[Dirvish] hardlink command

Paul Slootman paul at debian.org
Wed Sep 11 12:20:46 UTC 2013

On Wed 11 Sep 2013, Dave Howorth wrote:
> Paul Slootman wrote:

> > That's where the concept of branches come in.
> > - make a backup of a vanilla installation
> > - branch that to backups of specific servers, that are based on that
> >   vanilla installation.
> I've never used branches, so I have an idle curiosity question :)
> If I do as you suggest and set up two branches for two identical servers
> and then there is an update to say the kernel, do I get two copies of
> the new kernel, one in each branch, or does it somehow know to link
> them? Presumably there would be two new initrd files although the binary
> contents might be the same?

No, once branched they go their separate way. Only the files that
originally were identical are hardlinked. At least, that's my
understanding of of the situation ;-)

It's entirely possible that some additional magic is added that adds
--link-dest for each branch to the rsync invocation, but I haven't
checked that.

> I have to say, I've always thought of system images as being small
> enough that I don't worry about sharing backup space for them, but then
> I don't look after a few thousand servers or anything.

I used to look after a bit less than a thousand, although about half
were vservers that had been hashified, i.e. virtual servers that share a
common storage so that identical files can be hardlinked together.
Typically about 8 vservers would run on one hardware server.


More information about the Dirvish mailing list