Debian Upgrade May Cause Serious Breakage

daria42 writes "Debian developer Bill Allombert has e-mailed the Debian community saying he estimates about 30% of users upgrading from Debian Woody to Sarge will suffer 'serious breakage'. Allombert says the upgrade process suffers from a number of bugs reported before the release went live several days ago. Chief among the problems, he said, were cyclic dependencies and the fact that software installation tool apt depended heavily on the changing C++ libraries. Allombert wants developers to test the upgrade cycle continuously during development and not just during the freeze period just before release."

Comments

Evidence of problems with packaging systems

[AKAImBatman]

Chief among the problems, he said, were cyclic dependencies and the fact that software installation tool apt depended heavily on the changing C++ libraries.

Let this be a lesson to those of you who claimed that "APT is unbreakable." There's no such thing as an unbreakable technology. There is however, such a thing as a robust technology that resists failure. As packaging systems go, APT is fairly good. However, my belief is that packaging systems are inherently flawed.

What you want in an OS, is a method for determining the precise core upon which you can base your applications on. Such a core would effectively be an immutable set of system APIs that cannot be changed. The upshot to this situation is that the given system is verifyable. i.e. I can have a script go through and ensure that everything that should exist does exist. From that information, I can then do a delta to find out what exists that shouldn't exist.

This is in direct opposition to a packaging system that builds an OS out of inter-dependent components. The problem with such a strategy is that using inter-dependent components only works if you're building from scratch. As anyone who has managed a version control system can tell you, things get extremely complicated (and tend to require manual intervention) as soon as files start branching. The same thing happens in packaging systems as soon as you start doing upgrades to individual components. Soon you find yourself with a mess of mismatched dependencies which require constant manual intervention to solve. Not a good situation.

In the case of a defined core, you can simply wipe out the old core and replace it with the new one. As long as testing has been done to ensure that the new components are still backward compatible with old software, everything should work fine after the upgrade.

Food for thought, anyway. To the Debian team: Thanks for the new release! Even if there are some growing pains, it's still nice to see you back in the game. :-)

Re:Evidence of problems with packaging systems

[listen]

You again ;-)

Take a look at this Conary system. It has some interesting ideas that could certainly help in this kind of situation : especially transactions for upgrades. If a bit fails, the whole upgrade rolls back, and you can even rollback completed transactions.

I like this idea better than choosing some arbitrary core of code to upgrade as a massive lump, and statically linking hundreds of copies of anything not in the core into the separate apps. As to your verifiability detecting script, I see no reason this can not be done for a packaging system. And before you go on about corrupt databases, please remind yourself what a filesystem is: thats right, a corruptable database.

I will agree with you on compatibility: people should stop breaking ABI. I'm looking at you, Freetype...

I was waiting for Sarge but then came Ubuntu.

[Anonymous Coward]

Any reason why I should switch from Ubuntu to Debian?

This happens all the time

[Anonymous Coward]

in systems where politics begin to come before purpose.

It's not FreeBSD that's dying, but Debian.

Update took me two days ... grrr

[TekGoNos]

Nice to know I'm not alone.

Suddenly apt-get dist-upgrade didnt do anything good, I had to do an apt-get -f install multiple times until the dependancy stuff was sorted out. In the process, some packages (notably apache and ftpd) were simple de-installed and I had to re-select them manually.

Good for me that it was a server and apache and ftpd were the only important hand-selected packages. I fear for the desktop systems with several dozends of hand-selected packages.

So, I guess it is a good thing that Debian only releases a major update every two years :|

Re:Mixing Lilo and some kernel configs not nice ei

[Qzukk]

He's using SATA and the newer linux kernels moved SATA from IDE to the SCSI subsystem.

So all his fstab entries using /dev/hd* and his lilo root=/dev/hd* lines became wrong.

This is true.

[cdn-programmer]

I attempted an upgrade from woody to sarge about a month ago and it broke my system. I have 1000's of zombies running around. This shows up as a defunct process. Its not the end of the world mind you but you can't kill a zombie since it is already dead.

I have reported this and warned that there will be a lot of folks with broken systems. I was very surprised to hear that sarge went stable before this problem was sorted out.

A sarge install from scratch however is fine. Its just the upgrade that is broken and in more than one place.

Re:Cyclic Dependencies ?

[jdunn14]

I usually dont reply to ACs, but come on. Of course user has different meanings. People running protein folding simulations on super computers are users. So is grandma who needs a machine that automatically opens her email. As for the claim that open-source patches are better, they are, for those high end users who care to patch things themselves. For the rest of us, a distro will get the patched version of their binaries out to end users fast enough.

As for the playpen comment. Screw you, lots of people use linux for serious work, both desktop and commonly in the server room. Those supercomputers? not running windows. many running linux. Its getting to the point that you don't have to be an expert to run the system, yes its taken a while, but when it's made by the experts generally for the experts usability is not necessarily the top priority.

Yes I play on my linux box, try the newest stuff, etc. I also write web pages (rarely), do all my correspondence, write software for my job, and play media and games on linux. Basically I do the same things I would do on windows, and I find it less annoying to work with. This is mostly due to the fact that knowing enough about the internals means that I can swear at MS for not providing the expert level interface to go with their wizards. Sometimes a command line is more efficient (Ms may be getting the message if they really are revamping their CLI).

don't upgrade in place if you can help it

[cahiha]

If you really care about a system and minimal downtime due to upgrades, have two root partitions (it's only an extra 5-10G). Instead of upgrading, you make a clean install on the unused root partition. Clean installs generally work better to begin with, and you have the old install both mounted and bootable to figure out any problems and copy over configuration files.

As for complaints about this sort of thing, I still prefer Debian. I just spent several hours upgrading an OS X system from Jaguar to Tiger. A trivial file system inconsistency in HFS caused the installer to crash reproducibly and eventually required me to manually patch inodes (apparently a fairly common problem on Macintosh). And I'll have to wipe a Windows machine clean tomorrow because mysteriously hardware has stopped working and no amount of fiddling will do.

In comparison, these Debian upgrade woes seem minor. And unlike the Mac and Windows problems, the Debian upgrade problems will generally fix themselves after a few days when the package maintainers catch up.