1

I want to build a multi-tenant solution with global and custom roles.
The application's authorization will be built based on permissions.
Then there may be defined a role that combines multiple permissions.

I want to provide several predefined global roles that can be used by everybody.
Additionally, I want to provide tenants with the ability to define their own custom roles.

At the moment, I have the following draft of the schema design (postgres):

create table tenants (
  id uuid primary key,
  ...
);

create table permissions (
  code character varying(30) primary key
);

create table roles (
  id  uuid primary key,
  tenant_id  uuid null references tenants,
  ...
);

create unique index on roles (id, tenant_id) where tenant_id is not null;

create table role_permissions (
  role_id uuid not null references roles,
  permission_id character varying(30) not null references permissions,
  unique (role_id, permission_id)
);

create table users (
  id uuid not null,
  tenant_id uuid not null references tenants,
  ...
  primary key (id, tenant_id)
);

create table user_roles (
  tenant_id uuid not null,
  user_id uuid not null,
  role_tenant_id uuid null,
  role_id uuid not null references roles,
  foreign key (user_id, tenant_id) references users (id, tenant_id),
  check (customer_id = role_tenant_id or role_tenant_id is null)
);

In this schema I'm not able to correctly reference roles from user_roles.

Is it possible to implement such constraint in postgres?

Azamat Zhurtbayev
  • 233
  • 1
  • 9

1 Answers1

2

I assume that references to the customer table and customer_id starting with the definition of your users table really mean to refer to tenant and tenant_id.

At some point you need to trust your code to be correct. If that is not good enough for you, and you must have constraints, then this is what I would do:


create or replace function user_role_check(_user_id uuid, _role_id uuid)
  returns boolean as $$
  select count(*) = 1 
    from roles r
         join users u
           on u.tenant_id = r.tenant_id
   where u.id = _user_id
     and r.id = _role_id;
$$ language sql;

create table user_roles (
  id uuid not null primary key,
  user_id uuid references users(id),
  role_id uuid references roles(id),
  check (user_role_check(user_id, role_id)),
  unique (user_id, role_id)
);

Otherwise you are stuck duplicating tenant_id into user_roles.

Mike Organek
  • 11,647
  • 3
  • 11
  • 26
  • Thank you for your answer. Yes, you're right about field names. If there are no alternative solution have been provided after a few days, I will mark your answer as best one. – Azamat Zhurtbayev Sep 12 '20 at 17:39
  • Ok, I've implemented a check as you had suggested. Not exactly as you wrote but still the approach is correct and everything works in the right way. However I am not sure about performance side of this solution. Anyway there's no alternative solutions so far. So, I'm marking your answer as the best one. Thank you! – Azamat Zhurtbayev Sep 13 '20 at 09:09