Support checkpoint and restore

[Furisto]

Checkpoint and restore is supported by runc. We should also support these operations. There does not seem to be a crate that allows interacting with criu, so we probably have to write it ourselves. The go implementation would be a good starting point. If anyone has more information on this topic it would be appreciated.

[lizhemingi]

Is there any thing I can help ?

[utam0k]

@Furisto Have you started implementing this yet?

[Furisto]

I have started with the implementation. Will let you know if I need support or if we can divide it up.

[Furisto]

Hey @duduainankai I could use your help. What I have done so far:

• I have generated the code for the criu protobuf messages
• I can start criu in swrk mode
• I can send a message to criu to dump a simple process
• The message is processed and the process successfully dumped according to the criu logs. The image files are written to the output folder.
• I am waiting to receive a response from criu to tell me everything went well
  ... and nothing happens, I am never receiving a response

I am following the steps outlined here and in runc. Maybe you have an idea?

[lizhemingi]

Hey @duduainankai I could use your help. What I have done so far:

• I have generated the code for the criu protobuf messages
• I can start criu in swrk mode
• I can send a message to criu to dump a simple process
• The message is processed and the process successfully dumped according to the criu logs. The image files are written to the output folder.
• I am waiting to receive a response from criu to tell me everything went well
  ... and nothing happens, I am never receiving a response

I am following the steps outlined here and in runc. Maybe you have an idea?

Got. I will check on what you have listed and see what I can do. @Furisto

[adrianreber]

I started the discussion about a CRIU rust interface here checkpoint-restore/criu#1722

[Furisto]

@adrianreber Thanks! Will take a look.

[adrianreber]

I tried to checkpoint a container and it almost works. The main problem currently is that -d --detach is missing which does a setsid(). Are there any plans to implement --detach soon? With --detach checkpointing should be possible pretty easily.

[adrianreber]

Checkpointing works now, if I do setsid() after starting the container with stdin, stdout, stderr redirected to /dev/null.

It is also important that the youki re-opens /dev/null inside the container like I did it for crun: containers/crun@bbb1fa9

This is also needed: #623

[utam0k]

@adrianreber
Awesome!
Oh, I forgot about it 😭 Can I ask you to create an issue?

Are there any plans to implement --detach soon?

[adrianreber]

• Add --detach -d option #628
• Create device as 0666 and not 066 #627
• Re-open /dev/null inside the container #629

[YJDoc2]

Sorry to ping the issue subscribers here, but is this still relevant? Two of the thing mentioned in above have been merged/closed, and the first issue has a related PR which has been merged. If I recall correctly, there are also some integration tests regarding the checkpoint-restore functionality. If this support is done, can we close this issue? If not, what else might be needed for it to work?

[adrianreber]

I only implemented the checkpoint part. I did not find the time yet to implement the restore part of the code. Once restore is implemented this can be closed.