Hi, all. I hope RFC flow from me are not hitting some rate limiting.
Integration tests in our compary undercover that host uuid provided thru dmi
is not as unique as it should be. This is probably a manufacturer mistake but
still the case.
Here libvirt's config field host uuid comes to the rescue. It is a good option
in case of tweaking a few machines. In case of full-fledged mgmt it is probably
enough too. But in case your managment is managment of packages thru you
package manager it is not that convinent. The problem is that if one generates
uuid on package install then package update will not be easy. One probably
wants for updates to go smoothly in case user does not tweak anything but now
it is not possible as changing host uuid is visible as user intervention by
package managment. I'm talking now from my rpm based managment experience but
I imagine this could be no different in other packaging systems too. This is
probably could be overcomed thru some kind of reserse patching before update
and patching again after if package system is wise enougth to check file hashes
and not timestamps for example. But then this is should be done in this and
every packaging system.
So my proposition is simple - make it possible to take host uuid from distinct
file. I think keeping uuid could be the sole purpuse of it. Then we can
generate uuid on package installation and later updates need no any extra
actions at all.
Another benefit of keeping uuid in distinct place is that one can just copy
daemon configuration to a different host not being aware it needs to be
changed.