root@quatro src$ uname -a SunOS quatro 5.11 snv_133 i86pc i386 i86pc Solaris root@quatro src$ pkg search gcc Segmentation Fault (core dumped) root@quatro src$ exit exit benr@quatro src$ pkg search gcc Segmentation Fault benr@quatro src$ pfexec pkg search gcc INDEX ACTION VALUE PACKAGE description set GCC pkg:/SUNWgccruntime@3.4.3-0.97 description set GCC pkg:/developer/gcc/gcc-libgfortran@4.3.3-0.133 description set GCC pkg:/developer/gcc/gcc-libssp@4.3.3-0.133 description set GCC pkg:/developer/gcc/gcc-libgcc@4.3.3-0.133
Why does IPS still suck? Seriously, we can’t catch errors before segfaulting? And people wonder why I claim that IPS and AI are so immature. I just can’t wait for Oracle to cram this down my throat.
I want SX:CE back. Anybody in MPK17 listening!?!? Stop telling customers they are stupid for using post-install scripts, stop pontificating about how you know better. ZFS made claims to rightness and proved itself. IPS has yet to convince me…. its had years, and still has yet.
Update:
I’m getting email from folks asking about the underlying problem here. The issue is that Python freaks if you have alternate version installed an in your LD_LIBRARY_PATH or PYTHONPATH.
root@quatro ~$ pkg search virsh Segmentation Fault (core dumped) root@quatro ~$ unset PYTHONPATH root@quatro ~$ pkg search virsh Segmentation Fault (core dumped) root@quatro ~$ unset LD_LIBRARY_PATH root@quatro ~$ pkg search virsh INDEX ACTION VALUE PACKAGE basename file usr/bin/amd64/virsh pkg:/SUNWlibvirt@0.5.11-0.128 basename file usr/bin/i86/virsh pkg:/SUNWlibvirt@0.5.11-0.128
But that’s not the point I’m making… the point is that there isn’t a wrapper to catch these types of issues. IPS is a critical system utility and shouldn’t be derailed by something so simple. Its just immaturity.
I completely agree, we at Blastwave are totally unable to convert our stack to IPS mostly because it is scriptless. Safe installation and upgrade of production systems without installation scripts are nearly impossible.
Let’s hope that the future with Oracle will be brighter for OpenSolaris…
I feel your pain. I recently tried to image-update to 133, and after two hours of swapping (due to pkg using some completely obscene amount of RAM) my ssh connection dropped and the box became unresponsive.
Yeah, this was a piddly x86 box, and yeah, this was due to the “one time” package renaming, but… really? Is that an acceptable result? Should IPS really be encountering a situation where it requires over a gig of RAM to function?
IPS is a bunch of good intentions crammed into a half-baked implementation. It’s been “not quite there yet” for way too long, and Oracle needs to take fixing it seriously.
Amen. Amen. Though I don’t agree that IPS fails on its implementation (which, admittedly, is wretched.) I think it fails in its most fundamental concepts. It’s awful. I’m sorry, I hate to dismiss something so many people have worked so hard on, but it’s terrible. Ian Murdock has taken a superb enterprise O/S and turned it into Linux circa 1998. There’s a lot of change at Sun. This needs to change. (And they can bury all that pfexec crap in the same hole too.)
Jumpstart works too. Find it too hard? Think about a different job. Sys V package tools do what they need to as well. Extend them, speed them up (please!) but don’t just break everything. Put the default ksh back to ksh88 so I *know* all my scripts will work.
Forget the desktop. Forget IPS. Forget trying to please idiots off Slashdot who dismiss Solaris after five minutes because “tab doesn’t work”.
If I were a Sun customer now, I’d very seriously think about moving to Linux or BSD. I wouldn’t know when, if ever, I’d get the cool new stuff put into Solaris 10, and there’s no clear upgrade path to OpenSolaris. There are also MAJOR compatibility issues. It feels like a toy rather than an enterprise O/S. OpenSolaris will kill SunOS altogether, and long before it becomes the only option.
[@Jonathan Blanchard: I think the recommended way to do scripting at package install time under IPS is to install a transient SMF service to run the script.]
For me IPS is the worst of the new developments in OpenSolaris. It trades the old problems with the SysV packages for a whole load of new ones that should not have arisen had the system been designed correctly. The major problem for me is that it is just far too slow – at least an order of magnitude slower than the SysV packaging system. Having said that, it does, for the most part work. It should not still be worse in some areas than the old packaging system after so long in development.
@benr: Just did a pkg search gcc on a clean install of build 134 and it worked fine (both as the user created during OS install and a separate one I created now). So either the bug you came across was fixed in build 134 or there is something else causing the problem on your system. HTH
These two quotes from the comments sum it all up perfect:
“It’s been “not quite there yet” for way too long, and Oracle needs to take fixing it seriously.”
and
“I’d very seriously think about moving to Linux or BSD.”
If you have to dump so much of your experience to move to Solaris Next, why not just use it as the opportunity to move to Linux.
I know the idea is to rise like a Phoenix… but after so long IPS should have already succeeded, and it hasn’t.
Why don’t we get rid of the PATH statement while your at it? Then stupid admin mistakes can’t break the system.
The only thing this posts states is that the admin made a mistake and we need to toss out the OS because of it.
Grow Up.
alan
How to add a new package on your beloved SXCE:
1. Google to see if someone has the desired package in SVR4 format. Fat chance, right? Mess around with downloading 100s of MB’s in companion and supplemental CDs from Sun, looking for your package. Ouch!
2. Screw it. Download the source .gz and do the configure;make;make install ritual yourself. Take hours in hacking autoconf and working around Linux’isms that foil your attempts in a clean build. Even if you get lucky and everything works, you have no package metadata to help you 3 months later.
3. Resign yourself to grabbing the package from Blastwave but not before also getting the 200MB of other dependency crap that you don’t want. Expect to suffer CSW $LD_LIBRARARY_PATH hassles for the next two years until your system just breaks because of library incompatibilities.
I much prefer:
pkg install git
myself
Well to Alan Pae, I have just spent 4 days trying to get the latest and greatest dev version of your Open OS just to do a simple system update and to have package manager install studio 12 and net beans. Well day four and still error time out on package manger for any and all updates did the PKG_CLIENT_TIMEOUT trick to a ludicrous 500 and still only getting about 224.98MB oh wait 223.76MB oh wait 220.56MB, damn Error: 28 Reason Package manager taking to damn long so I died on you. End error.
This is why people hate your package manager IPS system. So what did I do? I logged into Sun grabbed Studio 12, Netbeans, the required Java JDK and manually installed them.
Welcome back SX:CE!!!!!!! and you want me to pkg install the entire system?
Get the damn GUI working!!! Hell even Solaris 10 U8 can do a nice GUI with system updates for SVR4 packages, So whats your problem?
Hey, Thanks Ben for your great site and I hope for the return of SX:CE, it is/was a great OS.
Keep the opinions comin’ guys. Solaris _IS_ a great OS, unquestionably. I simply think IPS hurts, rather than helps it. At least in its form today.
If I’m full of crap, I’m glad to hear it. Straighten me out.
Choosing Python for IPS implementation was probably not the best idea (performance wise).
@Pete: I agree with you. IPS is very easy to use and waaaay better than the old system, but PLEASE: Why is it so slow? Python? Who knows….
Slow, IPS Slow? Try doing a 200+ system update over a wireless 802.11G connection= ERROR!!!!! I have yet to get my system updated, no wired connection Real Tek driver not supported. So I sit here using pkgadd to do all my work. I can browse the web to get files but IPS is still sucking hind tit. If you are from the country you will understand “hind tit”. I don’t care if the OS is a BIG FREAKING download, its much better than dealing with not having an up to date system at the time of pressing. no I have a system that is behind no matter what I do. The only thing I am not happy about SX:CE or Solaris 10 is no firewire CD boot support. So this machine get none of the good stuff until I get a USB DVD drive. So I fight and fight Thankfully this is not a machine I use everyday, but if the OS would work like a real OS and provide me with what I need according to the advertisements I would be a whole lot happier. I have to ask why do you want Solaris exactly like Linux? I thought they were the competitor? I guess someone wants to do a Vista/Windows 7 photocopy of Mac OS X.
Well I guess be pisses about my comments if you want I love Solaris, but when you piss into the wind you can only guess what will happen.
Bizarre design decision in IPS: packages aren’t packages, just a list of files. So retrieving packages over the network takes much longer than it should due to connection setup and teardown.
Another strange one: no support for installing from local files (yet).
On the plus side the integration with ZFS and the automatic dependency checking are definitely an improvement, for me anyway, over Svr4 packages.
But what is the sense in checking dependancies when you cannot push data over the network? It takes half the amount of time to run dependancies (and I don’t care about extra cruft large disks are cheap) than it does to watch my IPS upgrade or install fail and never give me the result I want or need. I would rather search for my support files and have redundant support files than to not have a system up or program that I need to run my work.
I tried Nexenta which brings Opensolaris with proper package manager (apt-get).
IPS is a horror when you use a http proxy, mirroring the IPS repository is deadly slow (rsync X millions of files…).
Will they switch to dpkg/apt-get ?
The one thing I expected from IPS was speed. The big moan about upgrading (i.e. patching) Solaris boxes has always been how long it takes, but man, IPS is slow. Why on earth is it written in python? Maybe i’m old-fashioned, but shouldn’t core O/S tools be written in C? If you can’t code in C, I don’t want you writing my O/S.
Maybe one day IPS will be stable, solid, well documented, 10-100 times faster, won’t require a fast connection to a remote repository, will work sensibly with zones, and won’t break the system when you upgrade it. Maybe the installer will offer some kind of customization and an upgrade option.
I use heavily virtualized, minimal SXCE systems. I don’t have the luxury of shipping out all my data, rebuilding everything from scratch, and bringing the data back, or of learning over-complicated new systems on an OS I’ve used for 15 years. Therefore opensolaris is not fit for my use. It should not have become the only option until it was a drop-in replacement. I’d love to see Oracle can it and go back to SXCE and 10. If other people want to build distributions with other packaging tools and installers, they can.
@benr – LD_LIBRARY_PATH? I haven’t used that since Solaris 7!
Ben – with all due respect, if the fault lies with LD_LIBRARY_PATH it is hard to blame pkg. Yes, pkg has problems but so does the much older package manger SXCE uses. Nevertheless, I am sympathetic to the concerns of people who decry the abrupt termination of SXCE and the transition to a new package manager with (in my opinion) less regard than warranted to the impact of this change on large scale users. Everything has to be re-engineered to work with the new system. Even if the new system has some desirable features, and I think it does, it requires a lot of time and effort to move to the new system. What effort has been made to ease the transition? I don’t see evidence that much of anything has been done. No wonder there is frustration. So while I see reason to gripe, I don’t blame it all pkg.
At the risk of appearing unsympathetic (although we in the OpenSolaris/Solaris group do take all constructive criticism to heart), the fact is that SXCE wasn’t a product. If folks were treating it as such, that’s unfortunate. It was a release *under development* and that release is the next enterprise version of Solaris which for sake of argument might be called Solaris 11. But that release hasn’t come out and that’s the release you should be comparing with Solaris 10. And when that new enterprise release is made available, we expect there will be migration tools and strategies to make it easier to move Solaris 10 (or earlier systems) to the new thing.
As for abrupt termination, I believe there was fair warning given and the fact was that producing two entities was costly. We all want to get to that next enterprise release and producing only one development build will help speed that process along.
For those who want to let us know where things aren’t working, please file bugs as they’re trackable. Providing us as much data as possible is obviously important but among other things, that should include the version of pkg(5) (“pkg version” will provide that) along with the list of publishers you’re using (“pkg publisher”).
IPS was written by engineers who do not have any python background. It was not tested for reliability, performance or scalability. It was not even designed for customers who pay for patches, so has numerous sustainability issues. I hope Oracle takes appropriate steps to rectify all that. ‘it’ll be fixed soon’ does not convince our management anymore to continue with Opensolaris license.
Well David your comment about SX:CE does not make sense. So you pick one of the two dev OS’s to terminate? I and everyone else had been told that Opensolaris was Solaris 11 not SX:CE and if what you say is true why did you can SX:CE when it was the model for the next big release??????????????????????????
Apparently somebody did not want their pet project sent to the trash bin where it should be, and told to work on just the back bone parts IPS etc etc. I think that is what really happened and somebody made a bone head decision.
Say all you want about Opensolaris but until you can get a system updater that is in the 21st century and not back in the days of twig rubbing you will find less adoption and more complaining.
Users want something easy and if not bug free decent performance for new and great features.
To be reliant on features that are half baked in a situation were the old ways work you have to be pretty crazy to not utilize the old ways but now you have no choice but to deal with the half baked features.
Did anyone sit down and think about the consequences of backing people into this corner you have driven them to?
And the bugs, I guess if someone did not have such an infatuation with the linux userland and had to rehash a perfectly fine OS you would not have had to spend countless man hours on the rehash and instead spent it on squashing said bugs. spending three years to throw in gnu utilities? I guess the linux user base that never came was extremely important?
I guess I play the waiting game I already said my peace in the forums and we will see what happens.
Oracle please bring back the proper OS that should have lived through this ego fest that produced OpenSolaris…….. SX:CE revival.
I agree 100% about post install script. We have still to find a decent example from IPS about doing a post install replacement script and that is just ridiculous.
David – I understand the arguments you make and it is fairly obvious that maintaining SXCE would be expensive and this had to be difficult during tough times. The problem isn’t with SXCE per se but the fact that its cancellation implies that Solaris Next will be an IPS-based system. This means that at some point customers are going to deal with the transition costs unless they want to forever stay on Solaris 10. I think Oracle needs to reassure customers that they’ll be there when the time comes to ease the transition. The FUD factor has to be dealt with. We really don’t have a clear picture of what the future will bring and that worries us.
Fabien: Will they switch to dpkg/apt-get ?
Yeah, right. It’s NIH, so it would never go over with Solaris users. I mean, look at all the people here who are complaining that OpenSolaris has too many Linux-isms; there’s no way they’d accept a package manager that originated on Linux.
i_dont_do_internet_names: I’m not convinced the speed problems are caused by python. RedHat’s package tool, yum, is written in python and is reasonably fast. Something else is wrong here.
Rand Huntzinger: …with all due respect, if the fault lies with LD_LIBRARY_PATH it is hard to blame pkg.
Well, yes and no. I wouldn’t blame it for failing, but it’s disconcerting to see an important system tool segfault. It makes you wonder what other sorts of problems might be lurking. I tend to regard a program that segfaults as a program that’s likely to have dangerous buffer overflow issues and the like, because if they can’t be bothered to do error checking in this case, what else have they overlooked?
Chad wrote:
===
Well to Alan Pae, I have just spent 4 days trying to get the latest and greatest dev version of your Open OS just to do a simple system update and to have package manager install studio 12 and net beans. Well day four and still error time out on package manger for any and all updates did the PKG_CLIENT_TIMEOUT trick to a ludicrous 500 and still only getting about 224.98MB oh wait 223.76MB oh wait 220.56MB, damn Error: 28 Reason Package manager taking to damn long so I died on you. End error.
This is why people hate your package manager IPS system. So what did I do? I logged into Sun grabbed Studio 12, Netbeans, the required Java JDK and manually installed them.
===
I use pkg all the time, even to do system updates from old build to new build, all without a hitch.
I can move from build 133 to build 134 using:
pkg image-update be-name name
and then about an hour later I reboot and off I go. If I want to add a new pkg it’s painless. If I want to search for a pkg then pkg search -r whatever works for me.
There have been some announcements about “repo on a stick” and if you’re having issues maybe you should go that route.
I do this completely over wireless and like I said, I’ve personally never had an issue.
alan
David Comay writes:
“For those who want to let us know where things aren’t working, please file bugs as they’re trackable. Providing us as much data as possible is obviously important but among other things, that should include the version of pkg(5) (“pkg version” will provide that) along with the list of publishers you’re using (“pkg publisher”).”
Don’t you understand David? WE DO NOT WANT IPS!!!
Let me repeat that again: DITCH THE EFFORT. ABORT. ABORT. ALL HANDS ABANDON SHIP!!!
I want SX:CE back!!!
And Shawn Walker & Co. to stop doing what they are doing and to stop telling me that they know what is best for me, because they do not.
The point is that IPS being a “no scripting zone” is conceptually flawed.
And yes, it’s Python-C cobbled-together… substandard “solution”.
“Maybe i’m old-fashioned, but shouldn’t core O/S tools be written in C? If you can’t code in C, I don’t want you writing my O/S.”
That quote sums it up for me. I expect a “real” system to be designed and written with strict attention to detail, hopefully with a high performance language such as “C”…and NOT with the latest “fad” language. Certain applications written in Python have their place. But not a core utility. I dislike Linux for this (but tolerate it because I don’t regard Linux as being in the same ballpark as Solaris 10, HP-UX, or even a dead OS such as IRIX for that matter). But I do indded set higher standards for Unix in general, and if Solaris is going by the way of Linux, then why bother running it when I could simply run………Linux……….and get the same “fad language” programming style and wonderful “me-too” bug-fix quality. Uh huh. Sure. Let me jump right on OpenSolaris.
“The FUD factor has to be dealt with.”
The FUD factor? How’s this:
I’ve got HUNDREDS of CONFIGURATION PACKAGES which create databases, configure DNS servers, configure firewalls, and manipulate the platform in a consistently repeatable way. How am I able to achieve this? Why, with preinstall, postinstall, preremove, and postremove!
With IPS, all of that is out of the window. I can install a “transient SMF service”, but how will that “transient SMF service” know to trigger a remove action when I remove an IPS package?
How would I prevent the sysadmin from accidently running “svcadm enable” or “svcadm disable” and blowing an Oracle DB away?
IPS just busted from the ground up. The design of it is busted; the implementation is busted; and Ben is right, the busted implementation doesn’t work, and it has had pah-lenty of time to cook.
I keep writing this like a broken record: if they wanted to go IPS route, they should have BOUGHT sgi IRIX 6.5 inst and roboinst technology; it does everything IPS does, PLUS it understands System V pkgadd AND it understands HP-UX SD-UX depots, goes over the internet (if sysadmin wants to do that), and all that in addition to automatically resolving dependencies and handling namespaces and subsystems.
It’s 20 years old technology proven at the sysadmin battlegrounds and still lightyears ahead of the latest IPS.
UX-admin wrote:
“It’s 20 years old technology proven at the sysadmin battlegrounds and still lightyears ahead of the latest IPS.”
Another good point. Why do people see fit to reinvent wheels when there are tried-and-true technologies out there to be had? It seems to happen more and more these days. Everyone wants to ditch “the old ways” of Unix, only to produce “new” software systems FROM SCRATCH that really don’t cut it in productions environments that have matured over 20-30 years?
Many companies need the ability of customizable features in software packages. In fact, the right way to maintain a Unix system is to “never touch a running system” which means that change management needs to be implemented, and every change to the system must be traced. Many companies use native package tools for this purpose…to deliver consistently repeatable changes to production systems. So any good software deployment system MUST scale.
I verified that this works fine in 200906, I tried to post the script output but it says it is spam – it works for a grunt user to search for gcc…..
problem is not IPS (which could be, but that’s something else)
problem is generally low quality of these OpenSolaris builds since the release (200906)
The problems I have with SVR4 pkgadd are 1. no build receipt (like rpm .spec) to automate the src,patching,configure,build and install process 2. no package dependency auto resolution. 3. no package depot frameworks. IPS only address 2 and 3. and need CBE to address 1. IPS didn’t offer binary(package management) compatibility philosophy that Solaris users expected. This is the most costly change to customers and biggest complain I have. For now I am stayomg with TWW toolsets that resolve all three problems and offer package management compatibility.
Alan Wrote “use pkg all the time, even to do system updates from old build to new build, all without a hitch.
I can move from build 133 to build 134 using:
pkg image-update be-name name
and then about an hour later I reboot and off I go. If I want to add a new pkg it’s painless. If I want to search for a pkg then pkg search -r whatever works for me.
There have been some announcements about “repo on a stick” and if you’re having issues maybe you should go that route.
I do this completely over wireless and like I said, I’ve personally never had an issue.”
HOLY COW “and then about an hour later I reboot and off I go” An hour for 250MB of system updates?????? How long do you expect a user or admin to wait for the system to be ready??????????????? I thought we were using high speed internet not 56K modem connections I mean maybe ISDN speed for networks these days? I can see that searching for packages and dependencies are not a problem now that I know I will be waiting at least an hour for my system to be ready. Also why are you not using the package manager GUI? Is it just as broken as I am saying? Why would I want to waste an 8GB USB drive to hold a repository that should be fully functional to all users by now? Oh well back to my b130 SX:CE.
Also I never heard back on why David said that SX:CE was the model for the Solaris 11 (Next) So are you just developing a new desktop OS for Sun/Oracle (OpenSolaris) and keeping the good stuff (SX:CE) behind closed doors and ready to surprise us when the next Solaris update comes out? I think that would be the smartest decision you all have made since starting this project.
I never said that I waited an hour, I said about an hour later I reboot. I also don’t use an 8GB stick because the repo does work, for me.
Well a works for me is not a very business like perspective is it? A works for everyone is what it should be. Three years to rehash something that could have been easily obtained from some place else and it costs your customers this much pain. Not very business like!!!!
What is the difference between I waited and hour and about an hour later? Sounds about the same to me.
And yes the repo is over 7GB so if it doesn’t work you have to use an 8GB stick and that is a waste when the product should function as advertised for everyone not just the singular me’s.
Sorry for the anger Ben, I just cannot believe, the dog food we have been fed is filled with all this useless non-functional filler. It just waters down and kills the productivity of what should be a great product if the right product was being delivered. Just leave OpenSolaris to the desktop (think of the Java Desktop Release 2 for Linux) and put SX:CE back as the poster child for Solaris Next.
BTW this smells alot like the Java Desktop Release 2 for Linux you all did back in 2003-2004, sounds like someone in Sun has an infatuation with Linux.
Well the rest of the opensource community has been at this packaging thing for decades and have evolved robust, stable enterprise-grade solutions. Since IPS is a wheel re-invention driven by NIH tendencies it is not unreasonable to assume that this will take a similar amount of time to become robust and stable.
This is not to say that IPS does not have good things. On the contrary it simplifies many things and solves problems with SVR4 and brings a modern package distribution framework to Solaris. However these many good things are also present in the other robust stable systems that the community has developed.
Even SVR4 conceptually had many good features but the implementation sucked big time. I have worked with that codebase and I know. It should have been possible to even do a partial re-design and a re-write retaining better compatibility and ease of migration. SVR4 had bad performance issues and Casper Dik showed how with a little effort he drastically improved SVR4 performance.
One of the goals of IPS was to be portable. When we are at the task of designing a software management and distribution system for a specific OS, portability requirement seems quite weird. However IPS is also being used in other environments like managing user’s Java software stacks on other platforms like Windows. IMHO mixing these diverse goals was created unique troubles when our primary priority was to evolve a modern package management mechanism for Solaris.
The packaging dilemma existed at the very least since Solaris was open sourced in 2005.
pkgadd(1M) supported -d http://... for quite a few years.
How long would it have taken to extend and refine that functionality?
How long would it have taken add automatic dependency resolution with the backgraph algorithm?
If these last five years were used to polish/add these features, pkgadd would have been more than a viable alternative to any IPS inventions. Just imagine how much it costs to design and implement a whole packaging system from scratch!
In five years that were spent solving the packaging problems, pkgadd(1M) could have been more than enhanced…
However, when it comes to UGG Ultra Short making your purchase it is a good idea to buy a half or full size smaller than you would normally wear. This is because the sheepskin UGG Ultra Tall within these boots tends to give a little as it molds to the shape of one’s feet. By going for the half or full size smaller will ensure that your feet remain snug and warm at all times when wearing them.
While there are many things I like about OpenSolaris, the fact that much of what I liked about Solaris being broken or gone, really pisses me off. Things like Jumpstart, sparse root containers, CDE(hey it’s useful at times), separation of personalities (Solaris, GNU, BSD, etc.), and others. The biggest pain I would say has without a doubt been the IPS system. There are some nice features in it, like dependencies, automated ZFS snapshots for upgrades, only download the deltas, centralized repos, package fixing, package search, etc. However, it lacks fundamental features like pre/post scripts, install/uninstall scripts, Solaris sparse root zone support, etc. When coupled with the poorly written AI, it makes for an OS that can’t be deployed and fully automated anywhere near the level that Solaris 10 or below have been capable for a very long time. I’ve even seen the JET developers at Sun take the stance of waiting to see where IPS and AI go before even attempting to support this immature package and installation system we’re stuck with in OpenSolaris.
I think the intentions were good, but too much focus was placed on making a desktop OS and not an enterprise OS which is what Solaris is. I find it a sick joke that they even wrote IPS in Python. While it’s a good language, I would not use it for building OS utilities anymore than I would use it for running an ERP system. I’ve seen the same kind of goofiness with Perl and other languages being used for the wrong tasks. IPS should be written in C or C++, end of story. The IPS developers seriously needs to dig up SYSV package tools and bring back features that we are all dependent on. If you want to improve on it, great!. But don’t just toss it out the window because it’s old. I still believe that the old package management system could have been enhanced to do what IPS does, only better. Sometimes it makes sense to re-write things and other times it’s easier to enhance.
Honestly, if Sun/Oracle doesn’t fix the IPS and AI issues quickly, I doubt any major Sun shop will be happy to deploy or manage OpenSolaris. I’ve experienced enough problems going from b127 to b133. I had to manually force an upgrade to b128 and then I could use the normal command-line tools to upgrade. Basically, I’ve learned that before I click on the pretty GUI tools to do something useful for me, I better read the release notes because the quality of the OpenSolaris builds is sub-par.
Needless to say that if I get elected to the OGB, I will be making this a priority to fix. Without solid automated deployment frameworks, solid package management, and most importantly solid QA testing, OpenSolaris will not be capable of replacing Solaris 10.
Okay.. soapbox mode off;)
+1 on Alan Pae’s comments.
I can’t comment on the usability of IPS directly since I havent used it. We’re a Solaris 10 shop
.
But for people blaming the slowness of IPS on python and pining for the good old days of s5r4 packages…
I’d just like to point out that the majority of the internals of Solaris 10 patching subsystem is still written as shell scripts.
Just look at the insane number of short lived processes that get spawned while installing a patch.
So as far I’m concerned, even python has got to be a step forwards..
“I’d just like to point out that the majority of the internals of Solaris 10 patching subsystem is still written as shell scripts.
Just look at the insane number of short lived processes that get spawned while installing a patch.”
That’s because most people get patching backwards: sysadmins patch PRODUCTION SYSTEMS ad hoc, hoping that some patch XYZ will fix whichever problem they are encountering.
Guess what? That’s a sure-fire way to make an already bad problem even worse.
NEVER TOUCH A RUNNING SYSTEM.
The way it’s done right, and WORKS, is to have a single test/integration test environment, where the patch is first applied and tested on a STANDARDIZED BUILD OF THE OPERATING SYSTEM.
Then, if that works, it is integrated into that standardized build — into the Flash(TM) archive.
Wait, we’re not done yet! Where is the supporting documentation? The operating manual? Where is the requirements ID in the requirements management system?
Aaahhh, now that you submitted the documentation for the RID, and for the fix, now here’s your WINDOW for TESTING in product testing environment. You get to flash a target system from bare metal to a working OS, with your patch already applied to the Flash archive.
Works? Yeah? Can you show the test set documentation? M-hm. Is it sane? Ah, nope, sorry, this test failed. Return to sender! You have until the end of your PT window to fix this.
Works? Yes. Documentation of the tests is submitted? Yes. Is it complete? Yes. Did all tests pass? Yes.
OK, you have product acceptance. NOW you are ready to go into PROD. Here is your window. You are not allowed to deploy BEFORE or AFTER the window (deployment software will not allow it, and if you override, you better have a damn good reason, because it will report you to a change control authority.)
How do you deploy? Cluster node down – flash, cluster node up. Repeat with all the other cluster nodes. Downtime? ZERO. Work involved?
Watching pretty text from STDOUT scroll by in your web browser.
Lather, rinse, repeat for all the tens of thousands of systems in your environment.
Deploy your patch in seconds instead of hours or days.
Guess what? You’re not going to be hacking any PRODUCTION SYSTEMS with ad hoc patchadd(1M) commands today. That’s for amateurs, not system engineers.
“NEVER TOUCH A RUNNING SYSTEM”
Long story short, what is wrong with applying patches with Liveupgrade?
http://www.mbtmall.com
http://www.jordandi.com
Oh, please do not mention liveupgrade, another piece of sub-par tool that only “works” for those ego particular “me”s!
Given a zfs root system with seperate /var or /opt, or with local zones on seperate zpool/zfs, your luupgrade / luactivate will ruin your system with the “mount point syndrome” ( I dubbed it so) And, No, following link won’t always fix it, it will be back….
http://www.solarisinternals.com/wiki/index.php/ZFS_Troubleshooting_Guide#Live_Upgrade_with_Zones
Watch out “Mount point syndrome” in Solaris Live Upgrade with zones!
Although I have not tested a separate /var or /opt, using the LU with 12 zones in a separate disk with the conjunction of PCA patching, “We” never had a problem.
We use pca as well and love it more than any other patching tools provided by Sun/Oracle…..
For live upgrade, I am not saying that we never had a successful run for live upgrade. Like the point in this blog post, what we mean is even a very simple variation in the OS config will BREAK DOWN the live upgrade to its knees. To give you some examples:
Other than separate /opt or /var;
(1) Try mix QFS mounts when you build lucreate with -p (on another zpool) , it will try to traverse every single file in the QFS mounts and finally DIE because it is impossible to include all files in QFS mounts in Alternate BE (It never should be, if you create Alternate BE in the same zpool, it knows bypass QFS mounts…. , but again, you get my point that how easy live U can break)
(2) Try blend the zones with ZFS, if you ever tried “zfs set mountpoint=…” on the parent zone path before creating zones, sooner or later you WILL see something like following after luactivate to the alternate BE:
df: cannot statvfs /.alt.tmp.b-u2f.mnt/var/run: No such file or directory
df: cannot statvfs /.alt.tmp.b-NZf.mnt/var/run: No such file or directory
df: cannot statvfs /.alt.tmp.b-2Zf.mnt/var/run: No such file or directory
df: cannot statvfs /.alt.tmp.b-k4b.mnt/var/run: No such file or directory
df: cannot statvfs /.alt.tmp.b-8kc.mnt/var/run: No such file or directory
df: cannot statvfs /.alt.tmp.b-Zsc.mnt/var/run: No such file or directory
……
At this point, it is pretty much a broken LU environment waiting to be deleted….
The exquisite pieces are crafted from sterling silver and 18kt gold and are hallmarked at the Assay Office, Goldsmiths Hall, in London or Edinburgh. Speaking on the launch of the new boutique, Janet Jokat, Head of Retail, Links of London explained, ‘Over the last 17 years, valentine bracelet
A couple of points:
1. I don’t think C vs Python is going to make that much of a difference in performance, because any huge latency you’re going to see in a packaging system is going to be file operations. Remember how pkgadd was always changing /var/sadm/install/contents?
2. Throw me into the group that says LD_LIBRARY_PATH is bad. Maybe I could see having “pkg” being a wrapper script to make sure everything is okay, but to what end? You have to admit that in the end, your invocation of “pfexec pkg …” worked because the operating system handled the environment correctly.
Rebuild your custom Python build, and this time use “-R” on your library paths so you can properly isolate the two environments.
“2. Throw me into the group that says LD_LIBRARY_PATH is bad.”
LD_LIBRARY_PATH is bad. It’s evil. The correct way to link is -R$ORIGIN/../lib or -R$ORIGIN/../lib/64 for 32- and 64-bit payloads, respectively.
But that is not the point. The point is that mixing C and Python one gets an inconsistent salad – a mess.
That’s no engineering, no professionalism – that’s “HACK IT ‘TILL IT WORKS”, and by that point, one has already lost. In this particular case, all lose, because this affects everybody on the platform.
Another thing is maintenance. What is easier to maintain, an application written completely in one language, or an application written in multiple languages?
And further, if one is forced to rewrite parts of the application in C because Python can’t cut it – shouldn’t that serve as a warning to the developer that the tool he/she are using for the job are INADEQUATE?
Worse yet, Shawn doesn’t even come as far in his thinking to consider that possibility – he sets the breakpoint at “Python is the popular language to use, and as such the IPS project will benefit from many more developers.”
Wait a minute. Since when are language fads and their popularity criteria for an application as critical as a SOFTWARE MANAGEMENT SUBSYSTEM?
aaaaaaa
[url=http://www.3721.com/]uuuuuuuu[/url]
sports shoes
nike tn air max
There’s no reason IPS shouldn’t be faster. Python isn’t the bottleneck here, it’s the high number of disk I/Os and network round trips going on. SYS V packaging had its issues, but this has felt like a computer science experiment for way too long.
Jumpstart is needlessly complex compared to alternatives for Linux and a new package system that unifies automated installs seamlessly is a good goal.
All that said, the most ridiculous part of IPS is the fact that its repositories and metadata can’t be served by a regular ol’ HTTP server. Why a separate metadata server was determined to be necessary is beyond me. The complexity means there’s a dearth of mirrors, and even if you mirror locally you’re reliant on pkg.opensolaris.org being available for the metadata.
“Jumpstart is needlessly complex compared to alternatives for Linux and a new package system that unifies automated installs seamlessly is a good goal.”
I’ve just completed an eight-month project (as a consultant) for a company wishing to implement automated Linux installation, and even with expensive software from redhat, the underlying capabilities of Kickstart provisioning a PRIMITIVE compared to what JumpStart(TM) can be made to do, especially in combination with Flash(TM).
There is a reason WHY JumpStart(TM) has a steep learning curve, and it is because it can be made to do so much – completely automatically, for example with profiles and EASY automatic post-package deployment.
Meanwhile, beyond mere “add these RPMs to the system and create a RAID”, Kickstart is basically a primitive install-choice replicator/applicator application. It isn’t capable of making decisions on the fly (for example on architecture, partitioning, swap size), rather it is expected that the system administrator do all that hard work with scripts.
While that might be TOLERABLE for a system engineer, a contemporary sysadmin is incapable of designing such a complex provisioning system that will MAYBE come close to what JumpStart(TM) on Solaris can do.
So do not make the naive mistake of confusing a steep learning curve of JumpStart(TM) with lack of capability.
JumpStart(TM) is very powerful, and if the provisioning system is ARCHITECTED and IMPLEMENTED properly, JumpStart(TM) does serve as the foundation of a self-healing, self-provisioning network infrastructure.
oh cool, this information is really useful and definately is comment worthy! hehe. I’ll see if I can try to use some of this information for my own blog. Thanks!
We launched the 2010 latest and most fashionable True Religion jeans on our shop, welcome you to patronize.Here are many discounts for you.
Our aim is to provide high-quality products and excellent after-sales service.
http://www.jeanushop.com
Welcome to Moncler Down Jackets online shop.The 2010 latest and most fashionable Jackets,save up to 60% off. 7 days delivery to worldwide with free shipping.
Our aim is to provide customers high-quality products and excellent after-sales service.
http://www.monclerustore.com
Man proposes, God disposes.http://www.edfashionclothes.com/ ,Never be unduly elated by victory or depressed by defeat.Don’t paint the lily.
ed hardy clothing,and gucci shoes,all products are high quality but low price,welcome to http://www.uggjordanghd.com/.
Quality is better than quantity.http://www.handbags4buy.com/ A hedge between keeps friendship green.T
he cannot dwell there long.http://www.edhardy-buy.com/The spirit is willing, but the flesh is weak.
Nothing seek, nothing find http://www.uggboots4buy.com/.Birth is much, but breeding is more.
The time of life is short!to spend http://www.nikejordanshoes2sell.com/ that shortness basely,it would be too long.
On 11th June 2010, the FIFA 2010 World Cup in South Africa will begin.Why not celebrate your country qualifying by purchasing their world cup jersey? You could even purchase this for a loved one with their name and number printed on the football jersey!Our website will be the best place for you to come to find out all the latest news surrounding the 2010 World Cup in South Africa and where you should buy the latest international kits.
[[http://www.uniform-jersey.com]]
useful information
http://www.P90xhot.com
dont frustrate ,I think u doing well.
cheap mbt shoes
mbt
mbt shoes http://www.us-mbt.com
After a first phase of reforms agreed in 2008, developing countries have an around-44-percent share in the World Bank.
At the Pittsburgh G20 summit in September 2009 and the Istanbul Development Committee meeting in October 2009, the bank’s shareholders agreed to raise the voting rights to at least 47 percent for developing and transition countries.
“We were just pleased that we are getting close to reflecting China’s increasing share in world economy, and that is reflected in edited voting share,” World Bank President Robert Zoellick told Xinhua after the Development Committee meeting.
A 7.1-magnitude earthquake struck northwest China’s Qinghai Province on April 14. The earthquake had killed more than 2,200 people as of 5 p.m. Saturday, the rescue headquarters said late Saturday.
At present, an owners’ committee cannot be set up until all owners have moved into a community. The initial property management company is chosen by the developers and management fees are also decided without the owners’ involvement.
“In my opinion, it is good for us to be able to participate in the property management ahead of time,” said one of the Shijing Jiayuan owners. “But I am too young to put my name down for the committee. When I’m mature enough, I hope I can make a contribution to the owners’ committee to maintain the owners’ legitimate rights.”
great post
http://www.sellnikesbs.com
P90x .It really is not expensive if you factor in the cost
of a gym membership,P90x workout . The cost for P90X is
about three months of a paid gym membership but you get to
keep the program foreverP90x . You can try many of the
online sites, but it will be the same as buying from the
company or a Beachbody Coach. Make sure you are getting
original DVD’s. People are selling copies all over. The
problem is how long will they last, P90x workout ,and you
truly need the exercise and nutrition guide to even follow
the program. You can go to any site or you can go to
http://www.p90xmall.com and click on products. P90x dvd You can
order directly from the site,P90x dvd.
GHD straighteners was kmown as ghd flat iron, which was authorized online seller provides all kinds of hair straighteners,pink ghd,purple ghd,babyliss. By visiting ghd australia , you will find what you want and made yourself more beautiful.If you miss it ,you miss beauty.Buy a piece of ghd for yourself. Come and join us http://www.ghdoutlet-au.com/ to win the cheap ghd.
meizitang
zirantnag
lida daidaihua
online sale slimming capsule http://www.slimlida.com
good job,guy,it’s kind of u share this post with us which give useful information to us.God Job!I do hope u can publish more posts later.buy the way ,if you wanna buy cheap brand shoes online,you may visit our shoes store,i am sure you will find what u like.www.cheap-shoesonline.com
As we know, now GHD are loved by more and more people, which will save up to 45%.welcome to http://www.ghdoutlet-au.com/.
hi there, thanks for your great post, it help me so much! now i wonder if you linke the mbt shoes from my site: the url is http://www.thesuitshoes.com
A good post! http://www.offerreplicawatches.com.They sell Chanel Watches, Replica BRM Watches,offer Cartier Watches, Copy Ebel Watches, Replica Hublot Watches, Fake Tag Heuer Watches, wholesale Richard Mille Watches.
They also sell some very fashion watches, such as Montblanc Watches for sale,Chopard Watches, Dior Watches and Gucci Watches. Rolex watches are hot sell. If u are Armani Watches and Breitling Watches fan. You can still look at Omega Watches and Longines Watches, many new arrivals watches just come here.You can have a try on 58338312391273913. :[url=http://www.offerreplicawatches.com]www.offerreplicawatches.com[/url].
thank you for your post!http://www.mbtsonsale.com
I’m getting email from folks asking about the underlying problem here. [url=http://www.mydiscountjordanshoes.com/air-jordan-xv-c-79.html] air-jordan-xv
[/url]
thank you for your post!
HI! It is very good, useful things to share with everyone!
AVI to iPhone Converter is currently most ease-to-use converting tool which is able to Convert AVI to iPhone with professional
skills. What’s more, AVI to iPhone Converter has received a plenty of thanks from customers
since it turns up in the market.
http://www.avitoiphone.com
http://www.mp4toiphone.com
http://www.okayoutlet.com
Activists on ships raided by Israeli commandos cast doubt on Israel’s version of events that led to at least 10 deaths.
Supra shoes
http://www.suprashopping.com
http://www.idealhere.com/wholesale-Tattoo_c68
The 66-year-old Garrett has been the Trojans’ athletic director for 17 years, but he received caustic criticism for his handling of USC’s scandals. The school didn’t say whether Garrett retired or had been fired.
Both Garrett and Haden are former USC football players. Garrett won the Heisman Trophy in 1965, while Haden was the Trojans’ starting quarterback for three years in the 1970s under coach John McKay.
http://www.ed-hardy.cc/ed-hardy-clothing.html ed hardy clothes
http://www.ed-hardy.cc/ed-hardy-caps.html ed hardy hat
——— http://www.jorer.com ———-
JORER Inc. We are the best online dealer.about all kinds of Nike shox .run retailing and wholesale trade wordwidely for years. Free Shipping And Customs,Super Sale Off Retailing,With 1Week Delivery to your door.
http://www.mbtshoesaleonline.com
One day, a father and his little son were going home. At this age, the boy was interested in all kinds of things and was always asking questions. Now, he asked, “What’s the meaning of the word ‘Drunk’, dad?” “Well, my son,” his father replied, “look, there are standing two policemen. If I regard the two policemen as four then I am drunk.”
“But, dad,” the boy said, ” there’s only ONE policeman!”
http://www.coach-handbags-outlet.com/ coach bags outlet
sdfsf
sdfsdfsdf