Not logged in
Log in now
Create an account
Subscribe to LWN
LWN.net Weekly Edition for May 23, 2013
An "enum" for Python 3
An unexpected perf feature
LWN.net Weekly Edition for May 16, 2013
A look at the PyPy 2.0 release
RHEL customers expect a bit more process and stabilisation for their money.
The thing is customers pay for the userspace to remain crusty, as in the whole point of running an enterprise distro is that it isn't Ubuntu or Fedora, it doesn't change all the rules every 6 months.
Though you'll notice with RHEL5 there is already updates for desktop apps a lot more often, it gets latest firefox quite quickly for example.
RHEL 6: already obsolete ?
Posted Apr 22, 2010 1:58 UTC (Thu) by lwkejrlej (guest, #64237)
Good in theory, but in practice this isn't always the case. OpenOffice 3.2 contains fixes for bugs present in 3.1. So in effect RHEL 6 beta, after the "QA", is actually shipping a more buggy version of OpenOffice than what is currently available (and will be present in F13).
Similar reasoning applies to KDE 4.3 and Gnome 2.28, all of which have been superseded by less buggy versions.
Posted Apr 22, 2010 2:35 UTC (Thu) by airlied (subscriber, #9104)
Posted Apr 22, 2010 3:04 UTC (Thu) by lwkejrlej (guest, #64237)
By the rather long lists of bugs fixed ? Did the RHEL 6 "QA" address all of the bugs present in OO.org 3.1 that are listed as fixed for 3.2 ? Is Red Hat's QA team the same size as OO.org's development team ?
It can be argued that anything with new features carries the risk of bugs. However, recent OO.org, Gnome & KDE are evolutionary releases rather than groundbreaking or totally fresh codebases, and as such it's much more likely that the amount of bugs introduced is less than the number of bugs fixed.
Posted Apr 22, 2010 3:37 UTC (Thu) by rahulsundaram (subscriber, #21946)
Posted Apr 22, 2010 7:41 UTC (Thu) by evad (subscriber, #60553)
The argument is that staying with the old branch and only backporting just the fixes is a good thing. It works whilst the upstream supports it (usually about 6-12 months) but then you are forced to backport fixes to older versions that were never intended to have those fixes. Things break.
Red Hat have admitted this in person to me during subscription negotiations: Red Hat tend to break things badly by back porting critical bug and security patches (as well as new features). They fix security bugs and then they have to fix the fix because it exposed code paths that were never seen upstream.
So, customers should compare the latest stable versions of the software from upstream with an older version being maintained by Red Hat with new features/fixes (cherry picked ones, note!) backported over the top in a way that was never designed.
As you say, newer releases introduce *new* bugs (whilst fixing old ones) and *new* features. I don't really like the Enterprise Linux releases (they are obsolete at release and the package set is small) or the non-enterprise ones (Huge package set but insufficient testing, lots of new untested features, short lifespan). There has to be some middle ground we can get to.
Posted Apr 22, 2010 8:21 UTC (Thu) by rahulsundaram (subscriber, #21946)
That is indeed possible but not what I am suggesting. What I know instead is that even evolutionary updates introduce *new and different* bugs from the previous version while introducing new features and fixing some old bugs. Red Hat cannot simply just use the latest upstream user space for major codebases like a desktop environment or a office suite as has been suggested here. QA, hardware and software certifications take a lot of time and effort.
This is not to claim that backporting is a panacea but it is a solution that has worked better than anything else for the most part for the target market and Red Hat does update leaf packages like Firefox more often and introduce parallel installable versions of new software like Postgres. In the case of Postgres, some customers want the new features in it but simply upgrading to the latest version is not a option when the new version also introduces incompatibilities.
If more upstream projects would stop introducing so many (often avoidable) incompatibilties and take API/ ABI stability more seriously, it would certainly help but I am not sure there is a definitive movement towards that.
Posted Apr 22, 2010 2:37 UTC (Thu) by dag- (subscriber, #30207)
New releases also ship with new features that change behavior or even interfaces. An Enterprise distribution is supposed to keep behavior and interfaces stable and trusted. So shipping latest software goes against that need that business have. The last thing you want is that a kernel update suddenly impacts performance/behavior and outdo any performance tweaks you prepared and implemented on your 400 servers before.
Of course if you don't have that need, you may as well find yourself using a bleeding edge distribution, or one that is more often updated. But that is not what enterprises desire in general. It's a balance of what you think is important, and that might as well be different for different users.
If a typical hardware lifecycle is about 4 years, and commercial software is adapted to run on a new OS release only after 18 months of general availability, you cannot expect a company to standardize on new OS releases every 6 or 12 months. The preparation and migration of your production environment simply is too time-consuming (and too costly)...
Posted Apr 22, 2010 3:12 UTC (Thu) by lwkejrlej (guest, #64237)
I didn't imply that there should be a major update to a critical component, such as the kernel, every 6 to 12 months. That would be insane.
My point of contention is that there is user-space software available now that fixes bugs present in its earlier versions. Those earlier versions are present in RHEL 6 beta, and I do not see Red Hat's development team being able to backport all of the fixes from the newer version. It's simply easier to take the latest version, before Red Hat pours in the concrete and solidifies everything until 2015 / RHEL 7.
Posted Apr 22, 2010 4:59 UTC (Thu) by airlied (subscriber, #9104)
Again this isn't a Fedora or Ubuntu, a major amount of stabilisation work has gone into this already. You can't just shove major revisions of apps in the week before beta release. I think you seriously underestimate the amount of work that goes into this prior to the actual beta release. Its not a just a bunch of maintainers chucking shit at a packaging system and hoping it produces a coherent OS. Like I'll guarantee you can say the exact same things about RHEL6 when it finalises, in that there will of course be newer versions of all those components than it ships with, but that isn't what the RHEL OS is aimed at. Its aimed at people who don't care about constant version revisions, they want a stablised version with lots of bug fixes only, not new themes or applets or whatever KDE4.x+1 might drag in along with the bug fixes.
Posted Apr 22, 2010 5:54 UTC (Thu) by lwkejrlej (guest, #64237)
Fair enough, but can Red Hat guarantee that all the bug fixes present in package version N+1 are backported to version N, which happens to be shipped in RHEL 6 ?
I'm also not totally convinced about the QA process. Going by the embedded versions within the names of package rpms (and the patches in source rpms), it would appear that things such as gnome-panel haven't been really updated from the version in F12. gnome-panel crashes all the time whenever you do a minor interaction with one of its components. This is a highly visible bug that has existed for quite some time, and yet it hasn't been fixed.
Posted Apr 22, 2010 8:26 UTC (Thu) by rahulsundaram (subscriber, #21946)
That is never the goal. Just like new features, bug fixes are also cherry picked. Not everything gets the same priority. Upstream projects often exchange one set of known bugs with another set of unknown issues in newer releases. So just moving to a new upstream version doesn't solve the problem.
Posted Apr 23, 2010 14:12 UTC (Fri) by nix (subscriber, #2304)
Sure, for everything other than kernel and daemon bugs you can work around this with suitable library paths, and for kernel and daemon bugs you can work around it with a VM, but if you need a complete virtualized copy of the pre-updated RHEL to get things done, the updates really have done harm, even if all they did was fix bugs.
RHEL is for stability hounds, and often what matters there isn't that there are no bugs but that *the set of bugs does not change*, so the users can consistently work around them.
As a developer, I hate this crusty stuff, but I understand why the users like it.
Posted Apr 24, 2010 19:42 UTC (Sat) by madscientist (subscriber, #16861)
Posted Apr 24, 2010 20:24 UTC (Sat) by nix (subscriber, #2304)
Posted Apr 25, 2010 1:03 UTC (Sun) by dag- (subscriber, #30207)
In fact, as long as it is timely and digestible, and there is a clear roadmap for minor updates you've got something you can plan your organization around.
Posted Apr 22, 2010 8:06 UTC (Thu) by danieldk (guest, #27876)
In the months I have helped CentOS as a volunteer (and used it on servers) I have seen enough slip through Red Hat's QA that I don't believe in their approach anymore. Many bugfixes are fairly trivial, and can be verified and patched quickly.
Posted Apr 22, 2010 12:35 UTC (Thu) by mmahut (guest, #45550)
Posted Apr 23, 2010 10:00 UTC (Fri) by dag- (subscriber, #30207)
But the opposite (like eg. OpenSUSE, Fedora or Ubuntu) will never be more stable, more reliable or less bug-ridden. If you don't even try to freeze your code-base and instead have a rolling distribution, you can only assume software gets better (both more feature-rich and less bugs) but reality is very much different. Mostly because it will never be as well-tested in enterprise environments as a 2-year old RHEL release.
And yes, Red Hat does from time to time release new features or update software as well, and does introduce new drivers, but for a big part they refrain from making too big changes of which the impact cannot be overseen. Mostly in those areas their customers have no interest in, or Red Hat cannot guarantee to support it.
So the question is, does X past bugs or Y existing bugs in RHEL prove that Red Hat's Enterprise strategy is wrong ? Or does it validate to ship the latest and greatest software instead of trying to keep a stable environment for third parties to depend on ? I doubt it and most large companies seem to agree.
Developers and home users disagree, but they have very different needs.
Copyright © 2013, Eklektix, Inc.
Comments and public postings are copyrighted by their creators.
Linux is a registered trademark of Linus Torvalds