Installer should verify installability

[jdreed]

Currently, it is possible to end up with a b0rked machine (say, because the openafs modules aren't available for a new kernel version) via the PXE installer. This is bad. Thankfully, the difference is noticeable enough that it's an inconvenience as opposed to a major problem (ie: the greeter is missing, the ability to log in is missing, etc).

However, users who attempt to install Debathena should either end up with Debathena or with a message saying "Sorry, try again later."

We should make the installer more foolproof.

[jdreed]

At release-team, we talked about having the auto-debathena-whatever-ifier push openafs-modules directly into production.

But I'd still like the installer to do a bit more sanity checking.

[jdreed]

So, this is actually two issues:

• ensuring there's enough disk space
• ensuring critical modules (like openafs) exist in the repo.

The former is less of an issue, but is apparently a trivial fix according to amb. THe latter is much harder.

[jdreed]

Here's a terrible idea:

As was the case with Athena 9, the installer installs from a different repository, which is known good and updated periodically from production.

[andersk]

If production isn’t known good, we’re doing something wrong.

The openafs problem will go away with DKMS (#243).

[broder]

Replying to andersk:

If production isn’t known good, we’re doing something wrong.

I wish I shared your confidence in my perfection.

Yes, we shouldn't break things. But it doesn't mean that we don't. Or haven't. The fact that we shouldn't break things doesn't mean we shouldn't code protections for when we do.

[jdreed]

dkms "fixes" this.