rawhide install report 2007-12-20

[Jeremy Katz]

On Fri, 2007-12-21 at 06:37 -0500, G.Wolfe Woodbury wrote:
> 2) The install ultimately failed due to a missing piece in the images
>    (a missing cracklib dictionary  BZ 426444)

Fixed this up, thanks.  I hadn't noticed as the vast majority of the
install testing I've done (especially this week) is from the livecd.

> 3) The installer wastes a significant amount of resources to resize a logical
>    volume despite the fact that the "resized" lv is marked for reformatting
>    anyway. [BZ 426466]

Yep, on my list to be fixed... thanks for filing it so it doesn't get
lost, though.

>    The new horizontal bouncer seems to eat a lot of CPU time during the unneeded
>    fsck that took place too.  It may be "too cute for words."

The problem is that we really do want to provide some feedback here, but
parsing code for every filesystem checker, creator and resizer's output
isn't ideal.  We'll probably end up tweaking so that it takes a little
bit more output to make it "move"

> 4) The installer does not "automagically" enable the network adapter it used
>    to load anaconda from a NFS repo anymore.  You have to manually check on
>    the activate-on-boot box. [BZ 426473]

Yeah, this was a weirdo one that I fixed up a day or two ago.

> 5) It is ridiculous to immediately commit partitioning and format filesystems
>    rather than to defer the formatting much later in the install. [BZ 426474]
> 
>    The user has no time to reconsider partitioning on the basis of software
>    selections, and subsequent errors (e.g. BZ426444) or dependency hell leave
>    the target system in an unuseable state.

This wasn't a change that was made lightly, but it's a change that we
made because it's somewhat necessary for some of the things we're
working to enable.
  * If you do resizing, you really want to do it ASAP as resizing is a
much less predictable operation than just creating partitions and
filesystems.  So you want to enable the ability to fallback to changing
partitioning again.
  * The code paths for "I need swap early" vs "I can wait until the end
for partitions to be committed" were often divergent and not tested
leading to weird errors on the first case that didn't get tested because
all of our machines have lots of ram :)
  * Enabling the ability to add new CD-based repos to be installed from
at install time will require that we have a place to stick the installer
image so you can swap CDs.  Choices for that are basically RAM or disk.
And as we're talking about a file that's over 100 megs, RAM is pretty
unattractive.

That said, there are still a few of the pieces we need to finish up to
help offset the increased "risk".  One is that we're going to move
around some of the rest of the steps.  The sequence is going to be more
"configuration types of things", "partitioning", "package selection",
"install" as opposed to "configure", "partition", "configure", "package
selection", "install".  Which should help to avoid some of the perceived
problems. 

It's a big change to be sure and it does have some downsides, but it's
also a pretty necessary change

> The Anaconda Team deserves kudos for the rapid pace of recent improvements and
> the updated look and feel, but some significant holes have been left exposed.

Don't worry, we haven't stopped yet for the release :-)   First step is
getting all of the big changes landed, as much as we can by the alpha
next month and everything by the feature freeze.  And then comes going
through and ensuring everything is nicely polished.

Jeremy