Authorization FAQ

A list of common questions and answers related to application authorization.

1. What’s the difference between authentication and authorization?

Authentication is the mechanism for verifying who a user is. A username and password, for instance, verify that you can log in to an account.

Authorization is the mechanism for controlling what a user can do. For instance: after you’ve logged in, do you have access to a particular private git repository on GitHub?

If you’re curious to learn more about the differences between the two, see our chapter on What is Authorization?

2. What is a permission?

Having a permission is having the ability to perform an action on a particular resource. For instance—you have permission to read Oso’s code on GitHub (we’re open source!), but you don’t have permission to push directly to our source (please open a PR instead). On the other hand, you don’t have permission to read or push to the Microsoft Windows Kernel source on GitHub.

To make it easy to grant permissions, systems usually bundle permissions into groups. Role-based access control and attribute-based access control are both ways to assign groups of permissions.

3. What is a role, and what is role-based access control (RBAC)?

Roles are a way to group permissions and assign those permissions to users. Nearly every app uses roles! You’ve probably seen User, Manager, and Admin roles in the wild. Users can have many permissions in a single app—you might have an Owner role for repositories you create, but the Member role in your employer’s repositories.

To learn more about RBAC, see Authorization Academy's chapter on Role-Based Access Control.

4. What is an attribute, and what is attribute-based access control (ABAC)?

Attribute-based access control (ABAC) is a way to control access to a resource depending on its attributes. For instance, the policy that “anybody can read a repository if it is marked public” depends on the repository’s “public” attribute. Almost anything can be considered an attribute! Useful attributes are often relationships between your app’s objects. “You can read an issue if you are a contributor of the issue’s parent repository” is an attribute that depends on parent-child relationships.

5. What is a relationship, and what is relationship-based access control (ReBAC)?

Relationship-based access control (ReBAC) means organizing permissions based on relationships between resources. For instance, allowing only the user who created a post to edit it. Relationships look very much like roles. You can use relationships as a handy mental model for how to implement authorization and how to communicate it to your users. Relationships include data ownership, parent-child relationships, groups, and hierarchies.

For more on ReBAC, see the Authorization Academy chapter on Relationship-Based Access Control.

6. What is an access control list (ACL)?

An access control list (an ACL, often pronounced “ackel”) is a list of permissions attached to a resource. For instance, a repository may have a specific list of contributors and their permissions: “Alice: pull, push; Bob: pull; Carlos: pull.” ACLs can also include relations: “user Alice has relation Owner to object Repository.”

7. What is Google Zanzibar?

Zanzibar is Google’s internal, low-level authorization service. It’s a centralized service that provides an authorization API for Google’s application teams. Google has published a detailed description of the architecture, so there are several open- and closed-source re-implementations. Zanzibar-like systems are heavyweight authorization tools with high setup costs. Whether you should use Zanzibar-alikes or not depends on your application architecture.

If you’re curious about the precise details, we’ve walked through building Zanzibar from scratch to show you how it works!

8. How should I implement authorization in my microservice architecture?

It's always best to build authorization around your existing infrastructure.

If you already maintain a very performant way of exchanging data, like gRPC, you can use that to share authorization data between services.

If not, the simplest approach is to put your authorization data (like roles) in a JSON Web Token. This approach works well, but tokens have size limits.

If you outgrow a single token, you can use a shared authorization database. If you'd rather not maintain your own shared database, use an authorization as a service provider.

For more on this topic, take a look at best practices for authorization in microservices.

9. What’s the best way to model my authorization logic?

Instead of scattering authorization throughout your code, it's best to move your authorization logic into a policy. That policy will define the resources you’re controlling access to and the rules governing that access. When you need to make an authorization decision, you'll use your policy to make that decision.

To start, define your policy using role-based authorization (RBAC). Roles will fit most of the cases you'll run into in an app.

When you want to write authorization logic that depends on the existing relationships in your application, like granting access to repositories if you can access the parent organization, you can mix in relationship-based access control ReBAC). Relationships will cover nearly every common authorization case.

If access to a resource depends on an attribute that isn’t a role or relationship—for example, an is_public flag—you can write authorization rules based on arbitrary attributes of your resources.

10. How should I implement authorization in GraphQL?

In GraphQL, build your authorization logic as close to the data as possible. Usually, this means putting your authorization in the GraphQL API itself.

Putting your authorization logic at GraphQL's data access layer is very close to the data itself. That's the best way to authorize read requests, but that code may not have enough context to deal with authorization for write requests.

If you have a small application with basic authorization requirements, build your authorization in GraphQL resolvers. In a resolver, you'll have the context you need—the actor, the action, and the resource—to write your authorization logic correctly. However, you won’t be able to reuse your authorization code between resolvers.

If you find yourself writing lots of authorization logic in each resolver, consider moving your authorization to a GraphQL directive. Using directives will let you reuse authorization code throughout your schema.

In distributed or federated GraphQL, there are a whole new set of problems! We cover many of them in our guide to authorization patterns in GraphQL.

If you want to connect with thousands of other like-minded developers working on authorization, join the Oso community Slack! And if you have any questions about how to model your permissions, or just authorization in general, schedule a 1x1 with an Oso engineer.

Level up your authorization knowledge

Explore our docs

Access guides, example policies, and authorization best practies.
Read Oso docs

Read Authorization Academy

A series of technical guides for building application authorization.
See all chapters

Learn about Oso

A managed solution for centralizing authorization logic and protecting application data.
Dive into Oso