SU+J on 9.1-RC2 ISO

[Nathan Whitehorn]

There's an existing checkbox to disable it. There was substantial 
consensus for 9.0 that SUJ was something we wanted -- I'd personally be 
very hesitant to change the defaults without more input from FS people. 
I think this discussion should be moved to freebsd-fs@ or 
freebsd-current@ instead of stable@ since it's actually a filesystem 
issue not an installer issue.
-Nathan

On 11/03/12 14:09, Jeremy Chadwick wrote:
> (Please keep me CC'd, as I'm not subscribed to -stable)
>
> I've CC'd Nathan Whitehorn, who according to bsdinstall(8) is the
> author (not sure if maintainer) of the code.
>
> This default has already begun to bite users/SAs in the ass:
>
> http://lists.freebsd.org/pipermail/freebsd-questions/2012-November/246069.html
>
> SU+J (the journalling part specifically) needs to be disabled by default
> in the installer.  This default was a very bad choice and should not
> have been done.  It either indicates someone was out of touch with the
> rest of the issues surrounding the feature, or that someone
> intentionally decided "it's the best way to get people using it for
> testing" (I have seen this justification presented in the past, and it
> is the wrong approach).
>
> However, since some people DO want it (and those folks don't use dump),
> the installer should be modified to make SU+J support toggleable via a a
> checkbox.  The default, obviously, should be unchecked.
>
> If the user checks the checkbox, an ominous warning message should be
> displayed informing the user of the repercussions.  The only option at
> that point should be "OK", after which the checkbox is checked.
>
> Do not tell me "send patches".  This issue/problem has gone on long
> enough, and the community bitched hard/long enough, that the person who
> committed this default should be responsible for fixing it.
>
> We should operate under the assumption that this bug/problem will never
> be fixed.  It probably will be, but again, we must operate with the
> assumption that Kirk et al will require years to fix it.  (It has
> already been something like 9 months.  Or is it a year?)
>
> While I'm here -- does anyone remember the exact commit which was done
> sometime in the past 6-9 months which "made the installer work properly
> with SSDs" (re: partition alignment)?  I have questions about that; if I
> remember right, someone set the alignment size to 4KBytes, and that is
> completely 100% wrong -- it needs to be 1MByte or 2MBytes if you want to
> be extra cautious, which correlates with NAND erase block size,
> otherwise we're not really solving jack squat.
>