Add --sync option to podman rm

With the changes made recently to ensure Podman does not hit the
OCI runtime as often to sync state, we can find ourselves in a
situation where the runtime's state does not match ours.

Add a --sync flag to podman rm to ensure we can still remove
containers when this happens.

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

cmd/podman/rm.go

@@ -20,6 +20,10 @@ var (
 			Usage: "Force removal of a running container.  The default is false",
 		},
 		LatestFlag,
+		cli.BoolFlag{
+			Name:  "sync",
+			Usage: "Sync container state with OCI runtime before removing",
+		},
 		cli.BoolFlag{
 			Name:  "volumes, v",
 			Usage: "Remove the volumes associated with the container (Not implemented yet)",
@@ -73,6 +77,12 @@ func rmCmd(c *cli.Context) error {
 	for _, container := range delContainers {
 		con := container
 		f := func() error {
+			if c.Bool("sync") {
+				if err := con.Sync(); err != nil {
+					return err
+				}
+			}
+
 			return runtime.RemoveContainer(ctx, con, c.Bool("force"))
 		}
 

completions/bash/podman

@@ -1828,6 +1828,7 @@ _podman_rm() {
 	-h
 	--latest
 	-l
+	--sync
 	--volumes
 	-v
     "

docs/podman-rm.1.md

@@ -24,6 +24,13 @@ Remove all containers.  Can be used in conjunction with -f as well.
 Instead of providing the container name or ID, use the last created container. If you use methods other than Podman
 to run containers such as CRI-O, the last started container could be from either of those methods.
 
+**--sync**
+
+Force a sync of container state with the OCI runtime before attempting to remove.
+In some cases, a container's state in the runtime can become out of sync with Podman's state,
+which can cause Podman to refuse to remove containers because it believes they are still running.
+A sync will resolve this issue.
+
 **--volumes, -v**
 
 Remove the volumes associated with the container. (Not yet implemented)

libpod/container_api.go

@@ -675,22 +675,22 @@ func (c *Container) Batch(batchFunc func(*Container) error) error {
 	return err
 }
 
-// Sync updates the current state of the container, checking whether its state
-// has changed
-// Sync can only be used inside Batch() - otherwise, it will be done
-// automatically.
-// When called outside Batch(), Sync() is a no-op
+// Sync updates the status of a container by querying the OCI runtime.
+// If the container has not been created inside the OCI runtime, nothing will be
+// done.
+// Most of the time, Podman does not explicitly query the OCI runtime for
+// container status, and instead relies upon exit files created by conmon.
+// This can cause a disconnect between running state and what Podman sees in
+// cases where Conmon was killed unexpected, or runc was upgraded.
+// Running a manual Sync() ensures that container state will be correct in
+// such situations.
 func (c *Container) Sync() error {
-	if !c.batched {
-		return nil
-	}
-
 	// If runtime knows about the container, update its status in runtime
 	// And then save back to disk
 	if (c.state.State != ContainerStateUnknown) &&
-		(c.state.State != ContainerStateConfigured) {
+		(c.state.State != ContainerStateConfigured) &&
+		(c.state.State != ContainerStateExited) {
 		oldState := c.state.State
-		// TODO: optionally replace this with a stat for the exit file
 		if err := c.runtime.ociRuntime.updateContainerStatus(c, true); err != nil {
 			return err
 		}