Policy Service

The policy service provides REST API to evaluate/execute OPA policies written in the Rego language. The policy engine is extended with custom functions available for use through the Rego runtime during policy execution. The service also provides endpoints for exporting and importing policy bundles, for subscribing for policy changes and endpoints for policy administration (e.g. listing policies, lock/unlock specific policy).

It is developed using the Goa v3 framework and uses the Go OPA framework as a library.

A helper script named goagen.sh can be found inside the root directory of the service. It can be used to generate the transport layer code from the Goa DSL definitions in the design directory. The script should be executed everytime the design definitions are updated. It also generates updated OpenAPI documentation from the DSL.

OpenAPI documentation

In the local docker-compose environment, the Swagger URL is available at http://localhost:8081/swagger-ui/

High-level Overview

flowchart LR
	A([client]) -- HTTP --> B[Policy API]
	subgraph policy
		B --> C[(policies DB)]
	end
	C --sync--- D[Git Server]

Loading

Configuration

The Policy service is configured using the Configuration File. All configurations are expected as Environment variables specified in the configuration file. For managing the configuration data from ENV variables, envconfig library is used.

Policy Evaluation

The policy service exposes HTTP endpoints to evaluate/execute policies. The endpoint interface is conformant to the TSA requirements document.

To evaluate a policy a GET or POST request is sent to the evaluation URL. The example URL below is given for the local docker-compose environment. The host and port parts will be different for the different environments.

# URL with example policy repository, group, name and version
http://localhost:8081/policy/policies/xfsc/didresolve/1.0/evaluation

# URL with parameter placeholders
http://localhost:8081/policy/{repository}/{group}/{policy}/{version}/evaluation

There are four parameters in the URL specifying which exact policy should be evaluated - repository, group, policy and version. These parameters are also important during policy development (see below) as group and policy must be used as package name inside the policy source code file.

The body of the POST request can be empty, but if it's not empty, it must be JSON. It is passed directly to the policy execution runtime. Inside the policy it is accessed with the global variable name input. For example, if you pass to the evaluation endpoint the following JSON, it will be accessible by input.message:

{
  "message": "hello world"
}

Here is a complete example CURL request:

curl -X POST http://localhost:8081/policy/policies/xfsc/didresolve/1.0/evaluation -d '{"message":"hello world"}'

Note: If the version or any other subpath is missing in the folder structure (e.g. in combination with the repo feature), the service appends a "temp" group, which results in another structure of the URL. In this case the service fires "result empty" responses, because the package name must be corrected as well to the temp path.

Policy output JSON schema validation

The policy service exposes HTTP endpoint to validate the output of the policy. It uses JSON Schema to validate the JSON schema of the output of the policy. To execute the validation procedure, a new HTTP URL is automatically generated.

Evaluation URL pattern: {service_addr}/{repo}/{group}/{policyname}/{version}/evaluation
Validation URL pattern: {service_addr}/{repo}/{group}/{policyname}/{version}/validation

In order to use the validation endpoint, output-schema.json file following the JSON Schema specification must be present in the policy repository directory. For more information on policy development refer to: Policy Development

Policy Locking

The service exposes HTTP endpoints to lock and unlock policies. Locking a policy means that it's not allowed for evaluation (execution). Unlocking a policy allows its evaluation/execution to proceed normally.

Lock a policy with POST request:

curl -X POST http://localhost:8081/policy/policies/xfsc/didresolve/1.0/lock

Unlock a policy with DELETE request:

curl -X DELETE http://localhost:8081/policy/policies/xfsc/didresolve/1.0/lock

Policy Bundles

A policy bundle contains a Policy source code, static data, configuration and some additional meta-info necessary for verifying the origin of the policy (who signed the bundle).

The service allows to export a policy bundle by making HTTP request against a policy URL of the same format as is used for evaluation, but instead of using the evaluation word at the end, you replace it with export. That will pack the policy as ZIP bundle, and the ZIP bundle will be returned to the client.

wget http://localhost:8081/repository/policy/xfsc/didresolve/1.0/export

See here for more detailed overview of policy bundles export/import.

Policy Storage

Policies (rego source code and metadata) are stored in a storage, which is an interface and different implementations could be used. You can check the interface here.

Currently, there are two implementations of the storage interface:

• MongoDB
• Memory

Matrix for storage feature availability:

Feature	MongoDB	Memory
Policy Lock/Unlock	Yes	Yes*
Change Notifications	Yes	Yes*
Storage extension functions	Yes	Yes*
Automatic synchronization	Yes	N/A
Bundle import/export	Yes	Yes*

* Functionality is available only for the current instance of the policy service. Synchronization between instances of the policy service is not available.

In order to use another storage technology, one should implement the Storage interface.

flowchart LR
	A[Policy\nDeveloper] --git push/merge--> B[Git branch]
	subgraph Git Server
		B --> C[example/1.0/policy.rego]
		B --> D[example/2.0/policy.rego]
		B --> G[example/3.0/policy.rego]
	end
	C --> E[Sync]
	D --> E[Sync]
	G --> E[Sync]
	E --update--> F
	subgraph policy service
		F[(policies DB)]
	end

Loading

Subscribe for Policy Changes

The service allows external clients to subscribe for policy changes. A change event happens when policy source code, static data, export configuration or any state-related data is updated. Two subscription mechanisms are supported.

Cloud Events

When policy change event happens, a cloud event is sent using the CloudEvents Go SDK. It uses standard format for encoding/decoding events and typically uses a message queue under the hood for delivery of events. Our implementation is using NATS, but any broker supporting the CloudEvents specification should work.

Web Hooks

This mechanism allow clients to express interest in change events for a specific policy and when such change event is triggered, the policy service shall notify the client. A client subscribes for policy change by making an HTTP request to a policy endpoint with the suffix notifychage, for example:

$ curl https://mysvc.com/policy/policies/example/policyname/1.0/notifychange 
-d '{"webhook_url":"https://url-to-call-when-policy-changes.com"}'

This request creates a record in storage, so the Policy service will know which webhook URL to call when a policy change event happens.

The event type that will be received by the subscriber is per policy and is sent as JSON structure like:

{
  "repository": "policies",
  "group": "example",
  "name" : "mypolicy",
  "version": "1.0"
}

It is defined in notify.go

Policy Admin API

The API allows to inspect the internal state of the policies without requiring direct access to the policy database. This could be useful for debugging or checking what is the current source code or static data for a given policy. The endpoint is called with HTTP GET requests and supports basic filtering and expansion via query params:

GET /v1/policies?policyName=hello&locked=true&rego=true&data=true&dataConfig=true

All query parameters are optional.

Policy Development

• Policy Extensions Functions

Policies are written in the Rego language. Please refer to the OPA documentation for detailed overview of Rego and OPA capabilities.

Some conventions must be followed when writing policies.

1. The filename of the policy must follow rules for the naming and directory structure: the group, policy name and version are directories inside the Git repo and policy file must be named policy.rego. For example: /xfsc/example/1.0/policy.rego.
2. In the same directory there could be a data file containing static JSON, which is automatically available for use during policy evaluation by using the data variable. The file must be named data.json. Example: /xfsc/example/1.0/data.json
3. In the same directory there could be a configuration file containing information for getting static JSON data from external URL. The file must be named data-config.json. Example: /xfsc/example/1.0/data-config.json

Note that there should only be one of the two files data.json or data-config.json in the same directory. If both files exist in the same directory tha data from the data.json file will be eventually overwritten by the data acquired using the configuration from the data-config.json file.

4. In the same directory there could be a configuration file containing JSON schema for validating the policy evaluation output. The file must be named output-schema.json. Example: /xfsc/example/1.0/output-schema.json
5. The policy package name inside the policy source code file must exactly match the group and policy (name) of the policy.

What does it mean?

• Let's see an example for the 1st convention.

package xfsc.example

allow {
    input.message == "hello world"
}

Next, the filename must be /xfsc/example/1.0/policy.rego. When such file is synchronized with the policy service (storage), the naming convention allows the service to understand which part is the policy group, which part is policy name and which part is version.

If we create the above policy and store it in the Git repo as /xfsc/example/1.0/policy.rego, after the Git server is synchronized with the Policy Storage, the policy service will automatically expose URLs for working with the policy at:

http://localhost:8081/policy/xfsc/example/1.0/evaluation
http://localhost:8081/policy/xfsc/example/1.0/lock

• The 2nd rule for static data file naming is to make sure that file /xfsc/example/1.0/data.json is passed and is available to the evaluation runtime when a policy is evaluated at URL:

http://localhost:8081/policy/xfsc/example/1.0/evaluation

Static data is accessed within the Rego policy with data.someKey. Example: If the /xfsc/example/1.0/data.json file is:

{
  "name": "some name"
}

one could access the data using data.name within the Rego source code.

• The 3rd rule for configuration file is to provide configurations for getting static JSON data from external URL. The file must contain a URL, an HTTP method and a period, after which an HTTP request is made to get the latest data.

The period must be added as duration e.g. 10h, 1h30m etc.

The file MAY contain body for the request. Example file contents:

{
  "url": "http://example.com/data.json?page=3",
  "method": "GET",
  "period": "10h",
  "body": {
    "key": "value"
  }
}

This means that every 10 hours an HTTP request is going to be made on the given URL, with GET method and the result is going to be stored as static data for this policy and passed during evaluation.

• The 4th rule for policy output schema validation is to provide a JSON schema which will be used to validate the output of the policy.

Example file contents:

{
  "type": "object",
  "properties": {
    "foo": {
      "type": "string",
      "minLength": 5
    }
  }
}

This policy output would be valid: {"foo":"barbaz"}. This policy output would be invalid: {"foo":"bar"}.

• The 5th rule for package naming is needed so that a generic evaluation function can be mapped and used for evaluating all kinds of different policies. Without a package naming rule, there's no way the service can automatically generate HTTP endpoints for working with arbitrary dynamically uploaded policies.

Access HTTP Headers inside a policy

HTTP request headers are passed to the evaluation runtime on each request. They can be accessed through a built-in extension function named external.http.header(). It accepts as argument the name of the header in Canonical format. For example, inside Rego the value of a header named Authorization can be retrieved as follows:

package example.example

auth := external.http.header("Authorization")

Header names are passed to the Rego runtime in Canonical format. This means that the first character and any characters following a hyphen are uppercase and the rest are lowercase.

More examples, if the policy service receive a request with the following headers:

accept-encoding: gzip, deflate
Accept-Language: en-us
fOO: Bar
x-loCATion: Baz

Inside a policy these headers could be accessed as follows:

accept_encoding := external.http.header("Accept-Encoding")
accept_language := external.http.header("Accept-Language")
foo := external.http.header("Foo")
location := external.http.header("X-Location")

Policy Extensions Functions

A brief documentation for the available Rego extensions functions which can be used during policy development.

Policy Extensions Functions

You can also look at the source code in package regofunc to understand the inner-working and capabilities of the extension functions.

Build

Local binary

To make the service binary locally, you can run the following command from the root directory (you must have Go installed):

go build -o policy ./cmd/policy/...

Docker image

You can see the Dockerfile of the service under the deployment directory. There is one Dockerfile for use during local development with docker-compose and one for building an optimized production image: deployment/ci/Dockerfile.

Versioning

There is one global exported variable named Version in main.go. The variable is set to the latest tag or commit hash during the build process. You can look in the production Dockerfile to see how the Version is set during build. The version is printed in the service log on startup and can be used to verify which specific commit of the code is deployed.

Version should not be set or modified manually in the source code.

Logging

The service outputs all logs to stdout as defined by the best practices in the Cloud Native community. See here for more details 12 Factor App. From there logs could be processed as needed in the specific running environment. The standard log levels are [debug,info,warn,error,fatal] and info is the default level. If you want to set another log level, use the ENV configuration variable LOG_LEVEL to set it.

GDPR

GDPR

Dependencies

Dependencies

Deployment

Helm deployment documentation

License

Apache 2.0 license