Thread: Persistent identifiers for Postgres users

Persistent identifiers for Postgres users

From
Peter Geoghegan
Date:
Hello,

I maintain an app where database users correspond to actual users,
with privileges granted or denied to each. At the moment, records that
each user creates are identified as such by a text column that has a
default value of session_user(). I don't need to tell you that this is
suboptimal, because db users (as far as I'm aware) lack persistent
identifiers - names may change, users may be dropped, etc. Also, there
is no way that I am aware of to fake row level privileges by adding a
...AND id NOT IN (SELECT forbidden_department FROM user_priveleges
WHERE user_id = current_user_id() ) to relevant queries . Actually,
that approach is probably preferable to actual row level privileges,
as it allows me to deny access based on a domain-level concept,
departments.

Am I correct in my belief that postgres users lack persistent identifiers?

I believe that some other similar systems implement their own users
and privileges systems to achieve this, but I hesitate to do that. I
also hesitate to assume that the DB user name will never change, and
go ahead and use session_user() in lieu of a real persistent
identifier.

Regards,
Peter Geoghegan

Re: Persistent identifiers for Postgres users

From
Alvaro Herrera
Date:
Peter Geoghegan escribió:
> Hello,
>
> I maintain an app where database users correspond to actual users,
> with privileges granted or denied to each. At the moment, records that
> each user creates are identified as such by a text column that has a
> default value of session_user(). I don't need to tell you that this is
> suboptimal, because db users (as far as I'm aware) lack persistent
> identifiers - names may change, users may be dropped, etc. Also, there
> is no way that I am aware of to fake row level privileges by adding a
> ...AND id NOT IN (SELECT forbidden_department FROM user_priveleges
> WHERE user_id = current_user_id() ) to relevant queries . Actually,
> that approach is probably preferable to actual row level privileges,
> as it allows me to deny access based on a domain-level concept,
> departments.

You could use OIDs as identifiers for roles instead of names, but of
course you don't have any way to know that one of them is dropped.

--
Alvaro Herrera                                http://www.CommandPrompt.com/
The PostgreSQL Company - Command Prompt, Inc.