On Mon, 12 Jan 2004, Michal Pasternak <michal@pasternak.w.lub.pl> wrote:
> Martin Husemann [Mon, Jan 12, 2004 at 03:25:59PM +0100]:
> > I would imediately drop any package system that
> > fidles with my database content automagically in such situations.
> 
> What's the difference between pg_dump performed automatically and
> performed by hand?

You are not talking like a DBA. 
You are certainly not talking like a DBA who's responsible for a
production database.  
You are absolutely, positively not talking like a DBA responsible for a
production database with 10 GB of valuable data that is both widely used
and highly visible in the firm.  

Forget high availability, backup machines, fallback mechanisms.  If I'm a
DBA, and my sysadmin came to me and said he'd installed my new database as
requested, I'd expect to find my old one intact and unscathed.  The
probability that his pkgsrc script would correctly upgrade the data are so
close to zero, it's laughable.  To pick just one impossible problem:  How
is pkgsrc going to guess what devices to use?  

As Greg says, it would be nice if the authors took more account of data
migration issues.  But IMHO the only responsible path open to pkgsrc for
RDBMSs is to install the next version side-by-side with the old.  The DBA
will load the new database; the users will test it; it will go online. 
Then, after some days or weeks, the old one will be decommissioned.  

For anything that matters, running the old and new systems simultaneously
is what's going to happen.  The only question here is, how much will
pkgsrc do to make that easy?  

Respectfully yours, 

--jkl