[PATCH 00/80] Kernel based checkpoint/restart [v18]

Daniel Lezcano daniel.lezcano at free.fr
Wed Sep 30 11:01:09 PDT 2009


Dan Smith wrote:
> DL> If the checkpoint is done from the kernel, why the restart
> DL> wouldn't be in the kernel too ?
>
> I think thus far we have taken the approach of "if it can be done
> reasonably in userspace, then do it there" right?  
Well I am a little lost :)
The tty CR can be "reasonably" done in userspace I think. But it was 
done in the kernel, no ?

> Setup of the
> network devices is easy to do in userspace, allows more flexibility
> from a policy standpoint, and ensures that all existing security
> checks are performed. 
Yep, I agree. But you didn't answer the question, what are the network 
resources you plan to checkpoint / restart ?
eg. you let the container to setup your network, will you restore netdev 
statistics ? the mac address ? ipv4 ? ipv6 ?

Is it possible to do a detailed list of network resources you plan to CR 
with the different items you will address from userspace and kernel space ?

> Also, migration may be easier if the userspace
> bits can call custom hooks allowing for routing changes and other
> infrastructure-specific operations.
>   
You may have some problems with the connected sockets you will restore 
in this case.

> DL> Is there any documentation about the statefile format I can use if
> DL> I want to implement myself an userspace CR solution based on this
> DL> kernel patchset ?
>
> See linux-cr/include/linux/checkpoint_hdr.h and user-cr/restart.c.
>   
Argh ! I was hoping there was something else than the source code :)

Thanks
  -- Daniel



More information about the Containers mailing list