I'm a daily updater
|
| Author | Content |
|---|---|
| jezuch Sep 15, 2007 6:40 AM EDT |
I hope this is a rant against update notification applets, and not against people like me, who start (or, more recently, end) their day with an aptitude -u ;) I'm a daily updater and I'm not afraid to admit it! But, on the other hand, I take the time to carefully study the list of packages to upgrade (and/or remove and automatically install to satisfy dependencies), the changes in size (mostly increases) and changelogs (mailed to me by apt-listchanges). Am I a freak that I *enjoy* that? |
| Steven_Rosenber Sep 15, 2007 8:52 PM EDT |
On my Debian and Ubuntu systems with Update Manager, when I see the notification, I click it and get the updates. But on my Debian laptop -- no Synaptic, just apt and aptitude -- I don't always take the time to su to root, then apt-get update apt-get upgrade. Of course, if I happen to have seen some recent updates on the "big" boxes that look relevant, I would be more prone to update the laptop the next time I use it. I probably stick with Debian Stable instead of Testing because I don't want so many damn updates. Slackware is so new to me, I keep a close eye on security updates because it gives me an opportunity to work on the system. And nothing's automatic. Puppy Linux is totally different in this regard. If you wait a bit after a new version comes out, say 2.17, there will be a 2.17.1, and then you don't do anything until the next version comes out. Then you burn it if you want to, or don't if you don't. I'm running 2.17.1 now, but I'd just as soon stick with 2.14 because that has worked the best for me. (That was the last time gparted didn't take 15 minutes to read the drive info. That said, 2.17 has a lot to offer -- especially CUPS printing and better handling of SFS extensions.) |
| Sander_Marechal Sep 16, 2007 5:35 AM EDT |
The solution is easy: Use a stable distribution :-)Quoting:Presumably the assumption behind updaters is that the newest version of a software package is more secure and less buggy. But that, as anyone who explores his system soon finds out, is an unwarranted assumption. Conflicts between programs, unresolved dependencies, and broken systems all await those who avail themselves of over-ambitious upgrades. The truth is, except for security updates and fixes for specific problems, the average desktop user is likely to have fewer difficulties with an "if it ain't broke, don't fix it" philosophy. These problems don't exist on e.g. Debian Stable. They hardly ever put a new version of a program in the updates. Security fixes and the like are backported. Same goes for Red Hat IIRC. It sounds like Bruce Byfield is running some distro that's not stable. |
| number6x Sep 17, 2007 6:58 AM EDT |
I never update business production machines at a client's site through automatic updates. For people counting on their machines running without problems, all software changes must be tested before they are applied. Running a machine in a business environment, and running a personal desktop are different uses and they produce different requirements. Activity that is acceptable on a personal machine is not acceptable on a bank's ATM or on a vote counting machine. Automatic updates without user intervention are unnacceptable in a business environment. Any OS that does not allow a user to control the update is not enterprise ready. Most businesses with IT staff will run several layers of testing before software changes are allowed on a production machine. First the software change must be unit tested. The single software change is made on an isolated server that is running a copy of the production environment. Each shop should have a standard set of test jobs run after the change is applied. The results should be the same as the original production machine, unless some specific change from the update is expected. new test jobs should be run that test the exact nature of the update's changes. Second, the change will be system tested. There may be more than one piece of software being updated. After an update proves it does no damage on its own, it will be applied with all other updates scheduled for release at the same time on a set of test servers. All tests must be re-run and results should be as expected. Additional tests of software jobs that share files are usually added at this time. Third, the changes can be applied to production. Support staff needs to know the probable impact and should be alert for any issues that crop up. staff who are more familiar with the specific changes need to be on call to aid the support staff. Some shops I have worked at create a fourth layer between two and three. A pseudo-production environment. This starts out as an exact copy of production and has the changes applied to it before production. The data going into production is run through psuedo-production in parallel for a week or so. This allows testing under production level loads before implementing any updates. If you work for a financial institution or any corporation that deals with large transactions, you will probably have the same fear of automatic updates, at least for the business critical systems. The client I worked at who required the most rigorous testing was a telecom. Although I worked in the financial reporting department, their testing policies were based on the idea that when a customer picked up a receiver, they should get a dial tone. Things in production needed to work. No outages or down time were acceptable. We spent as much time in testing our code as we did in development, before it was allowed into production. On my personal machines, I just run synaptic and update. If anything breaks I can figure it out. |
| jdixon Sep 17, 2007 7:09 AM EDT |
> It sounds like Bruce Byfield is running some distro that's not stable. How many stable distributions are there? I know of Debian stable and Slackware, and of course the enterprise edition ones (RHEL and SuSE) qualify. From what I've seen, most of the other distros don't. I'm sure there are exceptions I don't know about though. I run Slackware, so unless I want to run the version in current, all I get are security upgrades and patches. |
| number6x Sep 17, 2007 7:21 AM EDT |
jdixon, Suse's SLED and Server are both stable. Things are in the Beta and the Open editions first. RHEL is stable, and not too cutting edge. Since RHEL is stable, CentOS and Scientific Linux inherit that stability. These distros are aimed at people who want stability over cutting edge. They are probably using the distros for boring work, or running a business and they don't want to risk breaking anything because there's a new nvidia driver. They could care less about new, unless it addresses a known error or problem that affects them and their work. Even then, they will probably test the update before applying it to all machines. You'll hear people complain about distros that are 'old and crufty'. These are probably the distros that will be making the most money. This is the same reason that businesses will stick with WinXP over Vista for the next year or so. XP is a known quantity, even as out of date as it is. Your bank wants its teller's terminals to operate correctly, not to be able to do fancy 3D effects. Boring and stable is more important to businesses than bleeding edge. |
| Steven_Rosenber Sep 17, 2007 9:39 AM EDT |
Usually people upgrade their operating system when there's a specific piece of software that they want to run but can't with their current OS. So as long as everything runs on XP, nobody's going to be in a hurry to upgrade to Vista. But if there's a performance "enhancement," i.e. things that didn't used to work now will, it's REALLY time to upgrade. That's why Windows 2000, for me was a must. When it comes to Windows, only with 2000 did I get a box with any kind of stability. It didn't crash every 10 minutes, USB services worked, it booted in less than 5 minutes. XP built on that. When it comes to Linux, if I wasn't testing a couple of distros a month with the purpose of writing about them, I'd be getting to the point of finding something and sticking with it. And that all depends on performance, hardware and software compatibility, and the box's intended use. Most people with Windows or Mac boxes never "change" operating systems because there really only is one choice, in their minds. And they'll go to great lengths to NOT reinstall their current operating systems, even if such action is warranted, due to: a) fear that they can't do it, b) fear that the box won't work (a variation on a.) and c) they're using pirated software and don't have the discs. When we install three or four Linux distros on a single box, replace one with another on a weekly or monthly basis, it's a bigger change than leaving the same Windows OS on a box for five to 10 years. But whether it's Ubuntu, Slackware, Gentoo, Debian, PCLinuxOS or Debian, at the heart of the system the kernel is the same (although it may be a different version of same). Unless it's BSD. But the fact that there are Linux distributions that get support for 10 years, seven years, five years, three years and six months means there is pretty much something for everyone. Some "stable" distros on some hardware do everything their users need. Others don't, and they'll never be patched. And who doesn't have a story about upgrading an OS and finding out that things didn't work as well as before? For me, the six-month release cycle is way too short. I know it's all about marketing, but I'd rather see a two-year release with fixes in the middle and a growing repository rather than an effort to reinvent the wheel twice a year. |
| Sander_Marechal Sep 17, 2007 11:24 AM EDT |
Quoting:> It sounds like Bruce Byfield is running some distro that's not stable. Plenty. "Stable" in the context of the author's complaint means a distribution that does not release new upstream versions as updates, but backports security fixes from the new upstream versions to the old versions that are included in the distro. The entire dependency argument is then moot, because no dependency ever changes. Even Ubuntu is "stable" in that regard. |
| Steven_Rosenber Sep 17, 2007 1:14 PM EDT |
Ubuntu has never broken for me as the result of an update. To the contrary, I've had more success upgrading Xubuntu from 6.10 to 7.04 through the Update Manager than I have had installing 7.04 from scratch. |
| Sander_Marechal Sep 17, 2007 3:44 PM EDT |
Steven: We're talking about updates, not upgrades. My point was that a "stable" distribution does not upgrade your package (to a new version) but only updates your packages (with backported securiy fixed and important bug fixes). |
| Steven_Rosenber Sep 17, 2007 4:19 PM EDT |
Question: When in the recent past has an update made things worse? I'm interested in any instances y'all can recall. |
| Scott_Ruecker Sep 17, 2007 4:27 PM EDT |
Only with Windows, after unintentionally upgrading the media player.. |
| Steven_Rosenber Sep 17, 2007 4:51 PM EDT |
My big ones: Going from Mac OS 10.3 to 10.4 broke Adobe Acrobat Distiller in Classic mode. And ... going from Windows 98 to 2000 screwed up ACPI shutdown. Now these are UPGRADES, not updates, I know, but still, it hurts. |
| jdixon Sep 17, 2007 6:48 PM EDT |
> My point was that a "stable" distribution does not upgrade your package (to a new version)... Not necessarily true. When the security fix upstream is "upgrade to version x.y.z", the new version is compiled on the same system, and there are no downstream dependencies, upgrading to the new version is called for. This frequently happens with Slackware, which I don't think anyone would deny is a stable distribution. I believe all of the patches to Slackware 12 since it has been released are upgrades to newer versions of the packages. |
| flufferbeer Sep 17, 2007 7:15 PM EDT |
Trying to install OpenOffice 2.0.4 in Debian after upgrading to the stable Etch 4.01 (now just where is the package that holds ca-certificates found, hmmm............) |
| Sander_Marechal Sep 17, 2007 9:49 PM EDT |
Quoting:Question: When in the recent past has an update made things worse? I'm interested in any instances y'all can recall. Try running Debian testing :-) Or any distro that has a continuous upgrade cycle really (Slackware/Zen come to mind but I admin that I haven't used them enough to know). Quoting:> My point was that a "stable" distribution does not upgrade your package (to a new version)... Even then, the correct policy of a stable distribution is to find out what the fix was in the new x.y.z version and backport that fix to version x.y.f that's used in the distribution. After which the distro releases x.y.f-2. After all, x.y.z. may have regressions as well as fixes. Only when the problem and the fix are so structural that backporting is not feasible (e.g. old versions of BIND) then the package should be upgraded to the new version in it's entirety. But that's a very rare and exceptional case. PS: What I described here is pretty much the exact policy taken by Red Hat and Debian. |
| jdixon Sep 18, 2007 12:43 AM EDT |
> What I described here is pretty much the exact policy taken by Red Hat and Debian. And, as far as I can tell, exactly the opposite of what Patrick does for Slackware. He relies on the upstream provider to make a security fix, and upgrades to the version which contains the fix. Only when that's not possible does he patch things himself. However, there are two conflicting definitions of version involved here, which could be causing communications problems. :) The term version can either be used to mean an upgrade from say Firefox 1.5.x to Firefox 1.5.y, or it can be used to mean an upgrade from Firefox 1.5.x to Firefox 2.0.y. Patrick doesn't do full version upgrades, only point upgrades. For instance, the lastest Samba patch was an upgrade from samba-3.0.25 to samba-3.0.26. |
| Sander_Marechal Sep 18, 2007 2:15 AM EDT |
AFAICT Debian and Red Hat don't even do point release upgrades. Even they can contain regressions or other updates besides the security and serious bug fixes. |
| number6x Sep 18, 2007 5:35 AM EDT |
My point was about the use of an OS in a business application. Let's say that your run a home grown application that reconciles all of your retail outlet daily transactions with your warehouse inventory and produces order reports for your buyers. This way they know what parts they need to order, and if they can save money by shipping surplus from one warehouse to another. This is the life blood of your retail operation. Lets say your OS vendor puts out a patch that upgrades the firewall shipped with the OS, and resets the defaults. Your homegrown app does not work with the default settings, but with your custom settings. Your app goes down and your inventory is out of whack for a few days while you figure out what is wrong. Maybe its a system that reconciles transactions for a bank's ATMs, or the system that handles incoming 911 calls for a city's police and fire. My point is not about your machines or the machines on a desk at some workplace, but about the OS's used in a businesses infrastructure. In these situations all changes must be tested in order to applied to the production system. Any OS that does not allow the users to control the application of updates is not ready for enterprise use. Users must be in control of the software that runs their business infrastructure, there is no acceptable reason for the software vendor taking control without the user's consent and knowledge. Any OS that allows what Windows Update does should not be used by a corporation for any enterprise class installations. Just ask Skype. http://www.freedom-to-tinker.com/?p=1190 |
| Sander_Marechal Sep 18, 2007 1:09 PM EDT |
Quoting:Just ask Skype. Well, that's what you get for relying on someone else's systems :-) |
| rijelkentaurus Sep 18, 2007 5:41 PM EDT |
Quoting: Well, that's what you get for relying on someone else's systems :-) That is a lesson foreseen by Larry Ellison at Oracle long before it became a problem for him and his company, hence Oracle's affinity for Linux. He knew his database would eventually be crippled in some form by MS. Quoting: AFAICT Debian and Red Hat don't even do point release upgrades. Even they can contain regressions or other updates besides the security and serious bug fixes. It seems to depend on the point update itself. No point expending effort where it's not needed, but they do a nice job of keeping their customers in mind. That's what happens when a company can move away from Red Hat to another Linux with relative ease: Good Customer Service. Something MS knows nothing about. http://www.redhat.com/security/updates/backporting/ |
Posting in this forum is limited to members of the group: [ForumMods, SITEADMINS, MEMBERS.]
Becoming a member of LXer is easy and free. Join Us!