Add ExecDied event and use it to retrieve exit codes

When making Exec Cleanup processes mandatory, I introduced a race
wherein attached exec sessions could be cleaned up and removed by
the cleanup process before the frontend had a chance to get their
exit code. Fortunately, we've dealt with this issue before in
containers, and the same solution can be applied here. I added an
event for an exec session's process exiting, `exec_died` (Docker
has an identical event, so this actually improves our
compatibility there) that includes the exit code of the exec
session. If the race happens and the exec session no longer
exists when we go to remove it, pick up exit code from the event
and exit cleanly.

Signed-off-by: Matthew Heon <mheon@redhat.com>

4 files changed, 82 insertions, 25 deletions

diff --git a/libpod/container_exec.go b/libpod/container_exec.go
index 09a5b9597..737bf74ad 100644
--- a/libpod/container_exec.go
+++ b/libpod/container_exec.go
@@ -1,6 +1,7 @@
 package libpod
 
 import (
+	"context"
 	"io/ioutil"
 	"net/http"
 	"os"
@@ -539,18 +540,7 @@ func (c *Container) ExecStop(sessionID string, timeout *uint) error {
 	var cleanupErr error
 
 	// Retrieve exit code and update status
-	exitCode, err := c.readExecExitCode(session.ID())
-	if err != nil {
-		cleanupErr = err
-	}
-	session.ExitCode = exitCode
-	session.PID = 0
-	session.State = define.ExecStateStopped
-
-	if err := c.save(); err != nil {
-		if cleanupErr != nil {
-			logrus.Errorf("Error stopping container %s exec session %s: %v", c.ID(), session.ID(), cleanupErr)
-		}
+	if err := retrieveAndWriteExecExitCode(c, session.ID()); err != nil {
 		cleanupErr = err
 	}
 
@@ -592,15 +582,7 @@ func (c *Container) ExecCleanup(sessionID string) error {
 			return errors.Wrapf(define.ErrExecSessionStateInvalid, "cannot clean up container %s exec session %s as it is running", c.ID(), session.ID())
 		}
 
-		exitCode, err := c.readExecExitCode(session.ID())
-		if err != nil {
-			return err
-		}
-		session.ExitCode = exitCode
-		session.PID = 0
-		session.State = define.ExecStateStopped
-
-		if err := c.save(); err != nil {
+		if err := retrieveAndWriteExecExitCode(c, session.ID()); err != nil {
 			return err
 		}
 	}
@@ -637,9 +619,9 @@ func (c *Container) ExecRemove(sessionID string, force bool) error {
 			return err
 		}
 		if !running {
-			session.State = define.ExecStateStopped
-			// TODO: should we retrieve exit code here?
-			// TODO: Might be worth saving state here.
+			if err := retrieveAndWriteExecExitCode(c, session.ID()); err != nil {
+				return err
+			}
 		}
 	}
 
@@ -653,6 +635,10 @@ func (c *Container) ExecRemove(sessionID string, force bool) error {
 			return err
 		}
 
+		if err := retrieveAndWriteExecExitCode(c, session.ID()); err != nil {
+			return err
+		}
+
 		if err := c.cleanupExecBundle(session.ID()); err != nil {
 			return err
 		}
@@ -757,6 +743,18 @@ func (c *Container) Exec(config *ExecConfig, streams *define.AttachStreams, resi
 
 	session, err := c.ExecSession(sessionID)
 	if err != nil {
+		if errors.Cause(err) == define.ErrNoSuchExecSession {
+			// TODO: If a proper Context is ever plumbed in here, we
+			// should use it.
+			// As things stand, though, it's not worth it - this
+			// should always terminate quickly since it's not
+			// streaming.
+			diedEvent, err := c.runtime.GetExecDiedEvent(context.Background(), c.ID(), sessionID)
+			if err != nil {
+				return -1, errors.Wrapf(err, "error retrieving exec session %s exit code", sessionID)
+			}
+			return diedEvent.ContainerExitCode, nil
+		}
 		return -1, err
 	}
 	exitCode := session.ExitCode
@@ -930,6 +928,8 @@ func (c *Container) getActiveExecSessions() ([]string, error) {
 				session.PID = 0
 				session.State = define.ExecStateStopped
 
+				c.newExecDiedEvent(session.ID(), exitCode)
+
 				needSave = true
 			}
 			if err := c.cleanupExecBundle(id); err != nil {
@@ -1039,6 +1039,22 @@ func writeExecExitCode(c *Container, sessionID string, exitCode int) error {
 		return errors.Wrapf(err, "error syncing container %s state to remove exec session %s", c.ID(), sessionID)
 	}
 
+	return justWriteExecExitCode(c, sessionID, exitCode)
+}
+
+func retrieveAndWriteExecExitCode(c *Container, sessionID string) error {
+	exitCode, err := c.readExecExitCode(sessionID)
+	if err != nil {
+		return err
+	}
+
+	return justWriteExecExitCode(c, sessionID, exitCode)
+}
+
+func justWriteExecExitCode(c *Container, sessionID string, exitCode int) error {
+	// Write an event first
+	c.newExecDiedEvent(sessionID, exitCode)
+
 	session, ok := c.state.ExecSessions[sessionID]
 	if !ok {
 		// Exec session already removed.
diff --git a/libpod/events.go b/libpod/events.go
index 839229674..22c51aeec 100644
--- a/libpod/events.go
+++ b/libpod/events.go
@@ -46,7 +46,22 @@ func (c *Container) newContainerExitedEvent(exitCode int32) {
 	e.Type = events.Container
 	e.ContainerExitCode = int(exitCode)
 	if err := c.runtime.eventer.Write(e); err != nil {
-		logrus.Errorf("unable to write pod event: %q", err)
+		logrus.Errorf("unable to write container exited event: %q", err)
+	}
+}
+
+// newExecDiedEvent creates a new event for an exec session's death
+func (c *Container) newExecDiedEvent(sessionID string, exitCode int) {
+	e := events.NewEvent(events.ExecDied)
+	e.ID = c.ID()
+	e.Name = c.Name()
+	e.Image = c.config.RootfsImageName
+	e.Type = events.Container
+	e.ContainerExitCode = exitCode
+	e.Attributes = make(map[string]string)
+	e.Attributes["execID"] = sessionID
+	if err := c.runtime.eventer.Write(e); err != nil {
+		logrus.Errorf("unable to write exec died event: %q", err)
 	}
 }
 
@@ -154,3 +169,25 @@ func (r *Runtime) GetLastContainerEvent(ctx context.Context, nameOrID string, co
 	// return the last element in the slice
 	return containerEvents[len(containerEvents)-1], nil
 }
+
+// GetExecDiedEvent takes a container name or ID, exec session ID, and returns
+// that exec session's Died event (if it has already occurred).
+func (r *Runtime) GetExecDiedEvent(ctx context.Context, nameOrID, execSessionID string) (*events.Event, error) {
+	filters := []string{
+		fmt.Sprintf("container=%s", nameOrID),
+		"event=exec_died",
+		"type=container",
+		fmt.Sprintf("label=execID=%s", execSessionID),
+	}
+
+	containerEvents, err := r.GetEvents(ctx, filters)
+	if err != nil {
+		return nil, err
+	}
+	// There *should* only be one event maximum.
+	// But... just in case... let's not blow up if there's more than one.
+	if len(containerEvents) < 1 {
+		return nil, errors.Wrapf(events.ErrEventNotFound, "exec died event for session %s (container %s) not found", execSessionID, nameOrID)
+	}
+	return containerEvents[len(containerEvents)-1], nil
+}
diff --git a/libpod/events/config.go b/libpod/events/config.go
index 085fa9d52..d88d7b6e3 100644
--- a/libpod/events/config.go
+++ b/libpod/events/config.go
@@ -127,6 +127,8 @@ const (
 	Create Status = "create"
 	// Exec ...
 	Exec Status = "exec"
+	// ExecDied indicates that an exec session in a container died.
+	ExecDied Status = "exec_died"
 	// Exited indicates that a container's process died
 	Exited Status = "died"
 	// Export ...
diff --git a/libpod/events/events.go b/libpod/events/events.go
index 01ea6a386..e03215eff 100644
--- a/libpod/events/events.go
+++ b/libpod/events/events.go
@@ -149,6 +149,8 @@ func StringToStatus(name string) (Status, error) {
 		return Create, nil
 	case Exec.String():
 		return Exec, nil
+	case ExecDied.String():
+		return ExecDied, nil
 	case Exited.String():
 		return Exited, nil
 	case Export.String():