Kernel Live Patch...

Kernel Live Patch...

[Dennis Ullrich]

Dear all,

I want to make a proposal which is similar to online kernel patching 
without reboot... The idea is to do something which could work similar 
to the live kernel patching method. I don't know how difficult it would 
be, to implement my proposal, but here it comes.

For example I have a HDD with one partiton and a Oracle DB on the ext4 
filesystem. I want to migrate it live to another storage (for example a 
LPFC LUN via Multipath) with all processes etc. on the FS.
My idea is to modify the IOStack that it is able to make a "bitwise" 
copy and "simply" move the Filesystem to its new place on a blockdevice.
The point is, to have control about the copy-progress and tell the 
IOStack which device is to use for the requested data, on every access. 
When the copy is done, the admin should be able to remove the old drive. 
All processes should be on the new blockdevice after the movement.

What do you think about this idea?


Best regards,
Dennis

P.S. I know that there are solutions (md, lvm) to do this, if its 
configured in the first place. But you will need a downtime for the 
final switch between new and old (because of the processes).
But in this case I really mean a function which is independent from any 
partiton table, filesystem etc.
and works only on blockdevices.

Re: Kernel Live Patch...

[Richard Weinberger]

On Fri, May 8, 2015 at 8:04 PM, Dennis Ullrich <dennis1ullrich@gmx.de> wrote:
> Dear all,
>
> I want to make a proposal which is similar to online kernel patching without
> reboot... The idea is to do something which could work similar to the live
> kernel patching method. I don't know how difficult it would be, to implement
> my proposal, but here it comes.
>
> For example I have a HDD with one partiton and a Oracle DB on the ext4
> filesystem. I want to migrate it live to another storage (for example a LPFC
> LUN via Multipath) with all processes etc. on the FS.
> My idea is to modify the IOStack that it is able to make a "bitwise" copy
> and "simply" move the Filesystem to its new place on a blockdevice.
> The point is, to have control about the copy-progress and tell the IOStack
> which device is to use for the requested data, on every access. When the
> copy is done, the admin should be able to remove the old drive. All
> processes should be on the new blockdevice after the movement.
>
> What do you think about this idea?

Why can't you stop the application for a jiffy?
Live patching and stuff is nice but if your service can't deal with
a short outage of a single application it is simple broken by design.

-- 
Thanks,
//richard