Re: Re: [COMMITTERS] pgsql/contrib/pg_dumpaccounts (Makefile README pg_dumpaccounts.sh) - Mailing list pgsql-hackers

Well, here in relatively minor form is the First Example of a Great 
Bridge Priority (which Tom, Bruce, and Jan have all predicted would 
come... ;-)

Our feeling is that DBAs will want to have the ability to backup user 
and group info, which you currently can't do with pg_dump.  You *can* do 
it with pg_dumpall - but only if you dump every database you've got at 
the same time.  Picture a professional environment where you might have 
many different databases running 24/7 - and doing a pg_dumpall across 
all of them at once just isn't practical.  Most DBAs would prefer to 
stagger their regular backups in such an environment, one database at a 
time.  Indeed, those backups are often on fixed schedules, at different 
times, for real business reasons.  And if you do that, you can't backup 
the aforementioned system catalogs.

That's what this pg_dumpaccounts is designed to do.  As you've seen, 
it's very simple - it does the same COPY stuff that pg_dumpall does 
before calling pg_dump, just without the pg_dump.  It's an inelegant 
solution, and shame on us for not catching the problem sooner.  But it 
*is* a problem, albeit perhaps one that current PostgreSQL users haven't 
run into yet.  We're concerned that people might have a false sense of 
security with pg_dump - that they might think if they backup one 
database, they're able to do a full restore.  They're not.  And like I 
said, there are situations when pg_dumpall isn't the appropriate solution.

We recognize this is a temporary hack - and fully expect it to go away 
in 7.1  We actually think that the final solution might be more 
appropriate in pg_dump itself than pg_dumpall, but that's obviously a 
much more breakable proposition (hence the separate utility).

I understand everyone's hesitation about adding a new utility this late 
in the process - and we're happy to be overruled on that (even if it's a 
discrete piece of code that wouldn't affect anything else...)  I'm not 
wild about putting it in /contrib, but if that's what everyone wants to 
do, ok.

Have we adequately explained the need for this?  Or do people think it's 
not necessary?

If it *is* necessary (or at least worthwhile), is it the consensus of 
the -hackers community that it go in /contrib?

Thanks,
Ned


Tom Lane wrote:

> Bruce Momjian <pgman@candle.pha.pa.us> writes:
> 
>> I think the issue is that we don't want to risk breaking pg_dumpall in a
>> minor release.
> 
> No we don't, but I agree with Peter that pg_dumpall is the place for
> this feature in the long run.  A separate contrib script is not going
> to get maintained.
> 
> What I want to know is why we are adding features at all in a minor
> release.  Especially 24 or so hours before release, when there is
> certainly no time for any testing worthy of the name.  Contrib or no
> contrib, I think this is a bad idea and a bad precedent.
> 
>             regards, tom lane
> 
> 


-- 
----------------------------------------------------
Ned Lilly                     e: ned@greatbridge.com
Vice President                w: www.greatbridge.com
Evangelism / Hacker Relations        v: 757.233.5523
Great Bridge, LLC                    f: 757.233.5555