[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

When stability is pointless

With apologies for cross-posting.

Dear all,

I have copied below the text of a blog post* I wrote a few minutes
ago, because it addresses an issue in Debian and Debian-derived
distros that I've encountered several times, and which no doubt many
people encounter frequently. It's an issue which seems, to judge by
forum posts and mailing list requests, to generate a fair amount of
traffic and expenditure of effort, but which has not really been
tackled in a co-ordinated way as far as I am aware.

Consequently, I've written this piece (which is intended for a general
audience) in order to raise awareness of the issue, and to seek a
co-ordinated effort to tackle it.

Yours respectfully,

Sam Kuper

*URL: http://www.sampablokuper.com/blog/2008/11/05/when-stability-is-pointless/

When stability is pointless

Many Linux distributions (and other software environments too) use
package managers to facilitate the installation, upgrading and
uninstallation of software packages as needed. At least, that's the

Why have package managers?

Are package managers necessary? Well, no. One way of managing software
is simply to install individual software programs/libraries as needed,
and allow each item to handle its own updating or uninstallation (or
even just leave that to the user to do manually). That's pretty much
how Windows handles things. It works OK if it isn't crucial that your
programs aren't able to communicate with each other beyond basic,
operating system level mechanisms like cut-and-paste. If your programs
depend on each other, however, you'd be in trouble if you removed a
piece of software that another piece depended on, or installed a piece
without installing its dependencies, etc.

Linux distributions usually have many pieces of software that are
inter-dependent. A package manager can keep track of those
dependencies and can, for instance, inform a user about to uninstall a
piece of software of which other packages will be affected by this.
That's important, because otherwise the user might accidentally
disable something crucial.
Package managers can also make upgrading software, to take onboard the
latest security patches, trivial, requiring only one or two commands
in order to automatically upgrade all the packages on an entire system
for which upgrades are available.

I haven't investigated the historical origins of package management,
so I can't claim to know the original reason why package managers came
to exist. But I can state from experience that package managers, when
they work well, make life easier than it would otherwise be for system
administrators running Linux systems with interdependent packages.


Package managers facilitate stability. Given the foregoing, this
should come as no surprise. Yet in the context of many Linux
distributions, stability means something more specific than the
colloquial "reliable, consistent" sort of meaning that the term
normally implies. Stability, in the context of these distributions,
means "unchanging, except with regard to security". One such
distribution is Debian.

Through the magic of package managers, Debian maintainers maintain
multiple packages of each original software item they want to make
available to users. The reason for the multiple packages is normally
to achieve stability. The stability is achieved by packaging a version
of the original software that has been tested sufficiently to warrant
confidence in its security: it is not known to compromise systems on
which it runs (except in cases where it may do so by explicit design).
This "stable" package is then offered to users for, typcially, many
years, and is not altered in any way except if a security flaw is
discovered in it. Only if a security is found will the package be
modified: to patch the flaw.

Meanwhile, the maintainer will keep an eye on new versions of the
original software. When the maintainer thinks a new version is worth
packaging, she may package it as an "unstable" package to begin with,
and then subsequently a "testing" package. If, as the release of a new
version of Debian is approaching, the "testing" package meets
sufficient criteria indicating its suitability for being regarded as
"stable", then it, too, is marked as "stable" for that new release of
Debian, and is thereafter treated as described above.

As such, Debian may have multiple packages for the same piece of
software: "stable" packages for current and old versions of Debian,
and "unstable" and/or "testing" packages. Usually, each of these
packages will be based upon a different version of the original
software. In a hypothetical case, the "stable" package for the current
Debian stable release might contain version 3.1 of the original
software (perhaps with security patches, as described above); the
previous Debian stable release might have packaged version 1.0 of the
original software. The current "testing" version of Debian might
package version 3.9 of the original software, and the "unstable"
version of Debian might package version 4.2 of the original software.


Sometimes, stability lets you down.

My perception is that the greatest problems with the system of
"stability" practised by Debian and other Linux communities arise when
the upstream developer has not maintained the documentation for
earlier versions of the software he has written. This leads to a
disconnect between users reliant on package managemers and interested
in dependability, and developers interested in making software that is
faster, more fully featured, or otherwise different from the earlier
versions of their software.

An example

Here is my scenario. I have a server running Ubuntu 8.04 LTS: a
"stable", recent release of a Debian-based Linux distribution. I wish
to install a security-related program called "psad" (short for "Port
Scan Attack Detector) on that server. However, the stable package of
psad for Ubuntu 8.04 turns out to house version 2.1 of psad. That
wouldn't bother me, except that… I can't set it up!

The reason I'm having difficulty setting it up is that the
documentation on installing psad refer not to version 2.1 but to
version 2.1.4, which requires setting up differently to 2.1. The
developer's recommendation is that I upgrade to a more recent version,
but two questions arise:

* how?

* why?

The how question has several possible answers. I could forego package
management altogether for psad and install version 2.1.4 straight from
the source code. But that would entail the problems outlined above:
the problems package management solves. Alternatively, I could perhaps
configure my server to attempt to use an "ubstable" or "testing"
Ubuntu psad package in place of the stable one for 8.04.

But why? Why should I have to do this, if the "stable" package is good
enough that it was chosen for "stable" release?
This is where the disconnect I alluded to above becomes apparent.
Developers may not always explicitly deprecate their old code, but
they nearly always do deprecate it anyway. The response of Michael
Rash, the psad developer, was characteristic of this.

Yet as far as users of stable distributions are concerned, that same
sofware - the oldish versions of the program - has not been
deprecated: it's still in the latest "stable" repositories.

There's a contradiction here.

Unfortunately, it's an unresolved contradiction, as far as I can tell.
I encountered it previously with rkhunter, also a piece of security
software I wanted to install onto a software running a "stable"
version of Linux. In that case, it was even worse: the upstream
developer had explicitly deprecated the version that was in the
current (read: un-deprecated) "stable" package, and was entirely
unwilling to support it.


Despite these problems, I'm still very grateful to - and have a lot of
respect for - the developers and maintainers of the software I use
(including the software I've mentioned above). Michael Rash's software
is innovative and free (in both senses of the word), and he was
certainly under no formal obligation to reply to my queries.
Furthermore, I have learned a good deal from reading his book.
Equally, rkhunter *can* be a useful tool even if you aren't using the
very latest version of it, and the maintainers who answered my queries
also did so voluntarily and courteously.
Yet the problem remains: when developers deprecate software that
maintainers have not deprecated, users are left in the lurch with
software they can't use, can't get much support for, or can't find
documetation for.


In some ways, I'm still very much a newcomer to the free software
world. It's only in the last 2-3 years that I've really begun feeding
back to the communities whose tools I use, and I have not become a
developer or maintainer myself. As such, I'm not yet in a position to
implement change to, say, Debian's release policy.
Yet, I do have some ideas for solving the problem I've outlined in this piece:

* Raise awareness of the problem.

* Propose that maintainers open a dialogue with the developers whose
software they package to request that developers continue to support,
or at least maintain documentation for, old versions of their software
until the maintainers have deprecated it.

* Ask the maintainers to shoulder this burden where the developers
cannot or will not do so.

* Seek further suggestions from the community.

In the first instance, I'm going to attempt to make some progress
towards realising these solutions by posting a link to this piece to
the Debian and Ubuntu mailing lists. Then I'll leave it in the
community's hands unless I find I have any more to add.

Reply to: