[CentOS] rpm --freshen issue (was: Re: Caught between a Red Hat and a CentOS)

Wed Oct 21 14:03:46 UTC 2009
Barry Brimer <lists at brimer.org>


On Wed, 21 Oct 2009, ken wrote:

>
> On 10/20/2009 12:15 PM Benjamin Franz wrote:
>> ken wrote:
>>> Okay, here's one. Maybe someone here can figure it out.
>>> Upgrading from 4.5 to 4.5.  From a 4.6 ISO I copied all the RPMs into a
>>> directory... let's call it c:/install :).   Now the oracle dba has
>>> strict parameters on what versions can be installed and which can't.
>>> The rpms in c:/install meet those requirements.  In addition, since this
>>> is a production machine, it can be down at most for one day.  So all I
>>> want to do is upgrade what's currently on the system.  Moreover, if
>>> something horks, I want two chances to back out (the second being asking
>>> the backup guy to put the system back to yesterday).  The command to do
>>> this would be
>>>
>>> rpm --freshen --repackage *
>>>
>>> run in that crazy c:/install directory (or what the redhat guy called, a
>>> "folder").  This command runs fine for one file which has no
>>> dependencies (i.e., change '*' to a specific rpm).  It also upgrades
>>> three or four co-dependent rpms if they're narrowly specified.  But if
>>> the file/rpm spec is '*', rpm complains about two missing dependencies
>>> and stops.
>>>
>>> Yeah, this directory contains 1507 rpms (IIRC)... which is a lot, but it
>>> should still work.  This is Linux, after all.  And there's plenty enough
>>> memory and cpu to handle it.
>>>
>>
>> Running
>>
>> rpm --freshen --repackage *
>>
>> for 1500+ rpms  probably exceeds the maximum character length for some
>> part of the system after expansion of the '*'  by the shell.
>
> That was my first suspicion too.  The redhat tech didn't bring that up
> though.  (That doesn't mean I'm going to ignore that as a possible
> workaround; the original conversation here was about tech support per
> se.  Of course I'm still seeking ways to do the job.  And so thanks for
> the suggestion.)
>
> I, too, recall reading some years back about a bash line length limit.
> Back then, a long time ago, it was 2048 characters.  So I ran "echo *"
> in that same install/ directory and the output included all 1507 files.
> So the problem's not with a bash command line length limit, but still
> pointing to the "rpm" command.
>
>
>
>>
>> Try breaking it up into smaller chunks (say two or three hundred at a
>> time). You can match subsets of the files using shell expansions like
>>
>> rpm --freshen --repackage [a-g]*
>>
>> and tweak the line for any dependency complaints manually.
>
> This solution occurred to me also.  And right now it's a top contender
> (along with another I'll mention shortly).  If the job environment were
> different, I'd go with it.  But my boss is making me jump through a lot
> of hoops for this project.  This upgrade from v.4.5 to v.4.6 needs to
> happen in a single, specified day *and* my boss needs to know how long
> it will take me to accomplish, this so the Oracle dba knows when he can
> start to on what he's got to do for this upgrade.  And I have at most
> fifteen hours (i.e., two working days) to come up with this fool-proof
> plan.  Plus, I don't have a test box to try things out on.  But I've had
> to do trickier stuff than this in the past with not dissimilar time
> constraints, so though I should be taking extra boxers to work, I'm not
> (yet).
>
> So what I was thinking of doing is scripting the solution you suggest
> above.  But then, if I'm going to script something, I might as well
> write a script that will take on the entire task wholistically.  I mean
> something like this:
>
> ls -1 install/ > what-to-upgrade.list  # create package list
> while read package | {upgrade package}  #just quasi-code here. Loop.
> if {there's nothing to upgrade}
>  remove pkg from what-to-upgrade.list
>  log this
>  continue
> fi
> if {there are dependencies}
> then for {each dependency} {upgrade package}  # yep, recursion
> fi
> else [upgrade package}         # simplest case, just upgrade one pkg
>
> The {upgrade package} function would be fairly simple (I think):
> - Find the correct package in the install/ directory (containing the
> RPMs for v.4.6).
> - Upgrade the 4.5 package with that correct 4.6 package.
> - Confirm that the 4.6 is installed.
> - Remove that package name from what-to-upgrade.list
> - Log that this package has been upgraded.
>
> I already see some bogus stuff here, but I'm writing this on the fly.
> Point is, it seems do-able, and probably within the time constraints.
> And then, what are the alternatives?
>
> One, suggested by the redhat tech (about whom there's more news...
> later), is to use up2date.  I read the manpage on it and it's pretty
> vague.  I'm sure I have, but I don't recall using it before, so I can't
> fill in the details which the manpage lacks.  Lastly, I don't see a way
> to test up2date to see if it will work within my (dba's) specific
> parameters.

If you add --aid on the end of your rpm command, and you are in the 
directory with the rpms, it should solve any dependency issues for you 
provided that the rpms that are needed are in the local directory.  If a 
pendency is resolved by something that you later try to upgrade i *think* 
rpm will just tell you that that particular package is already installed. 
I could be wrong.  Either way .. try the --aid.  Also, don't forget that 
you can add the --test flag to rpm to find out if rpm will work or not.

Hope this helps,
Barry