Expert partitioner gone wild in 12.2 RC1

Hi everyone,

I posted on this topic earlier but in French and, apparently, that wasn’t a good choice of language or them dudes are all on vacation. So here goes the same in English, in the hope to reach a larger audience. This post is meant to serve two purposes:

1/ to inform whoever’s maintaining the partitioner about a possible severe bug and a couple design issues,
2/ to warn the user community at large about the same and suggest a course of action to protect data.

A few days ago I tried installing openSUSE 12.2 RC1 on a brand new PC, UEFI-aware BIOS, and a new 500GB system disk purchased just for the occasion. I also have 5 other internal disks, totalling 12TB and containing 5TB worth of precious data, which I naturally installed in the box. Due to various incompatibilities between BIOS settings and bootloaders, I’ve had to restart that installation 4 or 5 times and what I’m about to describe only happened during the next-to-last installation.

The installation starts correctly and eventually invokes the partitioner. I select custom partitioning for experts, as I’ve always done since installing openSUSE 9.3 and all its successors for the past ten or so years. To make certain I’m about to change the partition table for the correct device, I go through the list of devices one by one, making sure sda through sdd have been attributed to disks labelled video1 through video4, sde attributed to disk labelled audio, and finally sdf attributed to my new 500GB system disk. During that checking procedure I also expand the list of partitions present on each disk and make sure they contain what they are supposed to contain, i.e. /video1 on device video1 and so on. Everything was ok.

I then proceed to propose a new partition table (GPT) and partition setup for that system disk: 30GB for /, 16GB for swap, 200GB for /home and the remainder, some 210GB, for a partition named /backup which is in fact more of a data cellar. I accept the changes and get back to the main partitioner window. Just before moving on with the rest of the installation, I again go through the list of devices (better safe than sorry) and that’s when I saw it: the partitioner had replaced partition /video1 on device video1 seen as sda with what looked like a default partition proposal. From memory it had 3 partitions: 10 or 20GB for /, 2 or 4GB (not sure) for swap and the rest of that 2TB disk for /home. /video1 had disappeared!

Naturally when faced with such unexpected behaviour, my first reflex was to press on the on/off switch, open the box, immediately unplug any disk containing data and not directly useful for the installation of openSUSE. Which is why I can’t provide screenshots or more accurate information as to those default-looking partition sizes and I’m definitely NOT going to try this heart-stopping experiment to help anyone reproduce that incident.

I then proceeded to re-install openSUSE from scratch on that single system disk, repartitioning from scratch, new partition table, etc, all other disks being unplugged during the operation. After reboot and proper startup of Suzie, I turned the whole thing off, reconnected my 5 data disks and rebooted. O surprise, disk video1 still contained partition /video1 with all its contents. It wasn’t really a surprise, since I hadn’t started any irreversible action on those disks but the period was tense nonetheless.

My advice to the user community at large: I’ve read from another user on this forum that he disconnects all his disks except the intended system disk during installations. Many others were making fun of him. He was absolutely correct though, I can tell you that. My only mistake was to have read his advice, thinking it made a lot of sense, and not applied it 2 days later during that chain of installations. Having said that, had I done what he suggests… I would never have seen what appears to be a nasty bug. :slight_smile:

To the maintainers and would-be re-designers of the partitioner:

1/ The application knows when a disk contains a GPT or MS-DOS partition type, for the simple reason that it offers a choice between the two options when expert-repartitioning a disk. Why doesn’t it show on the interface? Why is the only available method for a normal user to expert-repartition a disk in order to find out which type of partition table it contains? Why??? Ok, I think I know why, so my question should really be: when??? Before or after Suzie is made obsolete by the combined efforts of MicroCrap, mainboard and disk manufacturers?

As a side-note on this topic, I found out that the “disks” application (in system tools) enables users to see a cryptic partition type expressed in hexadecimal at first, just when a disk has been re-partitioned, which later changes to a non-descript “Basic data” once you reboot your system. I kinda s’pose that an informed guess would be that “Basic data” means GPT partition type but… ahem… users are best left in darkness about such things, right? Then again Suzie might lose her charms if her developers started calling a cat a cat.

2/ I’m speaking here as software designer, not necessarily Linux user. The main job of a partitioner is to change the partitioning scheme on connected disks. Yes, one may argue about the benefits of RAID, LVM, or include them in this conversation if one wishes, but within the context of using Suzie at home with 2 to 6 internal disks, most of us merely want the partitioner to be able to perform that “simplest” of tasks.

So we’re facing a state-shift situation, the famous before-state and after-state. Any 3-year old, provided his daddy’s named Linus, knows about before/after states. They’re more readily understandable when both faces of the situation are presented to the user and, what’s best, a dual view allows to spot at a glance those rare cases when the partitioner imposes a setup by default on a disk about which you had asked absolutely nothing in terms of changes.

So my input on the matter is: why not show these two states side by side on the interface, at least for the list of devices and the partitions they contain? Before on the left and after on the right, or above/below if you wish to cater to the Chinese market, with some sort of colour-coding such as red/green (change/no change on device). Not being able to see the previous setup of a disk, particularly when the partitioner decides of its own accord (in expert mode, I repeat) to implant a default setup where you had no intention to see one, is not really user-friendly. Could greatly be improved.

Usual disclaimer: I’ve been designing/writing/maintaining/debugging software for the last 30 years, do not drink when I touch a keyboard, always think twice before hitting a key or clicking on a mouse, so no… I didn’t dream this. Trust me, it just happened.

Hello again,
Maybe that’s the problem! lol!

Anyway, as you said “Let’s call a cat a cat” (or for english native speakers, I guess, “Let’s call a spade a spade”) and first move this thread to the Pre-Release/Beta subforum, where it actually belongs.

On 2012-07-23 02:06, mister dbk wrote:
>
> Hi everyone,
>
> I posted on this topic earlier but in French and, apparently, that
> wasn’t a good choice of language or them dudes are all on vacation. So
> here goes the same in English, in the hope to reach a larger audience.
> This post is meant to serve two purposes:
>
> 1/ to inform whoever’s maintaining the partitioner about a possible
> severe bug and a couple design issues,

Wrong place :slight_smile:

For two reasons. 1), there is a specialized forum on beta versions, so this post is off-topic
here. 2) Even if you post in the beta forum, you will not reach the maintainers, because they
do not use the forum, but the mail list.

And even then, the officiall method for communicating bugs is a Bugzilla :slight_smile:

What do they call that, red tape? :-p

> I then proceed to propose a new partition table (GPT) and partition
> setup for that system disk: 30GB for /, 16GB for swap, 200GB for /home

As that disk is only 500 GB, why use GPT? Support is still new, and it is not needed, you can
use traditional partitioning.

> and the remainder, some 210GB, for a partition named /backup which is in
> fact more of a data cellar. I accept the changes and get back to the
> main partitioner window. Just before moving on with the rest of the
> installation, I again go through the list of devices (better safe than
> sorry) and that’s when I saw it: the partitioner had replaced partition
> /video1 on device video1 seen as sda with what looked like a default
> partition proposal. From memory it had 3 partitions: 10 or 20GB for /, 2
> or 4GB (not sure) for swap and the rest of that 2TB disk for /home.
> /video1 had disappeared!

Wow. :frowning:

>
> Naturally when faced with such unexpected behaviour, my first reflex
> was to press on the on/off switch, open the box, immediately unplug any

No, you can simply tell the installer to do a fresh proposal or to delete the proposal. At
worst, halt, do not pull the plug.

> disk containing data and not directly useful for the installation of
> openSUSE. Which is why I can’t provide screenshots or more accurate
> information as to those default-looking partition sizes and I’m
> definitely NOT going to try this heart-stopping experiment to help
> anyone reproduce that incident.

You could simply have stopped there (nothing is written to the hard disk till after the red
window message box) and saved the logs to an external disk, for the report. After all, it is a
beta version and you are expected to do reports :wink:

> I then proceeded to re-install openSUSE from scratch on that single
> system disk, repartitioning from scratch, new partition table, etc, all
> other disks being unplugged during the operation. After reboot and
> proper startup of Suzie, I turned the whole thing off, reconnected my 5
> data disks and rebooted. O surprise, disk video1 still contained
> partition /video1 with all its contents. It wasn’t really a surprise,
> since I hadn’t started any irreversible action on those disks but the
> period was tense nonetheless.

No surprise at all to me. At that point changes to the disk as far off in the future of the
process.

> My advice to the user community at large: I’ve read from another user
> on this forum that he disconnects all his disks except the intended
> system disk during installations.

That procedure also has problems: you must make sure that disk is the boot disk in BIOS
parlance when you plug in the rest of the disks, or you are in trouble.


Cheers / Saludos,

Carlos E. R.
(from 11.4 x86_64 “Celadon” at Telcontar)

Now, you see, when you get the same answer in french and in english, it’s most likely to be the right answer (or at least the best answer we can provide). But it’s OK to continue in english. You’ll reach more readers for sure.

+1, Carlos
Same answer again. … and wrong advice of yours to the user community… although there are always exceptions and special cases.

> My only mistake was

thanks for taking the time to provide feedback, but:

-i never trust any partitioner with a single email, much less “5TB
worth of precious data”…that is i never expose any important data to
any partitioner…and to do so with pre-release/beta software is just
not even a reasonable idea! (30 years of experience? really? i just
barely have 20 years with personal computers and i’ve not trusted
testing software, much less testing-software-partitioners for over 19!!)

-but of course we all know you must have had a absolutely safe copy of
that data in at least one off-site location…and certainly you don’t
consider any “backup” kept in the same machine, or in the same
dwelling/structure as ‘safe’…we all know that no one with 30 years of
experience would make such a rookie mistake as that.

-so: yes, i disconnect drives when installing a new one–it is just
standard practice…and, anyone who laughs at that bit of caution
deserves my ignoring their lack of experience and their folly.

-agree with the others: wrong forum

-agree with the others: wrong place to address developers

-agree with the others: proceed to bugzilla and/or factory mail list

how could anyone be a SUSE user since 9.3 and not know those three
(four, with the mail list pointer) things?

incomprehensible, to me…of course, you are not the first 30 year
software veteran (and designer) who has come here having used SUSE
software since 9.x, who has learned so very little in all those years!

so, not listening to the wise poster on the French list who advised you
to unplug was not your “only mistake”…not even your only rookie mistake.

and btw, who is Suzie? and how did she get into into this??


dd

I’ve used the expert partitioner in 12.2 rc1 and had no trouble with it.
And the Proposal was everything I proposed, so I was able to proceed.
And the final Summaryconfirmed everything was OK to proceed.

On 07/23/2012 05:46 AM, caf4926 wrote:
> no trouble

somehow i am not surprised!

and, suppose if a bugzilla is logged it will be closed eventually with
“could not duplicate”

which is not to say it didn’t happen…lets see how many add a ‘me too’
to the bug…


dd

I’m not sure I understand what you mean. But trust me - sorry if I repeat what I already said in the french thread - this has again nothing to do with the partitionner, neither with openSUSE. A couple things have to become clear to everyone and are not IMO: you have several hard disks, connected to 2 SATA controllers (I think it is clear why sda became sdf and vice versa - as I explained in the french forum). We don’t know if these disks were partitioned with MBR or GUID. Apparently you ended up with GPT on your boot disk and you now have a UEFI booting Linux. Is that right? In this case you should have a EFI partition. Until 12.1, openSUSE used to format this partition in FAT16, which was clearly a bug. I hope - and assume - they fixed that one. It worked but surprisingly - for a change - it broke Windows standard. It also used to replace the protective MBR with an hybrid MBR, another kind of absurd bug. As you see, if we’re looking for bugs, we will always find some. Now if you don’t mind posting the output of fdisk -l, it will tell us more about your hard disks than any further explanation.

As for partition types, the MBR partition table use one byte, most often (but not always) shown in Hex, to identify the type of file system. (Too) many Linux file systems use 0x83 as their partition ID. This is IMO unfortunate. The GUID partition table use 128-bit values and the GUID (Globally Unique identifier) - not to be confused with the file system UUID! - is intended to be unique, as the name indicates, but is not! Please read this post http://forums.opensuse.org/english/get-technical-help-here/install-boot-login/475347-error-occurred-while-installing-grub-during-os-11-4-installation-error-25-asus-efi-bios-5.html#post2465994 with the example I posted there and have a look at this email by Rod Smith, the author of GPT fdisk: Need for a unique Linux GPT GUID type code (PATCH included). Does it describe the partition that was referred as “Basic-data” in your setup?

This thread is CLOSED fort he moment. It will be moveed to Pre-Release/Beta.

Moved, open for Business.