Merge pull request #2826 from mheon/restart_policy

Add restart policy for containers

8 files changed, 250 insertions, 13 deletions

diff --git a/libpod/container.go b/libpod/container.go
index 4bf9a1ba9..c07f4c78d 100644
--- a/libpod/container.go
+++ b/libpod/container.go
@@ -102,6 +102,20 @@ func (ns LinuxNS) String() string {
 	}
 }
 
+// Valid restart policy types.
+const (
+	// RestartPolicyNone indicates that no restart policy has been requested
+	// by a container.
+	RestartPolicyNone = ""
+	// RestartPolicyNo is identical in function to RestartPolicyNone.
+	RestartPolicyNo = "no"
+	// RestartPolicyAlways unconditionally restarts the container.
+	RestartPolicyAlways = "always"
+	// RestartPolicyOnFailure restarts the container on non-0 exit code,
+	// with an optional maximum number of retries.
+	RestartPolicyOnFailure = "on-failure"
+)
+
 // Container is a single OCI container.
 // All operations on a Container that access state must begin with a call to
 // syncContainer().
@@ -179,6 +193,16 @@ type ContainerState struct {
 	// This maps the path the file will be mounted to in the container to
 	// the path of the file on disk outside the container
 	BindMounts map[string]string `json:"bindMounts,omitempty"`
+	// StoppedByUser indicates whether the container was stopped by an
+	// explicit call to the Stop() API.
+	StoppedByUser bool `json:"stoppedByUser,omitempty"`
+	// RestartPolicyMatch indicates whether the conditions for restart
+	// policy have been met.
+	RestartPolicyMatch bool `json:"restartPolicyMatch,omitempty"`
+	// RestartCount is how many times the container was restarted by its
+	// restart policy. This is NOT incremented by normal container restarts
+	// (only by restart policy).
+	RestartCount uint `json:"restartCount,omitempty"`
 
 	// ExtensionStageHooks holds hooks which will be executed by libpod
 	// and not delegated to the OCI runtime.
@@ -346,6 +370,17 @@ type ContainerConfig struct {
 	LogPath string `json:"logPath"`
 	// File containing the conmon PID
 	ConmonPidFile string `json:"conmonPidFile,omitempty"`
+	// RestartPolicy indicates what action the container will take upon
+	// exiting naturally.
+	// Allowed options are "no" (take no action), "on-failure" (restart on
+	// non-zero exit code, up an a maximum of RestartRetries times),
+	// and "always" (always restart the container on any exit code).
+	// The empty string is treated as the default ("no")
+	RestartPolicy string `json:"restart_policy,omitempty"`
+	// RestartRetries indicates the number of attempts that will be made to
+	// restart the container. Used only if RestartPolicy is set to
+	// "on-failure".
+	RestartRetries uint `json:"restart_retries,omitempty"`
 	// TODO log options for log drivers
 
 	PostConfigureNetNS bool `json:"postConfigureNetNS"`
@@ -729,6 +764,17 @@ func (c *Container) LogPath() string {
 	return c.config.LogPath
 }
 
+// RestartPolicy returns the container's restart policy.
+func (c *Container) RestartPolicy() string {
+	return c.config.RestartPolicy
+}
+
+// RestartRetries returns the number of retries that will be attempted when
+// using the "on-failure" restart policy
+func (c *Container) RestartRetries() uint {
+	return c.config.RestartRetries
+}
+
 // RuntimeName returns the name of the runtime
 func (c *Container) RuntimeName() string {
 	return c.runtime.ociRuntime.name
@@ -1003,6 +1049,21 @@ func (c *Container) BindMounts() (map[string]string, error) {
 	return newMap, nil
 }
 
+// StoppedByUser returns whether the container was last stopped by an explicit
+// call to the Stop() API, or whether it exited naturally.
+func (c *Container) StoppedByUser() (bool, error) {
+	if !c.batched {
+		c.lock.Lock()
+		defer c.lock.Unlock()
+
+		if err := c.syncContainer(); err != nil {
+			return false, err
+		}
+	}
+
+	return c.state.StoppedByUser, nil
+}
+
 // Misc Accessors
 // Most will require locking
 
diff --git a/libpod/container_api.go b/libpod/container_api.go
index 5bfd869b3..5bb610aab 100644
--- a/libpod/container_api.go
+++ b/libpod/container_api.go
@@ -57,11 +57,11 @@ func (c *Container) Init(ctx context.Context) (err error) {
 
 	if c.state.State == ContainerStateStopped {
 		// Reinitialize the container
-		return c.reinit(ctx)
+		return c.reinit(ctx, false)
 	}
 
 	// Initialize the container for the first time
-	return c.init(ctx)
+	return c.init(ctx, false)
 }
 
 // Start starts a container.
@@ -199,8 +199,15 @@ func (c *Container) Kill(signal uint) error {
 	if c.state.State != ContainerStateRunning {
 		return errors.Wrapf(ErrCtrStateInvalid, "can only kill running containers")
 	}
+
 	defer c.newContainerEvent(events.Kill)
-	return c.runtime.ociRuntime.killContainer(c, signal)
+	if err := c.runtime.ociRuntime.killContainer(c, signal); err != nil {
+		return err
+	}
+
+	c.state.StoppedByUser = true
+
+	return c.save()
 }
 
 // Exec starts a new process inside the container
@@ -583,6 +590,7 @@ func (c *Container) Cleanup(ctx context.Context) error {
 	if !c.batched {
 		c.lock.Lock()
 		defer c.lock.Unlock()
+
 		if err := c.syncContainer(); err != nil {
 			return err
 		}
@@ -593,6 +601,19 @@ func (c *Container) Cleanup(ctx context.Context) error {
 		return errors.Wrapf(ErrCtrStateInvalid, "container %s is running or paused, refusing to clean up", c.ID())
 	}
 
+	// Handle restart policy.
+	// Returns a bool indicating whether we actually restarted.
+	// If we did, don't proceed to cleanup - just exit.
+	didRestart, err := c.handleRestartPolicy(ctx)
+	if err != nil {
+		return err
+	}
+	if didRestart {
+		return nil
+	}
+
+	// If we didn't restart, we perform a normal cleanup
+
 	// Check if we have active exec sessions
 	if len(c.state.ExecSessions) != 0 {
 		return errors.Wrapf(ErrCtrStateInvalid, "container %s has active exec sessions, refusing to clean up", c.ID())
@@ -754,7 +775,7 @@ func (c *Container) Refresh(ctx context.Context) error {
 		if err := c.prepare(); err != nil {
 			return err
 		}
-		if err := c.init(ctx); err != nil {
+		if err := c.init(ctx, false); err != nil {
 			return err
 		}
 	}
diff --git a/libpod/container_inspect.go b/libpod/container_inspect.go
index aa3a07888..a7369bfdd 100644
--- a/libpod/container_inspect.go
+++ b/libpod/container_inspect.go
@@ -95,6 +95,7 @@ func (c *Container) getContainerInspectData(size bool, driverData *inspect.Data)
 		LogPath:         config.LogPath,
 		ConmonPidFile:   config.ConmonPidFile,
 		Name:            config.Name,
+		RestartCount:    int32(runtimeInfo.RestartCount),
 		Driver:          driverData.Name,
 		MountLabel:      config.MountLabel,
 		ProcessLabel:    config.ProcessLabel,
diff --git a/libpod/container_internal.go b/libpod/container_internal.go
index a791df491..0b4e5763e 100644
--- a/libpod/container_internal.go
+++ b/libpod/container_internal.go
@@ -210,6 +210,90 @@ func (c *Container) handleExitFile(exitFile string, fi os.FileInfo) error {
 	return nil
 }
 
+// Handle container restart policy.
+// This is called when a container has exited, and was not explicitly stopped by
+// an API call to stop the container or pod it is in.
+func (c *Container) handleRestartPolicy(ctx context.Context) (restarted bool, err error) {
+	// If we did not get a restart policy match, exit immediately.
+	// Do the same if we're not a policy that restarts.
+	if !c.state.RestartPolicyMatch ||
+		c.config.RestartPolicy == RestartPolicyNo ||
+		c.config.RestartPolicy == RestartPolicyNone {
+		return false, nil
+	}
+
+	// If we're RestartPolicyOnFailure, we need to check retries and exit
+	// code.
+	if c.config.RestartPolicy == RestartPolicyOnFailure {
+		if c.state.ExitCode == 0 {
+			return false, nil
+		}
+
+		// If we don't have a max retries set, continue
+		if c.config.RestartRetries > 0 {
+			if c.state.RestartCount < c.config.RestartRetries {
+				logrus.Debugf("Container %s restart policy trigger: on retry %d (of %d)",
+					c.ID(), c.state.RestartCount, c.config.RestartRetries)
+			} else {
+				logrus.Debugf("Container %s restart policy trigger: retries exhausted", c.ID())
+				return false, nil
+			}
+		}
+	}
+
+	logrus.Debugf("Restarting container %s due to restart policy %s", c.ID(), c.config.RestartPolicy)
+
+	// Need to check if dependencies are alive.
+	if err = c.checkDependenciesAndHandleError(ctx); err != nil {
+		return false, err
+	}
+
+	// Is the container running again?
+	// If so, we don't have to do anything
+	if c.state.State == ContainerStateRunning || c.state.State == ContainerStatePaused {
+		return false, nil
+	} else if c.state.State == ContainerStateUnknown {
+		return false, errors.Wrapf(ErrInternal, "invalid container state encountered in restart attempt!")
+	}
+
+	c.newContainerEvent(events.Restart)
+
+	// Increment restart count
+	c.state.RestartCount = c.state.RestartCount + 1
+	logrus.Debugf("Container %s now on retry %d", c.ID(), c.state.RestartCount)
+	if err := c.save(); err != nil {
+		return false, err
+	}
+
+	defer func() {
+		if err != nil {
+			if err2 := c.cleanup(ctx); err2 != nil {
+				logrus.Errorf("error cleaning up container %s: %v", c.ID(), err2)
+			}
+		}
+	}()
+	if err := c.prepare(); err != nil {
+		return false, err
+	}
+
+	if c.state.State == ContainerStateStopped {
+		// Reinitialize the container if we need to
+		if err := c.reinit(ctx, true); err != nil {
+			return false, err
+		}
+	} else if c.state.State == ContainerStateConfigured ||
+		c.state.State == ContainerStateExited {
+		// Initialize the container
+		if err := c.init(ctx, true); err != nil {
+			return false, err
+		}
+	}
+	if err := c.start(); err != nil {
+		return false, err
+	}
+	return true, nil
+}
+
 // Sync this container with on-disk state and runtime status
 // Should only be called with container lock held
 // This function should suffice to ensure a container's state is accurate and
@@ -230,6 +314,14 @@ func (c *Container) syncContainer() error {
 		}
 		// Only save back to DB if state changed
 		if c.state.State != oldState {
+			// Check for a restart policy match
+			if c.config.RestartPolicy != RestartPolicyNone && c.config.RestartPolicy != RestartPolicyNo &&
+				(oldState == ContainerStateRunning || oldState == ContainerStatePaused) &&
+				(c.state.State == ContainerStateStopped || c.state.State == ContainerStateExited) &&
+				!c.state.StoppedByUser {
+				c.state.RestartPolicyMatch = true
+			}
+
 			if err := c.save(); err != nil {
 				return err
 			}
@@ -376,6 +468,9 @@ func resetState(state *ContainerState) error {
 	state.ExecSessions = make(map[string]*ExecSession)
 	state.NetworkStatus = nil
 	state.BindMounts = make(map[string]string)
+	state.StoppedByUser = false
+	state.RestartPolicyMatch = false
+	state.RestartCount = 0
 
 	return nil
 }
@@ -569,13 +664,13 @@ func (c *Container) prepareToStart(ctx context.Context, recursive bool) (err err
 
 	if c.state.State == ContainerStateStopped {
 		// Reinitialize the container if we need to
-		if err := c.reinit(ctx); err != nil {
+		if err := c.reinit(ctx, false); err != nil {
 			return err
 		}
 	} else if c.state.State == ContainerStateConfigured ||
 		c.state.State == ContainerStateExited {
 		// Or initialize it if necessary
-		if err := c.init(ctx); err != nil {
+		if err := c.init(ctx, false); err != nil {
 			return err
 		}
 	}
@@ -763,7 +858,7 @@ func (c *Container) completeNetworkSetup() error {
 }
 
 // Initialize a container, creating it in the runtime
-func (c *Container) init(ctx context.Context) error {
+func (c *Container) init(ctx context.Context, retainRetries bool) error {
 	span, _ := opentracing.StartSpanFromContext(ctx, "init")
 	span.SetTag("struct", "container")
 	defer span.Finish()
@@ -789,6 +884,12 @@ func (c *Container) init(ctx context.Context) error {
 	c.state.ExitCode = 0
 	c.state.Exited = false
 	c.state.State = ContainerStateCreated
+	c.state.StoppedByUser = false
+	c.state.RestartPolicyMatch = false
+
+	if !retainRetries {
+		c.state.RestartCount = 0
+	}
 
 	if err := c.save(); err != nil {
 		return err
@@ -851,7 +952,7 @@ func (c *Container) cleanupRuntime(ctx context.Context) error {
 // Should only be done on ContainerStateStopped containers.
 // Not necessary for ContainerStateExited - the container has already been
 // removed from the runtime, so init() can proceed freely.
-func (c *Container) reinit(ctx context.Context) error {
+func (c *Container) reinit(ctx context.Context, retainRetries bool) error {
 	span, _ := opentracing.StartSpanFromContext(ctx, "reinit")
 	span.SetTag("struct", "container")
 	defer span.Finish()
@@ -863,7 +964,7 @@ func (c *Container) reinit(ctx context.Context) error {
 	}
 
 	// Initialize the container again
-	return c.init(ctx)
+	return c.init(ctx, retainRetries)
 }
 
 // Initialize (if necessary) and start a container
@@ -901,12 +1002,12 @@ func (c *Container) initAndStart(ctx context.Context) (err error) {
 	if c.state.State == ContainerStateStopped {
 		logrus.Debugf("Recreating container %s in OCI runtime", c.ID())
 
-		if err := c.reinit(ctx); err != nil {
+		if err := c.reinit(ctx, false); err != nil {
 			return err
 		}
 	} else if c.state.State == ContainerStateConfigured ||
 		c.state.State == ContainerStateExited {
-		if err := c.init(ctx); err != nil {
+		if err := c.init(ctx, false); err != nil {
 			return err
 		}
 	}
@@ -950,6 +1051,11 @@ func (c *Container) stop(timeout uint) error {
 		return err
 	}
 
+	c.state.StoppedByUser = true
+	if err := c.save(); err != nil {
+		return errors.Wrapf(err, "error saving container %s state after stopping", c.ID())
+	}
+
 	// Wait until we have an exit file, and sync once we do
 	return c.waitForExitFileAndSync()
 }
@@ -986,6 +1092,8 @@ func (c *Container) restartWithTimeout(ctx context.Context, timeout uint) (err e
 		return errors.Wrapf(ErrCtrStateInvalid, "unable to restart a container in a paused or unknown state")
 	}
 
+	c.newContainerEvent(events.Restart)
+
 	if c.state.State == ContainerStateRunning {
 		if err := c.stop(timeout); err != nil {
 			return err
@@ -1004,13 +1112,13 @@ func (c *Container) restartWithTimeout(ctx context.Context, timeout uint) (err e
 
 	if c.state.State == ContainerStateStopped {
 		// Reinitialize the container if we need to
-		if err := c.reinit(ctx); err != nil {
+		if err := c.reinit(ctx, false); err != nil {
 			return err
 		}
 	} else if c.state.State == ContainerStateConfigured ||
 		c.state.State == ContainerStateExited {
 		// Initialize the container
-		if err := c.init(ctx); err != nil {
+		if err := c.init(ctx, false); err != nil {
 			return err
 		}
 	}
diff --git a/libpod/events/config.go b/libpod/events/config.go
index 36387e835..810988205 100644
--- a/libpod/events/config.go
+++ b/libpod/events/config.go
@@ -134,6 +134,8 @@ const (
 	// Renumber indicates that lock numbers were reallocated at user
 	// request.
 	Renumber Status = "renumber"
+	// Restart indicates the target was restarted via an API call.
+	Restart Status = "restart"
 	// Restore ...
 	Restore Status = "restore"
 	// Save ...
diff --git a/libpod/events/events.go b/libpod/events/events.go
index 202c9db4e..650a47bfb 100644
--- a/libpod/events/events.go
+++ b/libpod/events/events.go
@@ -144,6 +144,8 @@ func StringToStatus(name string) (Status, error) {
 		return Remove, nil
 	case Renumber.String():
 		return Renumber, nil
+	case Restart.String():
+		return Restart, nil
 	case Restore.String():
 		return Restore, nil
 	case Save.String():
diff --git a/libpod/options.go b/libpod/options.go
index 86c04db09..7ec7dfe63 100644
--- a/libpod/options.go
+++ b/libpod/options.go
@@ -1239,6 +1239,41 @@ func WithUseImageHosts() CtrCreateOption {
 	}
 }
 
+// WithRestartPolicy sets the container's restart policy. Valid values are
+// "no", "on-failure", and "always". The empty string is allowed, and will be
+// equivalent to "no".
+func WithRestartPolicy(policy string) CtrCreateOption {
+	return func(ctr *Container) error {
+		if ctr.valid {
+			return ErrCtrFinalized
+		}
+
+		switch policy {
+		case RestartPolicyNone, RestartPolicyNo, RestartPolicyOnFailure, RestartPolicyAlways:
+			ctr.config.RestartPolicy = policy
+		default:
+			return errors.Wrapf(ErrInvalidArg, "%q is not a valid restart policy", policy)
+		}
+
+		return nil
+	}
+}
+
+// WithRestartRetries sets the number of retries to use when restarting a
+// container with the "on-failure" restart policy.
+// 0 is an allowed value, and indicates infinite retries.
+func WithRestartRetries(tries uint) CtrCreateOption {
+	return func(ctr *Container) error {
+		if ctr.valid {
+			return ErrCtrFinalized
+		}
+
+		ctr.config.RestartRetries = tries
+
+		return nil
+	}
+}
+
 // withIsInfra sets the container to be an infra container. This means the container will be sometimes hidden
 // and expected to be the first container in the pod.
 func withIsInfra() CtrCreateOption {
diff --git a/libpod/pod_api.go b/libpod/pod_api.go
index 9a6baf23e..9ed5c88eb 100644
--- a/libpod/pod_api.go
+++ b/libpod/pod_api.go
@@ -364,6 +364,13 @@ func (p *Pod) Kill(signal uint) (map[string]error, error) {
 		}
 
 		logrus.Debugf("Killed container %s with signal %d", ctr.ID(), signal)
+
+		ctr.state.StoppedByUser = true
+		if err := ctr.save(); err != nil {
+			ctrErrors[ctr.ID()] = err
+		}
+
+		ctr.lock.Unlock()
 	}
 
 	if len(ctrErrors) > 0 {