lethal adventures in f18 alpha

[cornel panceac]

2012/9/15 Adam Williamson <awilliam at redhat.com>

> On 2012-09-14 22:29, cornel panceac wrote:
>
>> being tired of the dracut selinux infinite loop, i've decided this
>> morning to reinstall f18 from f18 x86 netinst disc. all attempts to
>> boot with standard options failed with a kernel crash (maybe my video
>> card is broken). i eventually managed to reach anaconda by booting in
>> safe graphics mode. there, i've completed the steps required (what is
>> the target device, what software, time zone) then pressed continue.
>> then, i've been hit by this:
>> https://bugzilla.redhat.com/**show_bug.cgi?id=857607<https://bugzilla.redhat.com/show_bug.cgi?id=857607>[1] . I've seen
>>
>> this before on same system in the previous 5 (?) years so that was not
>> a big surprise. a big surprise was that after i reboot, something
>> messed my existing linux partitions (actually one of them is missing).
>> what do i mean by 'messed'? well, first, grub2 fails to load a menu
>> saying that "no such partition" and offers a grub shell. then, in f17
>> rescue mode, the system complains that i have no linux partitions.
>> going to rescue mode's shell, the partitions are marked "Linux", but
>> the last one (#7) is missing. then i booted system rescue cd. there,
>> mount /dev/sdb{1,5,7} /mnt/mountpoint ended with something similar to
>> "broken ntfs signature". adding -t ext3 in the mix didn't help, error
>> was something like "bad superblock etc".
>>  THIS is something i don't remember to have ever seen on fedora
>> (alpha, beta or final).
>>
>
> You need to say what layout you had on the disk(s) previously, and what
> options you chose during install. It's impossible to know what happened.


well the layout was something like this:

p1 ->10gb -> scientific linux
p2 -> 0.5 gb  -> boot
p3 -> 1gb -> swap

p5 -> 140 gb -> /home -- that's the partition anaconda complained about and
errored out. also that's the partition i care more about :)

p6 -> 10 gb -> f18
p7 -> 70 gb -> / -- this is no longer present

the size are aproximated. the partitions that apparently are still there,
can not be mounted


> It's possible you simply hit this:
>
> https://bugzilla.redhat.com/**show_bug.cgi?id=855976<https://bugzilla.redhat.com/show_bug.cgi?id=855976>


looks similar but, i've never seen something like choosing what kind of
partitioning to use (custom, free space, replace linux ). it just crashed
and data was no longer there or accessible.

>
>
> Remember, this is an *Alpha*. You shouldn't deploy an Alpha of anything to
> any system you can't bear to lose. We do warn about this, regularly. You're
> also installing the Alpha before it's even been announced; I'm aiming to
> ensure this issue and other major ones in Alpha are explained in the
> release announcement when we release it, but we haven't, yet.


well, i always fail to mention this: one of the reason i report this kind
of things is because i want you to know about their existence. another
reson is that i want to learn why this happened and how can be avoided. in
this case, i also very much want to recover the lost data :D

>
>
>  from personal experience, i believe something is deeply wrong in the
>> anaconda logic. and of course, the back and forth mess is also a
>> problem. why not just continue to the next required step, with back
>> being an option, and at the end present a summary of completed and
>> incompleted steps, if any?
>>
>
> That isn't the design. There aren't steps that you go through one by one.
> The design is referred to as 'hub and spoke'. There is a main screen - the
> hub - and several spokes, some of which are optional, and which you can
> complete in any order, and which you can revisit any number of times, and
> which may potentially affect each other. The hub/spoke design was
> considered a better reflection of the capabilities of anaconda than the
> wizard-ish design.
>
>
>  btw, on systems with many hard disks (at least 2), if linux/anaconda
>> is unable to tell which is the first hard disk system will boot, why
>> not install grub on all hard disk, just to be sure?
>>
>
> That's a horrible idea...
>
> Bootloader installation selection is something else which isn't entirely
> implemented in newUI yet. But we certainly don't want to go around
> automatically squelching whatever's in the MBR of every single disk on the
> system. That certainly would be bad behaviour. And 'lethal'.
>

i do not really understand why are you sayng that. i mean, since we already
do not know if we are installing grub to the right drive, what's worst in
doing it two times in a row?  (or three, etc)

after all, this can be enabled by user's choice, who should know better if
there's is some special boot loader in mbr somewhere ...

otoh, this is the smallest of my problems, now :)
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.fedoraproject.org/pipermail/test/attachments/20120915/22a0671f/attachment.html>