patrole/HACKING.rst
Andreas Jaeger 2aad808184 Update hacking for Python3
The repo is Python 3 now, so update hacking to version 3.0 which
supports Python 3.

Update hacking extension for newer flake8. Remove vi header check, this
is enabled as H106 already.

Fix problems found.

Change-Id: Ie4ccf0a1075995f5624a838388b6b0b46343129a
2020-04-02 15:56:12 +00:00

7.4 KiB

Patrole Coding Guide

Patrole Specific Commandments

Patrole borrows the following commandments from Tempest; refer to Tempest's Commandments for more information:

Note

The original Tempest Commandments do not include Patrole-specific paths. Patrole-specific paths replace the Tempest-specific paths within Patrole's hacking checks.

  • [T102] Cannot import OpenStack python clients in patrole_tempest_plugin/tests/api
  • [T105] Tests cannot use setUpClass/tearDownClass
  • [T107] Check that a service tag isn't in the module path
  • [T108] Check no hyphen at the end of rand_name() argument
  • [T109] Cannot use testtools.skip decorator; instead use decorators.skip_because from tempest.lib
  • [T113] Check that tests use data_utils.rand_uuid() instead of uuid.uuid4()
  • [N322] Method's default argument shouldn't be mutable

The following are Patrole's specific Commandments:

  • [P100] The rbac_rule_validation.action decorator must be applied to all RBAC tests
  • [P101] RBAC test filenames must end with "_rbac.py"; for example, test_servers_rbac.py, not test_servers.py
  • [P102] RBAC test class names must end in 'RbacTest'
  • [P103] self.client must not be used as a client alias; this allows for code that is more maintainable and easier to read
  • [P104] RBAC extension test class names must end in 'ExtRbacTest'

Supported OpenStack Components

Patrole only offers in-tree integration testing coverage for the following components:

  • Cinder
  • Glance
  • Keystone
  • Neutron
  • Nova

Patrole currently has no stable library, so reliance upon Patrole's framework for external RBAC testing should be done with caution. Nonetheless, even when Patrole has a stable library, it will only offer in-tree RBAC testing for the components listed above.

Role Overriding

Correct role overriding is vital to correct RBAC testing within Patrole. If a test does not call self.override_role() within the RBAC test, followed by the API endpoint that enforces the expected policy action, then the test is not a valid Patrole test: The API endpoint under test will be performed with admin role, which is always wrong unless CONF.patrole.rbac_test_role is also admin.

Patrole does not have a hacking check for role overriding, but one may be added in the future.

Branchless Patrole Considerations

Like Tempest, Patrole is branchless. This is to better ensure API and RBAC consistency between releases because API and RBAC behavior should not change between releases. This means that the stable branches are also gated by the Patrole master branch, which also means that proposed commits to Patrole must work against both the master and all the currently supported stable branches of the projects. As such there are a few special considerations that have to be accounted for when pushing new changes to Patrole.

1. New Tests for new features

Patrole, like Tempest, implicitly tests new features because new policies oftentimes accompany new features. The same Tempest philosophy regarding feature flags and new features also applies to Patrole.

2. New Tests for new policies

When adding tests for new policies that were not in previous releases of the projects, the new test must be properly skipped with a feature flag. This involves using the testtools.skip(Unless|If) decorator above the test to check if the required policy is enabled. Similarly, a feature flag must be used whenever an OpenStack service covered by Patrole changes one of its policies in a backwards-incompatible way. If there isn't a method of selecting the new policy from the config file then there won't be a mechanism to disable the test with older stable releases and the new test won't be able to merge.

Introduction of a new feature flag requires specifying a default value for the corresponding config option that is appropriate in the latest OpenStack release. Because Patrole is branchless, the feature flag's default value will need to be overridden to a value that is appropriate in earlier releases in which the feature isn't available. In DevStack, this can be accomplished by modifying Patrole's lib installation script for previous branches (because DevStack is branched).

3. Bug fix on core project needing Patrole changes

When trying to land a bug fix which changes a tested API you'll have to use the following procedure:

  1. Propose change to the project, get a +2 on the change even with the test failing Patrole side.
  2. Propose skip to the relevant Patrole test which will only be approved after the corresponding change in the project has a +2.
  3. Land project change in master and all open stable branches (if required).
  4. Land changed test in Patrole.

Otherwise the bug fix won't be able to land in the project.

4. New Tests for existing features or policies

The same Tempest logic regarding new tests for existing features or policies also applies to Patrole.

Black Box vs. White Box Testing

Tempest is a black box testing framework, meaning that it is concerned with testing public API endpoints and doesn't concern itself with testing internal implementation details. Patrole, as a Tempest plugin, also falls underneath the category of black box testing. However, even with policy in code documentation, some degree of white box testing is required in order to correctly write RBAC tests.

This is because policy-in-code documentation, while useful in many respects, is usually quite brief and its main purpose is to help operators understand how to customize policy configuration rather than to help developers understand complex policy authorization work flows. For example, policy in code documentation doesn't make deriving multiple policies <multiple-policies> easy. Such documentation also doesn't usually mention that a specific parameter needs to be set, or that a particular microversion must be enabled, or that a particular set of prerequisite API or policy actions must be executed, in order for the policy under test to be enforced by the server. This means that test writers must account for the internal RBAC implementation in API code in order to correctly understand the complete RBAC work flow within an API.

Besides, as mentioned elsewhere <design-principles> in this documentation, not all services currently implement policy in code, making some degree of white box testing a "necessary evil" for writing robust RBAC tests.