(Jul 25 2005 11:36) Andy Bakun wrote:
> Well, swup is basing its decisions on the dependencies listed in the rpm
> file(s), so if mysql requires perl-5.8.5-5tr or even perl-5.8.5, rather
> than perl-5.8 or just perl, it's always going to update.  

Wrong. It's always going to update anyway, since perl is new and the
mysql package might benefit from that new perl.

> Unless swup is trying to be "too smart" by just looking at the package
> name and ignoring the version number and just adding the package name
> to the update list -- and voila, there's a new version, so update it
> too.  I don't know exactly how swup works in this regard, but what
> I've described seems overly complex.

Your description is not just overly complex, it's also just wrong. :)

1. Perl is updated
2. mysql requires /usr/bin/perl
3. mysql is scheduled for update as swup wants to keep your system as
   up-to-date as possible.

That's how simple it is. I'm considering rewriting this to:

1. Perl is updated
2. mysql requires /usr/bin/perl
3. swup tells the user that he should consider upgrading mysql as well.

Would that suit your needs better?

> Maybe, as far as mysql goes, support scripts should be moved into
> another RPM, and that RPM can require perl.  This will actually make it
> much more obvious what is being updated in the large package system that
> mysql is when there are multiple packages with interdependencies.  I
> think this would need to be thought out more.

Splitting support scripts into sub packages is something to consider since 
it will make it more transparent to the user why a package is updated.
However, it's a trade-off as well, since it makes it harder for the
users that expect the support scripts to be present after installing the
main package.

> A more realistic scenario is the recent zlib bug that required an
> update.  I wanted to update the zlib shared libraries as soon as
> possible (partly so I wouldn't forget/get into the habit of waiting to
> apply important updates, and partly because I wanted to update the
> minimal amount as outlined previously), but wait on updating mysql until
> a maintenance window.  Partitioned updates also decrease the required
> size of the maintenance window (especially if a problem crops up).
> 
> This is partly because I know that the zlib upgrade just adds new files,
> whereas the mysql upgrade stops and restarts mysql, and has postinstall
> scripts that create databases and warns about setting passwords and
> whatnot, and overall is a much more "complex" package.  (of course, I
> realize that in order to actually get the benefit of the zlib fixes, the
> services that use the zlib .so will have to be restarted anyway).

As mentioned before, I have no problem understanding why you want
to partition updates.

> The other thing is that I can say to the other sysadmins "make sure
> everything is updated on this machine EXCEPT mysql, we'll have the local
> mysql expert handle/be around during the upgrade in case anything goes
> wrong" (because I associate a "complex" package with increased
> "delicacy", prone to problems).  I suppose in that case I could put
> mysql in the exclude list in swup.conf, but then the swup.conf needs to
> be edited when you do want to actually have swup handle those packages
> (and then reverted when you're done) -- increases the chances of things
> going wrong because more files need to be touched.  

Wrong. when you want to include mysql, you run with '--ignore-filter'.
Edit the file once, smile forever.

> Also, adding something to the exclude list will have it be ignored in
> my cron job that sends me email saying something needs to be updated.

Sure about this? If I recall correctly, you will be notified that swup
skips those packages if update candidates are available.



> 
> > It might be that the behaviour you suggest with swup including as little
> > as possible when run with arguments, and running with no arguments
> > equals '.*', is the prefered behaviour. If we deside it is, then I can
> > change how swup does this soon enough. If --local-first does not work
> > the way I explained (remember that doing this was not the main reason
> > for --local-fist anyway), I will have to modify it to do that, since
> > having the option to do both is desired.
> 
> I will try it the next time I'm applying updates.
> 
> -- 
> Andy Bakun <[EMAIL PROTECTED]>
> 
> _______________________________________________
> tsl-discuss mailing list
> [email protected]
> http://lists.trustix.org/mailman/listinfo/tsl-discuss
> 

-- 
Christian H. Toldnes
Trustix Developer
_______________________________________________
tsl-discuss mailing list
[email protected]
http://lists.trustix.org/mailman/listinfo/tsl-discuss

Reply via email to