Add failure action for rolling updates

This changes the default behavior so that rolling updates will not
proceed once an updated task fails to start, or stops running during the
update. Users can use docker service inspect --pretty servicename to see
the update status, and if it pauses due to a failure, it will explain
that the update is paused, and show the task ID that caused it to pause.
It also shows the time since the update started.

A new --update-on-failure=(pause|continue) flag selects the
behavior. Pause means the update stops once a task fails, continue means
the old behavior of continuing the update anyway.

In the future this will be extended with additional behaviors like
automatic rollback, and flags controlling parameters like how many tasks
need to fail for the update to stop proceeding. This is a minimal
solution for 1.12.

Signed-off-by: Aaron Lehmann <aaron.lehmann@docker.com>
(cherry picked from commit 57ae29aa74e77ade3c91b1c77ba766512dae9ab4)
Signed-off-by: Tibor Vass <tibor@docker.com>

api/client/service/inspect.go

@@ -4,6 +4,7 @@ import (
 	"fmt"
 	"io"
 	"strings"
+	"time"
 
 	"golang.org/x/net/context"
 
@@ -101,6 +102,17 @@ func printService(out io.Writer, service swarm.Service) {
 			fmt.Fprintf(out, " Replicas:\t%d\n", *service.Spec.Mode.Replicated.Replicas)
 		}
 	}
+
+	if service.UpdateStatus.State != "" {
+		fmt.Fprintln(out, "Update status:")
+		fmt.Fprintf(out, " State:\t\t%s\n", service.UpdateStatus.State)
+		fmt.Fprintf(out, " Started:\t%s ago\n", strings.ToLower(units.HumanDuration(time.Since(service.UpdateStatus.StartedAt))))
+		if service.UpdateStatus.State == swarm.UpdateStateCompleted {
+			fmt.Fprintf(out, " Completed:\t%s ago\n", strings.ToLower(units.HumanDuration(time.Since(service.UpdateStatus.CompletedAt))))
+		}
+		fmt.Fprintf(out, " Message:\t%s\n", service.UpdateStatus.Message)
+	}
+
 	fmt.Fprintln(out, "Placement:")
 	if service.Spec.TaskTemplate.Placement != nil && len(service.Spec.TaskTemplate.Placement.Constraints) > 0 {
 		ioutils.FprintfIfNotEmpty(out, " Constraints\t: %s\n", strings.Join(service.Spec.TaskTemplate.Placement.Constraints, ", "))
@@ -110,6 +122,7 @@ func printService(out io.Writer, service swarm.Service) {
 	if service.Spec.UpdateConfig.Delay.Nanoseconds() > 0 {
 		fmt.Fprintf(out, " Delay:\t\t%s\n", service.Spec.UpdateConfig.Delay)
 	}
+	fmt.Fprintf(out, " On failure:\t%s\n", service.Spec.UpdateConfig.FailureAction)
 	fmt.Fprintf(out, "ContainerSpec:\n")
 	printContainerSpec(out, service.Spec.TaskTemplate.ContainerSpec)
 

api/client/service/opts.go

@@ -274,6 +274,7 @@ func (m *MountOpt) Value() []swarm.Mount {
 type updateOptions struct {
 	parallelism uint64
 	delay       time.Duration
+	onFailure   string
 }
 
 type resourceOptions struct {
@@ -457,6 +458,7 @@ func (opts *serviceOptions) ToService() (swarm.ServiceSpec, error) {
 		UpdateConfig: &swarm.UpdateConfig{
 			Parallelism:   opts.update.parallelism,
 			Delay:         opts.update.delay,
+			FailureAction: opts.update.onFailure,
 		},
 		Networks:     convertNetworks(opts.networks),
 		EndpointSpec: opts.endpoint.ToEndpointSpec(),
@@ -503,6 +505,7 @@ func addServiceFlags(cmd *cobra.Command, opts *serviceOptions) {
 
 	flags.Uint64Var(&opts.update.parallelism, flagUpdateParallelism, 1, "Maximum number of tasks updated simultaneously (0 to update all at once)")
 	flags.DurationVar(&opts.update.delay, flagUpdateDelay, time.Duration(0), "Delay between updates")
+	flags.StringVar(&opts.update.onFailure, flagUpdateFailureAction, "pause", "Action on update failure (pause|continue)")
 
 	flags.StringVar(&opts.endpoint.mode, flagEndpointMode, "", "Endpoint mode (vip or dnsrr)")
 
@@ -545,6 +548,7 @@ const (
 	flagRestartWindow       = "restart-window"
 	flagStopGracePeriod     = "stop-grace-period"
 	flagUpdateDelay         = "update-delay"
+	flagUpdateFailureAction = "update-failure-action"
 	flagUpdateParallelism   = "update-parallelism"
 	flagUser                = "user"
 	flagRegistryAuth        = "with-registry-auth"

api/client/service/update.go

@@ -191,12 +191,13 @@ func updateService(flags *pflag.FlagSet, spec *swarm.ServiceSpec) error {
 		return err
 	}
 
-	if anyChanged(flags, flagUpdateParallelism, flagUpdateDelay) {
+	if anyChanged(flags, flagUpdateParallelism, flagUpdateDelay, flagUpdateFailureAction) {
 		if spec.UpdateConfig == nil {
 			spec.UpdateConfig = &swarm.UpdateConfig{}
 		}
 		updateUint64(flagUpdateParallelism, &spec.UpdateConfig.Parallelism)
 		updateDuration(flagUpdateDelay, &spec.UpdateConfig.Delay)
+		updateString(flagUpdateFailureAction, &spec.UpdateConfig.FailureAction)
 	}
 
 	updateNetworks(flags, &spec.Networks)

contrib/completion/bash/docker

@@ -1726,6 +1726,7 @@ _docker_service_update() {
 		--restart-window
 		--stop-grace-period
 		--update-delay
+		--update-failure-action
 		--update-parallelism
 		--user -u
 		--workdir -w

contrib/completion/zsh/_docker

@@ -1095,6 +1095,7 @@ __docker_service_subcommand() {
         "($help)--restart-window=[Window used to evaluate the restart policy]:window: "
         "($help)--stop-grace-period=[Time to wait before force killing a container]:grace period: "
         "($help)--update-delay=[Delay between updates]:delay: "
+        "($help)--update-failure-action=[Action on update failure]:mode:(pause continue)"
         "($help)--update-parallelism=[Maximum number of tasks updated simultaneously]:number: "
         "($help -u --user)"{-u=,--user=}"[Username or UID]:user:_users"
         "($help)--with-registry-auth[Send registry authentication details to swarm agents]"

daemon/cluster/convert/service.go

@@ -53,6 +53,13 @@ func ServiceFromGRPC(s swarmapi.Service) types.Service {
 		}
 
 		service.Spec.UpdateConfig.Delay, _ = ptypes.Duration(&s.Spec.Update.Delay)
+
+		switch s.Spec.Update.FailureAction {
+		case swarmapi.UpdateConfig_PAUSE:
+			service.Spec.UpdateConfig.FailureAction = types.UpdateFailureActionPause
+		case swarmapi.UpdateConfig_CONTINUE:
+			service.Spec.UpdateConfig.FailureAction = types.UpdateFailureActionContinue
+		}
 	}
 
 	// Mode
@@ -65,6 +72,23 @@ func ServiceFromGRPC(s swarmapi.Service) types.Service {
 		}
 	}
 
+	// UpdateStatus
+	service.UpdateStatus = types.UpdateStatus{}
+	if s.UpdateStatus != nil {
+		switch s.UpdateStatus.State {
+		case swarmapi.UpdateStatus_UPDATING:
+			service.UpdateStatus.State = types.UpdateStateUpdating
+		case swarmapi.UpdateStatus_PAUSED:
+			service.UpdateStatus.State = types.UpdateStatePaused
+		case swarmapi.UpdateStatus_COMPLETED:
+			service.UpdateStatus.State = types.UpdateStateCompleted
+		}
+
+		service.UpdateStatus.StartedAt, _ = ptypes.Timestamp(s.UpdateStatus.StartedAt)
+		service.UpdateStatus.CompletedAt, _ = ptypes.Timestamp(s.UpdateStatus.CompletedAt)
+		service.UpdateStatus.Message = s.UpdateStatus.Message
+	}
+
 	return service
 }
 
@@ -111,9 +135,19 @@ func ServiceSpecToGRPC(s types.ServiceSpec) (swarmapi.ServiceSpec, error) {
 	}
 
 	if s.UpdateConfig != nil {
+		var failureAction swarmapi.UpdateConfig_FailureAction
+		switch s.UpdateConfig.FailureAction {
+		case types.UpdateFailureActionPause, "":
+			failureAction = swarmapi.UpdateConfig_PAUSE
+		case types.UpdateFailureActionContinue:
+			failureAction = swarmapi.UpdateConfig_CONTINUE
+		default:
+			return swarmapi.ServiceSpec{}, fmt.Errorf("unrecongized update failure action %s", s.UpdateConfig.FailureAction)
+		}
 		spec.Update = &swarmapi.UpdateConfig{
 			Parallelism:   s.UpdateConfig.Parallelism,
 			Delay:         *ptypes.DurationProto(s.UpdateConfig.Delay),
+			FailureAction: failureAction,
 		}
 	}
 

docs/reference/api/docker_remote_api_v1.24.md

@@ -3967,7 +3967,8 @@ Create a service
       },
       "UpdateConfig": {
         "Delay": 30000000000.0,
-        "Parallelism": 2
+        "Parallelism": 2,
+        "FailureAction": "pause"
       },
       "EndpointSpec": {
         "Ports": [
@@ -4057,6 +4058,8 @@ JSON Parameters:
     - **Parallelism** – Maximum number of tasks to be updated in one iteration (0 means unlimited
       parallelism).
     - **Delay** – Amount of time between updates.
+    - **FailureAction** - Action to take if an updated task fails to run, or stops running during the
+      update. Values are `continue` and `pause`.
 - **Networks** – Array of network names or IDs to attach the service to.
 - **Endpoint** – Properties that can be configured to access and load balance a service.
     - **Spec** –

docs/reference/commandline/service_create.md

@@ -42,6 +42,7 @@ Options:
       --restart-window value           Window used to evaluate the restart policy (default none)
       --stop-grace-period value        Time to wait before force killing a container (default none)
       --update-delay duration          Delay between updates
+      --update-failure-action string   Action on update failure (pause|continue) (default "pause")
       --update-parallelism uint        Maximum number of tasks updated simultaneously (0 to update all at once) (default 1)
   -u, --user string                    Username or UID
       --with-registry-auth             Send registry authentication details to Swarm agents

docs/reference/commandline/service_update.md

@@ -48,6 +48,7 @@ Options:
       --restart-window value           Window used to evaluate the restart policy (default none)
       --stop-grace-period value        Time to wait before force killing a container (default none)
       --update-delay duration          Delay between updates
+      --update-failure-action string   Action on update failure (pause|continue) (default "pause")
       --update-parallelism uint        Maximum number of tasks updated simultaneously (0 to update all at once) (default 1)
   -u, --user string                    Username or UID
       --with-registry-auth             Send registry authentication details to Swarm agents

integration-cli/docker_api_swarm_test.go

@@ -768,6 +768,7 @@ func serviceForUpdate(s *swarm.Service) {
 		UpdateConfig: &swarm.UpdateConfig{
 			Parallelism:   2,
 			Delay:         8 * time.Second,
+			FailureAction: swarm.UpdateFailureActionContinue,
 		},
 	}
 	s.Spec.Name = "updatetest"