Feature proposal: Rebootless Installer

[Douglas McClendon]

Jesse Keating wrote:
> On Tue, 2009-07-14 at 11:19 -0400, Colin Walters wrote:
>> There's no such step in the live install (right?).    Now, it would
>> likely make sense to have such a mechanism, but it would need design.
>> I forget offhand too how the PackageKit updater works in the live
>> image (do we disable it by default?).
> 
> With the Anaconda live installer, it's the raw unmodified live
> filesystem that is "copied" to the newly partitioned disk(s).  Ergo any
> rpm updating you do on the LiveOS before you start the installer will be
> lost.
> 
> It appears that Doglas' installer uses the in use LiveOS so that the
> changes you make while running the LiveOS before starting the install
> will be carried over into the install.

Douglas, JDog, whatever...  But correct characterization, with the added 
'changes you make while running the LiveOS before, during, and after 
starting the install will be carried over into the install'.

> 
> That in itself seems interesting enough to explore and see if we can't
> get the same functionality, to some extent into Anaconda.

Yes, this is a subtle distinction you have made.  An entirely different 
feature, could be to simply run the current LiveOS installer as is, 
except instead of using the pristine (i.e. pre-boot, no changes from the 
running LiveOS) image, you could use a snapshot of the image as it is at 
the time of the start of installation.  That would go back to your 
wording above instead of my correction, i.e. only changes made prior to 
the start of installation would get copied over, _and_ you would then 
need to reboot into the new OS as is currently normal.


> 
> As for rebootless, that's interesting too.  I'm glad you were able to
> demo the technology within your installer, but I'd worry about the
> duplication of effort/code to have yet another program that is designed
> to discover disks,...  There is a lot of logic code in
> Anaconda to get the partitioning correct ....

This is all I believe 100% in line with what I said in the 
progenitor/top post.  If this feature is to remain in Fedora in the 
longterm, Anaconda is absolutely the right place for it, for all the 
reasons you mentioned.

However, even though over a year ago I had skimmed all the relevant 
parts of anaconda, the reason I chose to go the route I did were

a) lack of complexity.  The first iteration of my installer was just a 
100 line bash script.  Thus narrowing my task to just writing a GUI 
front end for it, simplified things.

b) educational value- This was a great, relatively simple, pygtk-glade 
exercise for me.

c) the whole unionfs issue.  There is *very real possibility* that 
Fedora may decide to go the ubuntu LiveOS architecture once a solid 
unionfs is in mainline.  Depending on implementation- if going with the 
ubuntu native squashfs for instance, it may then be impossible to have 
RebootlessInstallation as an additional optional feature on the same 
LiveOS.

I will be more than happy making or helping to make the anaconda 
implementation of this feature, if I can be assured that it will have 
some longevity.  I really think going with zyx-liveinstaller (my thing) 
for F12 is the right answer to put this feature out there before people, 
to determine if it is something that we want to really start doing some 
serious additions to anaconda for.

Perhaps jkatz or others can and don't mind just whipping up an anaconda 
implementation in the next few weeks.  But I didn't see much(any) 
interest on fedora-livecd-list, and I'm making a wild guess here- that 
everybody (especially those getting paychecks) have plenty on their 
plate in the near future already.

> 
> I think your work is great and useful though!

That means a lot.  Back at you 10X, I know you and others here have done 
vast amounts of very hard, often unrewarding work keeping Fedora in shape.

peace...

-dmc