Thread: SET ROLE and search_path

Hi,

 

I’m trying to implement a PostgreSQL multi-tenant database that will be accessed by a Web Application.

The users that will login will belong to different companies and a schema was created in the database for each company.

However, I would like the Web Application to connect with a single Postgres login.

Let’s say that I have 2 companies : comp1 and comp2 with their respective schema (comp1 / comp2).

Then, the web application connects with web_app login which has been granted comp1 and comp2 roles….

Depending on the user connecting to the application, I would like to use SET ROLE comp1 / SET ROLE comp2 in order to get access to the relevant data only.

However, it seems that SET ROLE does not change the search_path (which is different for comp1 and comp2).

 

Is there any way to change the search_path in an easy way (in a procedure) after SET ROLE has been executed.

Am I missing anything with SET ROLE.

When search_path contains “$user”, does it refer to session_user or current_user ?

 

Thanks for any advice

 

Patrick

Does your role definition assign a search_path?

create role comp1;
alter role comp1 set search_path=comp1,base,public;

Every re-use of the postgres connection must start by resetting the search_path.  I find it easier to log in as comp1.  Some jiggery-pokery involved in passwords but no one in company #1 needs to know the user name let alone password.

I have this exact setup, and I use roles / schema names that match so the $user var works with the search path when I set role as my application user.

> When search_path contains “$user”, does it refer to session_user or current_user ?

It uses current_user, not session_user. Works perfectly with set_role for me.

Am I (again) alone in finding this a bit hokey?  That a user name just happens to be a schema name ...

---

Rob Sargent <robjsargent@gmail.com> writes:
> Am I (again) alone in finding this a bit hokey?  That a user name just 
> happens to be a schema name ...

That's actually strongly encouraged by the SQL spec, if memory serves.

            regards, tom lane

Greetings,

* Tom Lane (tgl@sss.pgh.pa.us) wrote:
> Rob Sargent <robjsargent@gmail.com> writes:
> > Am I (again) alone in finding this a bit hokey?  That a user name just
> > happens to be a schema name ...
>
> That's actually strongly encouraged by the SQL spec, if memory serves.

... and all-but-required by some other database systems.  Doesn't make
it a good idea tho, imv.

Thanks,

Stephen