On Fri, Aug 28, 2009 at 1:19 AM, Robert Milkowski<[email protected]> wrote:
> Martin Bochnig wrote:
>>
>> On Fri, Aug 28, 2009 at 12:53 AM, Philip Brown<[email protected]> wrote:
>>
>>>
>>> A long while ago, I brought up the unpopular point, that the real problem
>>> with patching, is not "SVR4 patching is broken", but "Sun's patching
>>> teams
>>> make poor choices about how to structure patches(and group packages)".
>>>
>>> I was basically shouted down, for daring to criticize Sun, or the
>>> direction
>>> IPS was heading in.
>>>
>>> No one bothered to confirm or deny whether what I said, was *true* or
>>> not.
>>>
>>> Here's some hard data to prove the truth of the point I was trying to
>>> make.
>>>
>>> My reason in bringing this up again, in this forum, is to ask the IPS
>>> team
>>> how they plan to make IPS magically fix this sort of problem, when it is
>>> caused by the sun packaging/patch teams' PROCESSES, not lack of
>>> technology.
>>>
>>> My guess is the reply will be , "disk space is cheap, so just do full
>>> installs everywhere". But I'll state ahead of time that in my opinion,
>>> this
>>> is not an appropriate response to customers
>>>
>>>  * * * * * * *
>>>
>>> The real-life problem:
>>> (and note that, being real-life, the problem description is somewhat LONG
>>> :-)
>>>
>>>
>>> I'm looking to patch a bunch of solaris 10 sparc servers.
>>> They have a "server oriented install" set of packages. To give some idea
>>> of
>>> the scope of the install:
>>>
>>> # pkginfo|grep SUNW|wc -l
>>>    592
>>> # (whereas a full install, would have 2000+ packages)
>>>
>>> They have SUNWxim installed. Even though a local X server is not
>>> installed.
>>>
>>> Because, being a university, our customers may very well wish to run
>>> UTF-8
>>> locales for some programs, remotely.
>>> That means we have things like
>>>  system      SUNWeuxwe UTF-8 X Window Environment
>>> installed. Which require SUNWxim.
>>>
>>> (I also vaguely recall that some java things complain without SUNWxim
>>> installed, but i could be mistaken)
>>>
>>> So far so good.. but now I want to install a recommended patch cluster.
>>> Which includes patch 121975-01
>>>
>>>
>>> patching 121975-01 FAILS. Because it depends on 121975-01.
>>> patch 121975-01 fails,  because SUNWdtdte is not installed.
>>>
>>> But we dont WANT SUNWdtdte installed! That includes stuff like dtlogin,
>>> and
>>> all kinds of other cruft that dont belong on our servers!
>>>
>>>
>>> SUNWxim does not depend on SUNWdtdte. Therefore, a patch for SUNWxim, has
>>> no
>>> business pulling in an implied dependancy on SUNWdtdte either.
>>> But it does, and the patch fails without it.
>>>
>>> That fairly tightly fits my own personal definition of
>>>  "broken patch creation policies/procedures".
>>>
>>> So, how is switching to the "better technology" of IPS, going to solve
>>> this
>>> problem of broken patch creation process inside of Sun?
>>>
>>
>>
>>
>> Hey Phil: Very easy! As no "--no-deps" is "supported", your scenario
>> would not even be possible anymore, not at all. That's the solution
>> (really).
>>
>
> As Shawn and Bart posted today --nodeps will be supported via 'pkgrecv
> --nodeps ...; pkg install' which is fair enough and does make sense. It is
> actually supported even right now - it is just that it is too ceymbersome in
> a current form for an average sysadmin. But Shawn expressed clearly they are
> open to suggestion to make it easier and even proposed a solution which
> sounds really good.
>
> Philip - the answer is relatively easy - as there are no patches in IPS. If
> there will be a bug fix for SUNWxim it will be provided in form of a new
> SUNWxim package (different timestamp or version) so you will just upgrade a
> specific package. Even better - as not necessarily all files in a new
> package will be different pkg will (it already does) fetch only files which
> changed making an upgrade even quicker. Providing bug-fixes in form of a new
> package version has been common on many other platforms like Linux and has
> been working really well for them. The only main issue has been lack of
> reliable rollback with such an approach except for installing old package
> again which not always is the best option. In opensolaris thanks to BE you
> can actually upgrade your system (or selected packages) on a new BE which is
> a clone of current OS and reboot to it when you ready - if you want a 100%
> (kind of) guarantee to rollback the change you reboot to old BE - I really
> love it and I've been using it for some time now.
>
>
>
> --
> Robert Milkowski
> http://milek.blogspot.com



Ok Robert, Shawn, Dave and Bart.
I'm more or less convinced now.

Congrats to IPS)


rgds.
_______________________________________________
pkg-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/pkg-discuss

Reply via email to