		return "pre-update"
+	case PostUpdate:
+		return "post-update"
+	case PostCheck:
+		return "post-check"
+	default:
+		return fmt.Sprintf("invalid(%d)", p)
+	}
+}