The default environment is supposed to be set by the installclass, but if you look at anaconda.log you'll notice some lines along the lines of:

WARN anaconda: module import of fedora-server failed: <class 'ImportError'>

because you can't have a hyphen in a module name in python. Can this be renamed to fedora_server.py?

I can fix that, but any idea why it suddenly started happening on Rawhide and is fine on Fedora 23 and 24?

Oh whoops, never mind on the name. Part of the reason for all the weird import stuff in installclasses.py is to handle names like that.

The reason it stopped working just recently is because blivet-2.0 hit rawhide just recently. The problem is not the name, but the imports.

dshea@spectre:~/fedpkg/fedora-productimg-server$ python3 fedora-server.py 
Traceback (most recent call last):
  File "fedora-server.py", line 10, in <module>
    from blivet.autopart import swapSuggestion
ImportError: cannot import name 'swapSuggestion'

It's swap_suggestion now.

Thanks for the help, David. Updated package is built in Rawhide now that should fix this.

Well, since that fedora-productimg-server went out, Rawhide Server installs have failed to boot, showing a tmux error:

https://openqa.fedoraproject.org/tests/14221/modules/_boot_to_anaconda/steps/8

all other flavors don't hit that, so I'm gonna say you broke it. :)

(In reply to Adam Williamson from comment #5)
> Well, since that fedora-productimg-server went out, Rawhide Server installs
> have failed to boot, showing a tmux error:
> 
> https://openqa.fedoraproject.org/tests/14221/modules/_boot_to_anaconda/steps/
> 8
> 
> all other flavors don't hit that, so I'm gonna say you broke it. :)

That doesn't make any sense... we pushed the same change to the Atomic installer, so that should have broken too, if this was the problem.

I'll look into it...

Now you mention it, Atomic does in fact fail the same way:

https://openqa.fedoraproject.org/tests/14367
https://openqa.fedoraproject.org/tests/14367/modules/_boot_to_anaconda/steps/7

but Everything boot and Workstation boot do not.

When I finally got into the VM, I tried running anaconda directly from the terminal (non-X). I got this crash:


Traceback (most recent call last):
  File "/sbin/anaconda", line 1099, in <module>
    setupDisplay(anaconda, opts, addon_paths)
  File "/sbin/anaconda", line 580, in setupDisplay
    anaconda.initInterface(addons)
  File "/usr/lib64/python3.5/site-packages/pyanaconda/anaconda.py", line 240, in initInterface
    self._intf = TextUserInterface(self.storage, self.payload,
  File "/usr/lib64/python3.5/site-packages/pyanaconda/anaconda.py", line 176, in storage
    self._storage.setDefaultFSType(self.instClass.defaultFS)
AttributeError: 'Blivet' object has no attribute 'setDefaultFSType'



Looks like something was missed when converting to Blivet 2.0.

(In reply to Stephen Gallagher from comment #8)
> When I finally got into the VM, I tried running anaconda directly from the
> terminal (non-X). I got this crash:
> 
> 
> Traceback (most recent call last):
>   File "/sbin/anaconda", line 1099, in <module>
>     setupDisplay(anaconda, opts, addon_paths)
>   File "/sbin/anaconda", line 580, in setupDisplay
>     anaconda.initInterface(addons)
>   File "/usr/lib64/python3.5/site-packages/pyanaconda/anaconda.py", line
> 240, in initInterface
>     self._intf = TextUserInterface(self.storage, self.payload,
>   File "/usr/lib64/python3.5/site-packages/pyanaconda/anaconda.py", line
> 176, in storage
>     self._storage.setDefaultFSType(self.instClass.defaultFS)
> AttributeError: 'Blivet' object has no attribute 'setDefaultFSType'
> 
> 
> 
> Looks like something was missed when converting to Blivet 2.0.

After talking with David Lehman, we realized that this was *indirectly* caused by fedora-productimg-server; We do set a defaultFS that differs from upstream, so this code-path is executed and it was missed in the Blivet 2.0 conversion.

Reassigning to anaconda to sort out.

I'm not sure what's going on with Atomic, since it doesn't follow this code-path. Please open a separate bug to track that one.

At this point we probably ought to create a new bug, since this is nothing at all to do with the original bug here. I'll close this and open a new one. Leaving this as ON_QA so we can test the fix for the *original* problem once the new problem is fixed.

For Atomic (rawhide), https://github.com/projectatomic/fedora-productimg-atomic should be updated for Blivet 2.0. I think.

(In reply to Radek Vykydal from comment #11)
> For Atomic (rawhide),
> https://github.com/projectatomic/fedora-productimg-atomic should be updated
> for Blivet 2.0. I think.

I already built http://koji.fedoraproject.org/koji/buildinfo?buildID=756654 for fedora-productimg-atomic. This was tracked in BZ #1329288

For the record the Atomic problem here is one we already knew about, https://bugzilla.redhat.com/show_bug.cgi?id=1327460 . The factor that briefly confused us is that "tmux.conf:20: no current client" seems to be the new "pane is dead" for tmux 2.2 - we now we see that message instead of "pane is dead" when anaconda crashes early. But there were two completely different *causes* of the early crashes. So now we have this bug for the blivet port bug that Server hits, we have 1327460 for the Atomic bug, and I also filed https://bugzilla.redhat.com/show_bug.cgi?id=1330300 for the fact that with tmux 2.2, we don't see the anaconda crash output on the screen as we did before.

the original problem here is long fixed.