checkpoint: set default work-dir to image-path

[liusdu]

Sump/restrore logs are very useful when process fails.
So put them to image path defaultly，otherwise these files
will be deleted when container exits.

Signed-off-by: Liu Hua [email protected]'

Proposed changelog entry

• checkpoint/restore: if --work-path is not specified, it defaults to the value of --image-path (same as criu does).

[kolyshkin]

What prevents you from using --work-path instead of --image-path to achieve the same results? If --image-path is not set, it defaults to the value of --work-path, see https://criu.org/Directories).

Ahh I see -- standalone criu binary works as described above, but runc c/r doesn't.

I think we need to modify runc to NOT set WorkDirectory (and CriuOpts.WorkDirFd) if --work-path is not specified. This way runc checkpoint and runc restore will work the same way as criu, which I think is a sane way.

@liusdu can you implement the above?

[liusdu]

What prevents you from using --work-path instead of --image-path to achieve the same results? If --image-path is not set, it defaults to the value of --work-path, see https://criu.org/Directories).

Hi @kolyshkin, thanks to you reply. if --images-dir is not set, it defaults to the value of "$(pwd)", not --work-dir. The description is as following:

This is the one specified with the -D|--images-dir $path option. If not given CRIU uses current working directory.

And if --work-dir is not set, it defaults to the values to --images-dir see https://github.com/checkpoint-restore/criu/blob/criu-dev/criu/crtools.c#L133

Ahh I see -- standalone criu binary works as described above, but runc c/r doesn't.

I think we need to modify runc to NOT set WorkDirectory (and CriuOpts.WorkDirFd) if --work-path is not specified. This way runc checkpoint and runc restore will work the same way as criu, which I think is a sane way.

@liusdu can you implement the above?

Yes, It can fix it this way~

[liusdu]

@kolyshkin done, please take a look.

[kolyshkin]

@liusdu needs a rebase

[liusdu]

@kolyshkin sorry for this, done~

[kolyshkin]

Suggested change

	// if criuOpts.WorkDirectory is not set. let criu sets it.
	// If criuOpts.WorkDirectory is not set, criu default is used
	// (which is to reuse ImagesDirectory as WorkDirectory).

[kolyshkin]

I think you need to amend the commit message to explain how it works (i.e. by not setting the WorkDirectory we let criu use its default, which is to reuse ImagesDirectory, or some such).

[kolyshkin]

A few typos in the commit message:

Dump/restrore logs are very useful when process fails.

restore

New runc puts these files in c.root, which will be deleted

I think you mean "Now".

when container exits. So if checkpinting/restoring failed,

checkpointing

[kolyshkin]

LGTM

@adrianreber PTAL

[adrianreber]

The only thing I am worried about is if Docker or Podman are relying on this somehow. Would be good to verify that neither of them break with this change.

[liusdu]

@adrianreber Podman and Docker/Containerd all set WorkDirectory. So this change does not affect them.
And I think behavior of containers is not good enough, because we can not get restore log when restoring failed.

1. Podman:

• Code
  Podman set WorkDirectory to ctr.bundlePath()

       // workPath will be used to store dump.log and stats-dump
        workPath := ctr.bundlePath()
        logrus.Debugf("Writing checkpoint to %s", imagePath)
        logrus.Debugf("Writing checkpoint logs to %s", workPath)
        logrus.Debugf("Pre-dump the container %t", options.PreCheckPoint)
        args := []string{}
        args = append(args, r.runtimeFlags...)
        args = append(args, "checkpoint")
        args = append(args, "--image-path")
        args = append(args, imagePath)
        args = append(args, "--work-path")
        args = append(args, workPath)

• behavior:

# restore command
podman  container restore  06fd1a24a1f0992c9349e7de45796b80e253a10fc1b7adcfc89052cd460a4288
# output
Error: OCI runtime error: criu failed: type NOTIFY errno 0
log file: /var/lib/containers/storage/overlay-containers/06fd1a24a1f0992c9349e7de45796b80e253a10fc1b7adcfc89052cd460a4288/userdata/restore.log
# restore.log is perisist on disk
ls -l /var/lib/containers/storage/overlay-containers/06fd1a24a1f0992c9349e7de45796b80e253a10fc1b7adcfc89052cd460a4288/userdata/restore.log 
-rw------- 1 root root 25921 6月  17 11:03 /var/lib/containers/storage/overlay-containers/06fd1a24a1f0992c9349e7de45796b80e253a10fc1b7adcfc89052cd460a4288/userdata/restore.log

2. Containerd:

• code

if p.CriuWorkPath == "" {
              // if criu work path not set, use container WorkDir
              p.CriuWorkPath = p.WorkDir
      }

• behavior

# restore 
docker start --checkpoint checkpoint1 c1b1a73a91
# restore output
Error response from daemon: failed to retrieve OCI runtime container pid: open /var/run/docker/containerd/daemon/io.containerd.runtime.v1.linux/moby/c1b1a73a91b8bbbf61318f7bfa391a8ce6ad42952aba10076372e60e3a34b7ae/init.pid: no such file or directory: unknown
#  I can not found restore.log
find  /var  -name restore.log

[adrianreber]

Thanks for looking into it. So Docker and Podman both seem to always set WorkDirectory and ImagesDirectory.

This PR should not lead to different behaviour on the container engine level.

Patch LGTM

[mrunalp]

LGTM