On Fri, Jun 26, 2020 at 3:22 PM Przemek Klosowski via devel
I remember that two issues that made me apprehensive wrt. BTRFS were its
handling of the 'disk full' situation, and lack of a staightforward
'fsck' workflow. I think the first issue has been resolved, and we
probably just need some docs and scripts that handle file system
corruption by remounting R/O and printing some suggestions what to do next.
A medium term goal is to make systemd and the desktop environment more
tolerant to starting up read-only, and even though this is a limited
environment the user isn't just stuck at a prompt. SUSE/openSUSE can
today boot read-only snapshots as part of its rollback strategy but
I'm not sure how/why it works or whether it's adaptable.
A short term goal, possibly even a requirement for the proposal, is
some kind of message at a dracut prompt to at least give the user
something to go on, in sequence, including even 'join us on
#fedora-btrfs' or whatever. A bigger problem is that right now (a) new
installs don't set a password for root user, and (b) systemd emergency
target requires a root user login to get to a prompt. It has to be a
mount *failure* to get to a dracut prompt where we could show some
messages. There is this middle area where the user is stuck no matter
the file system.
Some of these are long standing problems, but they're perhaps being
spotlit by the change.
For what it's worth, this is really needed, and overdue. I have
repeatedly failed Fedora OS release upgrades on different machines by
running out of root fs space. I think the default / is around 50GB, and
it's too easy to fill: during OS update we need space for three copies
of each package: the old version, the downloaded new version, and the
space to install the new version.
75G on new installs today but yes there are many folks still with a
50G root volume at /
And changing this to 80+G is sorta 'kick the can' but also as it turns
out it doesn't really fix the problem that well and puts pressure on
/home in cases where the laptop drive is kinda small. There are other
valid ways to solve this single problem, e.g. a single plain ext4 or
xfs volume. But both of those leave things on the table users benefit
Of course it isn't all about features. If it's just a feature contest
btrfs wins somewhat dramatically. What's going to make the feature
successful is the community backing it up. The change needs the desire
and resources of Fedora more than just features. A dozen owners on the
proposal hopefully gives confidence that it's serious, but it's going
to take more than that.