JSON is great for a wire format, but awful for a configuration file. oslo.policy can easily also support a safe subset of YAML.
The policy files for the services can be complicated and hard to use. Even knowing what operations are protected by what rules is difficult since the operation name (e.g., identity:create_user) isn’t the same as the operation the user performs (e.g., PUT /v3/users). The sample policy files would be a lot more usable if we could simply put a comment in the file saying what the user operation is for the rule, but the only format allowed (JSON) doesn’t support comments.
As a deployer, I should be able to comment the policy file with descriptions for my custom roles and rules.
As a developer, I should be able to provide a sample policy file that’s self-describing. As in, add comments to the sample policy file.
Rather than parse the policy file using the JSON parser, oslo.policy will parse it using the YAML parser (provided by PyYAML). Since JSON is a subset of YAML existing JSON files will continue to work. Projects and deployers can switch to simplified YAML and document using comments if they want.
oslo.policy will be changed to use PyYAML’s safe_load() to parse the policy file rather than the JSON parser ( jsonutils.loads() ).
For some reason the name of the method to read the policy file includes the format. This was incorrect to begin with and is made even less accurate since the format is YAML, so rename load_json() to load() (keeping the old method name around but deprecated, using debtcollector).
oslo.policy defines a policy_file config option which defaults to policy.json. So the default behavior is to look for a policy.json file. The new default behavior will look first for policy.yaml and if that doesn’t exist it will look for policy.json. As such, the default will be removed and the new default behavior will be described in the help text.
Files to change:
We could preprocess the JSON files to remove lines that look like comments. But then the format isn’t really JSON and is not a standard format.
We could have separate loaders for JSON vs YAML and use the parser based on the file extension. This is possible but is unnecessary since JSON is a subset of YAML.
The oslo_policy.policy.Rules class’s load_json method will be renamed to load. load_json will be deprecated.
In its most general form, YAML allows the document to contain executable code that’s then run. We don’t want our policy files to be allowed to contain executable code. As such, we’ll use safe_load() instead of load().
I don’t know if the YAML parser is a lot slower, but since it supports several representations for the same result I assume it takes more work to parse it. The policy file is read when the server starts and also whenever the file changes (it used to be read on every request, but that’s been changed to check the modification time), so I don’t think this is going to be noticeable.
The different projects all have sample policy.json files. These files should be renamed to policy.yaml, the format changed to simplified YAML, and comments added. This doesn’t have to happen immediately since policy.json will continue to work.
The projects are already loading their policy files during gate testing.
If the documentation mentions that the policy file is in JSON format then that can be changed to say YAML format. Any policy sample files should be changed to the simpler YAML format rather than JSON.
oslo.policy will depend on PyYAML. This is already in global-requirements.
This work is licensed under a Creative Commons Attribution 3.0 Unported License. http://creativecommons.org/licenses/by/3.0/legalcode