On Fri, May 13, 2022 at 3:20 PM przemek klosowski via devel
<devel(a)lists.fedoraproject.org> wrote:
I had issues with this for several releases now, although I never got a
broken upgrade like you're reporting: every time so far I got a message
'you need additional 4GB on /' before the actual upgrade started. It
would be sad if the free space estimation code didn't work and allowed
broken, half-upgraded systems!!
I wonder if the message pointing to `/` is ambiguous? Because if you
have a separate /var, then referring to `/` isn't really giving the
correct advice is it?
Note that as of Fedora 36, the rpmdb is in /usr which *is* on `/` -
but note that the switch doesn't happen until the first boot following
the upgrade to Fedora 36. So /var is still the active location for
rpmdb during the upgrade.
Unfortunately, I believe that the current upgrade workflow requires
a
root disk three times the total installed package size: each package is
there as the original version, the RPM of the upgrade in the cache
directory, and as a new image before the old one is purged.
Yeah I'm not sure it's currently possible to do an atomic update with
dnf or rpm, hence rpm-ostree. Or txnupd and btrfs snapshots
(essentially dnf and rpm keep doing what they're doing, but create a
snapshot of the running system, and update the snapshot out of band,
rather than stomping on the currently running OS).
--
Chris Murphy