On 9 Jun 2000, Guillaume Cottenceau wrote:
> We're thinking of something like HelixUpdate -- but we can't come to a
> whole agreement at MandrakeSoft, some of us think we should only install
> only applicative stuff,
> some others think we should handle a full upgrade
> process this way (with the machine up).
Great idea, that!
I tried to upgrade a server from 6.1 to 7.0. I first grabbed rpmdrake and
used it to uninstall a maximum of things with automatical dependencies,
and then upgrade needed package, without halting the services or rebooting
the server.
Then I had problems with xfs (and others)...
Services did continue to work, but we had a 6 hours power failure a few
days later. Lilo did not work (it is a full SCSI machine), did not find
initrd image for scsi module or something.
After booting from floppy, I copied /etc to another partition, then
installed 7.0 on mke2fs'ed / and /usr partition at a time I could afford
having the server down.
Still, an upgrade without reboot could be appreciated, but I guess it is
*much* more difficult to *guarantee* it will work with any combination of
glibc/processes/server-processes... is it bad to launch a /bin/bash in a
rpm upgrade script when glibc is upgraded and the other rpms not ? I guess
a major "unresolved symbol" syndrom may occur... :-(
Someone told me that Debian updates itself completely without rebooting
(except kernel change, I guess). How do they do that ?
--
Stéphane Gourichon - Laboratoire d'Informatique de Paris 6 - Équipe AnimatLab
