On Mar 1, 2014, at 1:12 AM, Adam Williamson <awilliam(a)redhat.com> wrote:
On Fri, 2014-02-28 at 20:11 -0700, Chris Murphy wrote:
> There are advantages for server using XFS, even for the smaller
> percent (?) who may end up using the default installation path.
> There's no negative I think of for Workstation using XFS. So I'd say
> make them both XFS.
The xfs negative I can see is the resizing thing. That's about it,
really. (For anyone who's not aware: you can't shrink xfs partitions,
currently. That means that people won't be able to shrink their Fedora
install to install something else alongside it later).
I see the "install something else alongside Fedora" made most straightforward
with plain ext4. [1]
LVMthinp is a technically better workaround, even better than ext4 shrink, but it's
esoteric. It requires the same capability as [1] but in addition the 2nd installer needs
to: a. understand LVMthinp, b. permit thinpool overcomit which even anaconda presently
doesn't. [2]
Therefore if plaint ext4 isn't on the table for Workstation, we're back to
Workstation (ideally) mimicking what Server does.
Chris Murphy
[1] Even with ext4+LVM it would mean either a.) two something's sharing one /boot,
i.e. doable but not best practices; or .b) the followup installer has do LVM gymnastics
galore to extract space for a new plain partition. I don't even know if anaconda
custom partitioning does that.
[2] Anaconda is probably being overly conservative, by disallowing any amount of
overcommitting, which is the whole idea behind thin provisioning: setting the maximum
practical size for each file system's LV, and do a one time format, and the LV only
takes from the thinpool what it actually uses. It obviates both grow and shrink resizing.