[CRIU] Docker checkpoint restore demo

[Saied Kazemi]

Nice!  An interesting use case of leaving the container running after
checkpoint.

It looks like C/R is not a priority for Docker and it may take a long
time before this functionality is merged.  If this is indeed the case,
it'd be major pain to constantly rebase as both Docker and CRIU are
changing rapidly.

--Saied


On Tue, May 12, 2015 at 11:48 AM, Ross Boucher <rboucher at gmail.com> wrote:
> I just thought I'd share a quick video of what I'm using Criu + Docker for,
> in case anyone was interested, and to say thanks for all the help:
>
> https://www.youtube.com/watch?v=F2L6JLFuFWs
>
> When the document get executed, the code blocks are sent to a server, which
> spins up a new Docker container and then runs the code. In between each
> block, the running process is checkpointed (but left running). Every block
> also has its own checksum, so on subsequent runs, we can examine the
> checksums in order, and if we have a matching checkpoint we restore from
> that point in the document and resume executing the new code.
>
> If we aren't resuming, we pull a container from a pool of running containers
> that are waiting for their first code block. When resuming, we can create a
> new container from the base image and restore directly into it without first
> needing to start it up. We even commit the docker filesystem each time if it
> has changed, so we choose the correct docker image to restore into.
>
> All of the necessary changes in libcontainer are pending on the criu branch,
> and my branch for support in docker is available here:
>
> https://github.com/boucher/docker/tree/boucher-cr
>
> Thanks again for the help. I'm looking forward to seeing criu support get
> merged into Docker.
>
> -Ross
>
>