Обсуждение: AW: Proposal for enhancements of privilege system

> > Imho this is an area where it does make sense to look at what other
> > db's do, because it makes the toolwriters life so much easier if pg
> > behaves like some other common db.
> 
> The defined interface to the privilege system is GRANT, REVOKE, and
> "access denied" (and a couple of INFORMATION_SCHEMA views, 
> eventually).
> I don't see how other db's play into this.

Of course the grant revoke is the same. But administrative tools usually
allow you to dump schema, all rights, triggers ... for an object and thus
need 
access to the system tables containing the grants.

> 
> > Other db's usually use a char array for priaction and don't have
> > priisgrantable, but code it into priaction. Or they use a bitfield.
> > This has the advantage of only producing one row per table.
> 
> That's the price I'm willing to pay for abstraction, 
> extensibility, and
> verifyability. But I'm open for better ideas.

Imho this is an area that is extremly sensitive to performance,
the rights have to be checked for each access.

Andreas

On Tue, 30 May 2000, Zeugswetter Andreas SB wrote:

> > 
> > > Other db's usually use a char array for priaction and don't have
> > > priisgrantable, but code it into priaction. Or they use a bitfield.
> > > This has the advantage of only producing one row per table.
> > 
> > That's the price I'm willing to pay for abstraction, 
> > extensibility, and
> > verifyability. But I'm open for better ideas.
> 
> Imho this is an area that is extremly sensitive to performance,
> the rights have to be checked for each access.
Yes, but I believe that Peter's idea is good. System tables are used for
each access not only for ACL, and performance problem is a problem for
system cache not primary for privilege system.
I look forward set privilege for columns and functions. Large multiuser
projects need it.
                        Karel

On Tue, 30 May 2000, Zeugswetter Andreas SB wrote:

> Of course the grant revoke is the same. But administrative tools
> usually allow you to dump schema, all rights, triggers ... for an
> object and thus need access to the system tables containing the
> grants.

That's what you use the information schema views for. Also, of course,
we're light years away from having anything like a portable pg_dump.

> Imho this is an area that is extremly sensitive to performance, the
> rights have to be checked for each access.

But using some sort of arrays is going to make it slower in any case since
you can't use indexes on those.


-- 
Peter Eisentraut                  Sernanders väg 10:115
peter_e@gmx.net                   75262 Uppsala
http://yi.org/peter-e/            Sweden