[PLUG] CCC

Russell Senior russell at personaltelco.net
Mon Feb 25 06:37:01 UTC 2019


On Sun, Feb 24, 2019 at 8:04 PM Rigel Hope <gnu at rigelhope.org> wrote:

> "For this use case it probably wouldn't even work since it can't do
> network transfers."
>
> pipe to tar, pipe to netcat. sha256sum afterwards.
>

We needed to transition some VMs from one machine to another one, a year or
so ago. The two machines were in the same data center but in different
racks at different ISPs. When we cut over the VMs I wanted to minimize
downtime. The VM storage allocations were something like 200GB each. Also,
I discovered that rsync didn't know how to work on block devices. I think
there was a patch floating around that let it work on block devices, but
for whatever reason it wasn't practical. So, I dd'd the mounted block
device, knowing full well that it wasn't stable, to a file on host A, and
rsync'd that file across to host B. This took hours. Most of the block
device would still be the same when I was ready to cut over.

When we were ready to roll (in the middle of the night, hopefully), I
shutdown the VM on host A, dd'd the block device to the file again, rsync'd
the file across the network to a file on host B, dd'd that file to host B's
block device and booted the VM back up on host B. The total downtime per VM
was about 20 minutes, iirc. I'm sure I'm leaving out some details, as it
was a while ago. I'm recalling copying and hand editing some kvm/qemu
metadata as well.

I'm not knocking "backup software", but I felt happier knowing that if
something went wrong somewhere in the process, I'd know how to fix it.

The main idea I try to follow is to leave a path of retreat and "bring the
60m rope".
https://www.coldmountainkit.com/knowledge/historical-facts/toni-kurz-died-on-the-eiger/

-- 
Russell Senior
russell at personaltelco.net



More information about the PLUG mailing list