Open Policy Agent, Part II - Developing Policies
Go behind the scenes of Open Policy Agent.
Join the DZone community and get the full member experience.Join For Free
In the previous part of the series, we explored Open Policy Agent and implemented an ACL-based access control for our application. In this entry, I am going to share with you some of the discoveries that I made while evaluating Open Policy Agent in regards to policy design and development.
After evaluating policy rules, OPA returns a result of the policy decision to your application. This result is a JSON structure. Based on your requirements, this JSON structure can contain a single member holding a true or false (authorized/not authorized) value. However, you can create policies whose evaluation results in an arbitrarily complex JSON document. For example, OPA can return a list of nodes on which Kubernetes should schedule a workload.
In microservice applications, OAuth 2.0 is a rather popular authorization framework used to secure services' APIs. It typically leverages JSON Web Tokens (JWT) to convey claims. OPA comes with built-in functions that can decode the token and validate its signature and expiration time. Furthermore, your policy rules can make decisions based on the claims included in the token. Just forward the token as an input to OPA and offload the entire token processing from your application!
You may also like: Everything About HTTPS and SSL (Java).
OPA makes policy decisions based on the data stored in memory. In the case of large data sets, replicating all the data in memory can be impractical. While evaluating policy rules, is OPA able to reach out to an external data store to get additional data for decision making? For example, send a query to LDAP to grab additional attributes or look up data in an SQL database?
Based on my research, I think there are two possible approaches for leveraging external data sources in OPA. First, there is a built-in HTTP function that can fetch data from external HTTP services during policy evaluation. Second, you can leverage Partial Evaluation, as described in this blog post.
While partially evaluating policies, OPA doesn't return a complete policy decision, but, instead, it returns a set of conditions. It is left to you to translate this set of conditions into a query appropriate for your datastore and execute the query in order to obtain the final policy decision. Note that regardless of which approach you choose, reaching out to external data stores will have negative impact on latency and reliability of your solution. Caching data in OPA's memory is always a better option assuming that it suits your use case.
If you have raw data that would be difficult to write a policy against, you can pre-process that data into a form that better suits the policy writing before importing it into OPA. Moreover, if you have multiple sources of data, e.g. data from LDAP and Active Directory, you can merge them outside of OPA and load the merged form into OPA.
RBAC (Role-Based Access Control) and ABAC (Attribute-Based Access Control) are two frequently used policy models. Are you wondering if you can implement them using OPA? Of course you can! Follow these two links to find sample implementations of RBAC and ABAC.
Hierarchical group permissions are commonly found in practice, e.g. parent group permissions are a superset of child group permissions. These models can be elegantly described using recursive rules. However, at the time of this writing, OPA doesn't support recursion in policies.
While learning the OPA's Rego language, I appreciated the built-in interactive shell (REPL) that I could use to write and test my policies instantly. Just type
opa run, and you are good to go. Alternatively, you can go online and utilize the Rego Playground too.
If you are dealing with complex policies, how do you ensure that you implemented your policies correctly? OPA allows you to write test cases that you can run against your policies. You can use data mocking and calculate test coverage. See also the command
Is the evaluation of your policies too slow? OPA comes with a profiler to report on time spent on evaluating policy expressions. See also the
opa eval command.
OPA comes with a formatting tool
opa fmt to format Rego policy files. You don't need to fight battles with other developers about how the Rego files should be formatted!
OPA is a relatively new project; however, additional tooling and integrations with OPA are showing up quickly. If you like to use Visual Studio Code, there is a feature-rich VS Code plugin available for you. Rego syntax highlighting is available for several other editors like VIM, Atom, and TextMate.
In this blog post, I shared with you several tips and approaches for how to design policies in Open Policy Agent. In the final article in the series, we will focus on how you can integrate Open Policy Agent with your application.
If you have any comments or questions, please use the comment section below. I look forward to hearing from you.
Published at DZone with permission of Ales Nosek. See the original article here.
Opinions expressed by DZone contributors are their own.
Top 10 Engineering KPIs Technical Leaders Should Know
How To Approach Java, Databases, and SQL [Video]
Effective Java Collection Framework: Best Practices and Tips