[CRIU] About p_haul_docker work-flow?

[Pavel Emelyanov]

On 01/11/2016 07:19 AM, Hui Kang wrote:
> Yes, but some minor inline comment.
> 
> - Hui
> 
> On Sun, Jan 10, 2016 at 9:01 PM, Dengguangxing <dengguangxing at huawei.com> wrote:
>> Hi Hui,
>>
>> I am studying P.HAUL recently. Going through p_haul_docker source-code,
>> I get the docker work-flow like this:
>>
>> 1. criu pre-dump and dump, send images to target host
> 
> AFAIK, docker does not support pre-dump. So phaul only rsyncs the
> container filesystem.

Actually for pre-dump we don't need docker support. Just knowing the
pid of the root process is enough.

>> 2. send container rootfs, container-config, execdriver meta-images to target host
>> 3. on target host, restart docker daemon to reload new-added container according to container files above
>> 4. call criu restore ...
>>
>> Not sure if I get this right?
>>
>> And restarting docker daemon may do harm to current-running containers. Are there other ways to achieve this?
>> like create a new container according to specific config and perform restore to it?
> 
> I think a new container with the same config and rootfs is possible.
> But a better solution is to modify docker.
> 
> - Hui
> 
>>
>> Thanks
>>
>> Deng Guangxing
>>
> _______________________________________________
> CRIU mailing list
> CRIU at openvz.org
> https://lists.openvz.org/mailman/listinfo/criu
> .
>