Hacking on cloud-init

This document describes how to contribute changes to cloud-init. It assumes you have a GitHub account, and refers to your GitHub user as GH_USER throughout.

Submitting your first pull request

Follow these steps to submit your first pull request to cloud-init:

  • To contribute to cloud-init, you must sign the Canonical contributor license agreement

    • If you have already signed it as an individual, your Launchpad user will be listed in the contributor-agreement-canonical group. (Unfortunately there is no easy way to check if an organization or company you are doing work for has signed.)
    • When signing it:
      • ensure that you fill in the GitHub username field.
      • when prompted for ‘Project contact’ or ‘Canonical Project Manager’, enter ‘Rick Harding’.
    • If your company has signed the CLA for you, please contact us to help in verifying which Launchpad/GitHub accounts are associated with the company.
    • For any questions or help with the process, please email Rick Harding with the subject, “Cloud-Init CLA”
    • You also may contact user rick_h in the #cloud-init channel on the Freenode IRC network.
  • Configure git with your email and name for commit messages.

    Your name will appear in commit messages and will also be used in changelogs or release notes. Give yourself credit!:

    git config user.name "Your Name"
    git config user.email "Your Email"
    
  • Sign into your GitHub account

  • Fork the upstream repository on Github and clicking on the Fork button

  • Create a new remote pointing to your personal GitHub repository.

    git clone git://github.com/canonical/cloud-init
    cd cloud-init
    git remote add GH_USER git@github.com:GH_USER/cloud-init.git
    git push GH_USER master
    
  • Read through the cloud-init Code Review Process, so you understand how your changes will end up in cloud-init’s codebase.

  • Submit your first cloud-init pull request, adding yourself to the in-repository list that we use to track CLA signatures: tools/.github-cla-signers

    • See PR #344 and PR #345 for examples of what this pull request should look like.
    • Note that .github-cla-signers is sorted alphabetically.
    • (If you already have a change that you want to submit, you can also include the change to tools/.github-cla-signers in that pull request, there is no need for two separate PRs.)

Transferring CLA Signatures from Launchpad to Github

For existing contributors who have signed the agreement in Launchpad before the Github username field was included, we need to verify the link between your Launchpad account and your GitHub account. To enable us to do this, we ask that you create a branch with both your Launchpad and GitHub usernames against both the Launchpad and GitHub cloud-init repositories. We’ve added a tool (tools/migrate-lp-user-to-github) to the cloud-init repository to handle this migration as automatically as possible.

The cloud-init team will review the two merge proposals and verify that the CLA has been signed for the Launchpad user and record the associated GitHub account.

Do these things for each feature or bug

  • Create a new topic branch for your work:

    git checkout -b my-topic-branch
    
  • Make and commit your changes (note, you can make multiple commits, fixes, more commits.):

    git commit
    
  • Run unit tests and lint/formatting checks with tox:

    tox
    
  • Push your changes to your personal GitHub repository:

    git push -u GH_USER my-topic-branch
    
  • Use your browser to create a merge request:

    • Open the branch on GitHub

      • You can see a web view of your repository and navigate to the branch at:

        https://github.com/GH_USER/cloud-init/tree/my-topic-branch

    • Click ‘Pull Request`

    • Fill out the pull request title, summarizing the change and a longer message indicating important details about the changes included, like

      Activate the frobnicator.
      
      The frobnicator was previously inactive and now runs by default.
      This may save the world some day.  Then, list the bugs you fixed
      as footers with syntax as shown here.
      
      The commit message should be one summary line of less than
      74 characters followed by a blank line, and then one or more
      paragraphs describing the change and why it was needed.
      
      This is the message that will be used on the commit when it
      is sqaushed and merged into trunk.
      
      LP: #1
      

      Note that the project continues to use LP: #NNNNN format for closing launchpad bugs rather than GitHub Issues.

    • Click ‘Create Pull Request`

Then, someone in the Ubuntu Server team will review your changes and follow up in the pull request. Look at the Code Review Process doc to understand the following steps.

Feel free to ping and/or join #cloud-init on freenode irc if you have any questions.

Design

This section captures design decisions that are helpful to know when hacking on cloud-init.

Cloud Config Modules

  • Any new modules should use underscores in any new config options and not hyphens (e.g. new_option and not new-option).

Unit Testing

cloud-init uses pytest to run its tests, and has tests written both as unittest.TestCase sub-classes and as un-subclassed pytest tests. The following guidelines should be followed:

  • For ease of organisation and greater accessibility for developers not familiar with pytest, all cloud-init unit tests must be contained within test classes

    • Put another way, module-level test functions should not be used
  • pytest test classes should use pytest fixtures to share functionality instead of inheritance

  • As all tests are contained within classes, it is acceptable to mix TestCase test classes and pytest test classes within the same test file

  • pytest tests should use bare assert statements, to take advantage of pytest’s assertion introspection

    • For == and other commutative assertions, the expected value should be placed before the value under test: assert expected_value == function_under_test()
  • As we still support Ubuntu 16.04 (Xenial Xerus), we can only use pytest features that are available in v2.8.7. This is an inexhaustive list of ways in which this may catch you out:

    • Support for using yield in pytest.fixture functions was only introduced in pytest 3.0. Such functions must instead use the pytest.yield_fixture decorator.
    • Only the following built-in fixtures are available [1]:
      • cache
      • capfd
      • caplog (provided by python3-pytest-catchlog on xenial)
      • capsys
      • monkeypatch
      • pytestconfig
      • record_xml_property
      • recwarn
      • tmpdir_factory
      • tmpdir
    • On xenial, the objects returned by the tmpdir fixture cannot be used where paths are required; they are rejected as invalid paths. You must instead use their .strpath attribute.
      • For example, instead of util.write_file(tmpdir.join("some_file"), ...), you should write util.write_file(tmpdir.join("some_file").strpath, ...).
    • The pytest.param function cannot be used. It was introduced in pytest 3.1, which means it is not available on xenial. The more limited mechanism it replaced was removed in pytest 4.0, so is not available in focal or later. The only available alternatives are to write mark-requiring test instances as completely separate tests, without utilising parameterisation, or to apply the mark to the entire parameterized test (and therefore every test instance).
  • Variables/parameter names for Mock or MagicMock instances should start with m_ to clearly distinguish them from non-mock variables

    • For example, m_readurl (which would be a mock for readurl)
  • The assert_* methods that are available on Mock and MagicMock objects should be avoided, as typos in these method names may not raise AttributeError (and so can cause tests to silently pass). An important exception: if a Mock is autospecced then misspelled assertion methods will raise an AttributeError, so these assertion methods may be used on autospecced Mock objects.

    For non-autospecced Mock s, these substitutions can be used (m is assumed to be a Mock):

    • m.assert_any_call(*args, **kwargs) => assert mock.call(*args, **kwargs) in m.call_args_list
    • m.assert_called() => assert 0 != m.call_count
    • m.assert_called_once() => assert 1 == m.call_count
    • m.assert_called_once_with(*args, **kwargs) => assert [mock.call(*args, **kwargs)] == m.call_args_list
    • m.assert_called_with(*args, **kwargs) => assert mock.call(*args, **kwargs) == m.call_args_list[-1]
    • m.assert_has_calls(call_list, any_order=True) => for call in call_list: assert call in m.call_args_list
      • m.assert_has_calls(...) and m.assert_has_calls(..., any_order=False) are not easily replicated in a single statement, so their use when appropriate is acceptable.
    • m.assert_not_called() => assert 0 == m.call_count
  • Test arguments should be ordered as follows:

    • mock.patch arguments. When used as a decorator, mock.patch partially applies its generated Mock object as the first argument, so these arguments must go first.
    • pytest.mark.parametrize arguments, in the order specified to the parametrize decorator. These arguments are also provided by a decorator, so it’s natural that they sit next to the mock.patch arguments.
    • Fixture arguments, alphabetically. These are not provided by a decorator, so they are last, and their order has no defined meaning, so we default to alphabetical.
  • It follows from this ordering of test arguments (so that we retain the property that arguments left-to-right correspond to decorators bottom-to-top) that test decorators should be ordered as follows:

    • pytest.mark.parametrize
    • mock.patch
  • When there are multiple patch calls in a test file for the module it is testing, it may be desirable to capture the shared string prefix for these patch calls in a module-level variable. If used, such variables should be named M_PATH or, for datasource tests, DS_PATH.

Type Annotations

The cloud-init codebase uses Python’s annotation support for storing type annotations in the style specified by PEP-484. Their use in the codebase is encouraged but with one important caveat: types from the typing module cannot be used.

cloud-init still supports Python 3.4, which doesn’t have the typing module in the stdlib. This means that the use of any types from the typing module in the codebase would require installation of an additional Python module on platforms using Python 3.4. As such platforms are generally in maintenance mode, the introduction of a new dependency may act as a break in compatibility in practical terms.

Similarly, only function annotations are appropriate for use, as the variable annotations specified in PEP-526 were introduced in Python 3.6.

[1]

This list of fixtures (with markup) can be reproduced by running:

py.test-3 --fixtures -q | grep "^[^ -]" | grep -v '\(no\|capturelog\)' | sort | sed 's/.*/* ``\0``/'

in a xenial lxd container with python3-pytest-catchlog installed.

Feature Flags

Feature flags are used as a way to easily toggle configuration at build time. They are provided to accommodate feature deprecation and downstream configuration changes.

Currently used upstream values for feature flags are set in cloudinit/features.py. Overrides to these values (typically via quilt patch) can be placed in a file called feature_overrides.py in the same directory. Any value set in feature_overrides.py will override the original value set in features.py.

Each flag should include a short comment regarding the reason for the flag and intended lifetime.

Tests are required for new feature flags, and tests must verify all valid states of a flag, not just the default state.

cloudinit.features.ALLOW_EC2_MIRRORS_ON_NON_AWS_INSTANCE_TYPES = False

When configuring apt mirrors, if ALLOW_EC2_MIRRORS_ON_NON_AWS_INSTANCE_TYPES is True cloud-init will detect that a datasource’s availability_zone property looks like an EC2 availability zone and set the ec2_region variable when generating mirror URLs; this can lead to incorrect mirrors being configured in clouds whose AZs follow EC2’s naming pattern.

As of 20.3, ALLOW_EC2_MIRRORS_ON_NON_AWS_INSTANCE_TYPES is False so we no longer include ec2_region in mirror determination on non-AWS cloud platforms.

If the old behavior is desired, users can provide the appropriate mirrors via apt: directives in cloud-config.

cloudinit.features.ERROR_ON_USER_DATA_FAILURE = True

If there is a failure in obtaining user data (i.e., #include or decompress fails) and ERROR_ON_USER_DATA_FAILURE is False, cloud-init will log a warning and proceed. If it is True, cloud-init will instead raise an exception.

As of 20.3, ERROR_ON_USER_DATA_FAILURE is True.

(This flag can be removed after Focal is no longer supported.)

Ongoing Refactors

This captures ongoing refactoring projects in the codebase. This is intended as documentation for developers involved in the refactoring, but also for other developers who may interact with the code being refactored in the meantime.

cloudinit.net -> cloudinit.distros.networking Hierarchy

cloudinit.net was imported from the curtin codebase as a chunk, and then modified enough that it integrated with the rest of the cloud-init codebase. Over the ~4 years since, the fact that it is not fully integrated into the Distro hierarchy has caused several issues.

The common pattern of these problems is that the commands used for networking are different across distributions and operating systems. This has lead to cloudinit.net developing its own “distro determination” logic: get_interfaces_by_mac is probably the clearest example of this. Currently, these differences are primarily split along Linux/BSD lines. However, it would be short-sighted to only refactor in a way that captures this difference: we can anticipate that differences will develop between Linux-based distros in future, or there may already be differences in tooling that we currently work around in less obvious ways.

The high-level plan is to introduce a hierarchy of networking classes in cloudinit.distros.networking, which each Distro subclass will reference. These will capture the differences between networking on our various distros, while still allowing easy reuse of code between distros that share functionality (e.g. most of the Linux networking behaviour). Distro objects will instantiate the networking classes at self.net, so callers will call distro.net.<func> instead of cloudinit.net.<func>; this will necessitate access to an instantiated Distro object.

An implementation note: there may be external consumers of the cloudinit.net module. We don’t consider this a public API, so we will be removing it as part of this refactor. However, we will ensure that the new API is complete from its introduction, so that any such consumers can move over to it wholesale. (Note, however, that this new API is still not considered public or stable, and may not replicate the existing API exactly.)

In more detail:

  • The root of this hierarchy will be the cloudinit.distros.networking.Networking class. This class will have a corresponding method for every cloudinit.net function that we identify to be involved in refactoring. Initially, these methods’ implementations will simply call the corresponding cloudinit.net function. (This gives us the complete API from day one, for existing consumers.)
  • As the biggest differentiator in behaviour, the next layer of the hierarchy will be two subclasses: LinuxNetworking and BSDNetworking. These will be introduced in the initial PR.
  • When a difference in behaviour for a particular distro is identified, a new Networking subclass will be created. This new class should generally subclass either LinuxNetworking or BSDNetworking.
  • To be clear: Networking subclasses will only be created when needed, we will not create a full hierarchy of per-Distro subclasses up-front.
  • Each Distro class will have a class variable (cls.networking_cls) which points at the appropriate networking class (initially this will be either LinuxNetworking or BSDNetworking).
  • When Distro classes are instantiated, they will instantiate cls.networking_cls and store the instance at self.net. (This will be implemented in cloudinit.distros.Distro.__init__.)
  • A helper function will be added which will determine the appropriate Distro subclass for the current system, instantiate it and return its net attribute. (This is the entry point for existing consumers to migrate to.)
  • Callers of refactored functions will change from calling cloudinit.net.<func> to distro.net.<func>, where distro is an instance of the appropriate Distro class for this system. (This will require making such an instance available to callers, which will constitute a large part of the work in this project.)

After the initial structure is in place, the work in this refactor will consist of replacing the cloudinit.net.some_func call in each cloudinit.distros.networking.Networking method with the actual implementation. This can be done incrementally, one function at a time:

  • pick an unmigrated cloudinit.distros.networking.Networking method
  • find it in the the list of bugs tagged net-refactor and assign yourself to it (see Managing Work/Tracking Progress below for more details)
  • refactor all of its callers to call the distro.net.<func> method on Distro instead of the cloudinit.net.<func> function. (This is likely to be the most time-consuming step, as it may require plumbing Distro objects through to places that previously have not consumed them.)
  • refactor its implementation from cloudinit.net into the Networking hierarchy (e.g. if it has an if/else on BSD, this is the time to put the implementations in their respective subclasses)
    • if part of the method contains distro-independent logic, then you may need to create new methods to capture this distro-specific logic; we don’t want to replicate common logic in different Networking subclasses
    • if after the refactor, the method on the root Networking class no longer has any implementation, it should be converted to an abstractmethod
  • ensure that the new implementation has unit tests (either by moving existing tests, or by writing new ones)
  • ensure that the new implementation has a docstring
  • add any appropriate type annotations
    • note that we must follow the constraints described in the “Type Annotations” section above, so you may not be able to write complete annotations
    • we have type aliases defined in cloudinit.distros.networking which should be used when applicable
  • finally, remove it (and any other now-unused functions) from cloudinit.net (to avoid having two parallel implementations)

cloudinit.net Functions/Classes

The functions/classes that need refactoring break down into some broad categories:

  • helpers for accessing /sys (that should not be on the top-level Networking class as they are Linux-specific):
    • get_sys_class_path
    • sys_dev_path
    • read_sys_net
    • read_sys_net_safe
    • read_sys_net_int
  • those that directly access /sys (via helpers) and should (IMO) be included in the API of the Networking class:
    • generate_fallback_config
      • the config_driver parameter is used and passed as a boolean, so we can change the default value to False (instead of None)
    • get_ib_interface_hwaddr
    • get_interface_mac
    • interface_has_own_mac
    • is_bond
    • is_bridge
    • is_physical
    • is_renamed
    • is_up
    • is_vlan
    • wait_for_physdevs
  • those that directly access /sys (via helpers) but may be Linux-specific concepts or names:
    • get_master
    • device_devid
    • device_driver
  • those that directly use ip:
    • _get_current_rename_info
      • this has non-distro-specific logic so should potentially be refactored to use helpers on self instead of ip directly (rather than being wholesale reimplemented in each of BSDNetworking or LinuxNetworking)
      • we can also remove the check_downable argument, it’s never specified so is always True
    • _rename_interfaces
      • this has several internal helper functions which use ip directly, and it calls _get_current_rename_info. That said, there appears to be a lot of non-distro-specific logic that could live in a function on Networking, so this will require some careful refactoring to avoid duplicating that logic in each of BSDNetworking and LinuxNetworking.
      • only the renames and current_info parameters are ever passed in (and current_info only by tests), so we can remove the others from the definition
    • EphemeralIPv4Network
      • this is another case where it mixes distro-specific and non-specific functionality. Specifically, __init__, __enter__ and __exit__ are non-specific, and the remaining methods are distro-specific.
      • when refactoring this, the need to track cleanup_cmds likely means that the distro-specific behaviour cannot be captured only in the Networking class. See this comment in PR #363 for more thoughts.
  • those that implicitly use /sys via their call dependencies:
    • master_is_bridge_or_bond
      • appends to get_master return value, which is a /sys path
    • extract_physdevs
      • calls device_driver and device_devid in both _version_* impls
    • apply_network_config_names
      • calls extract_physdevs
      • there is already a Distro.apply_network_config_names which in the default implementation calls this function; this and its BSD subclass implementations should be refactored at the same time
      • the strict_present and strict_busy parameters are never passed, nor are they used in the function definition, so they can be removed
    • get_interfaces
      • calls device_driver, device_devid amongst others
    • get_ib_hwaddrs_by_interface
      • calls get_interfaces
  • those that may fall into the above categories, but whose use is only related to netfailover (which relies on a Linux-specific network driver, so is unlikely to be relevant elsewhere without a substantial refactor; these probably only need implementing in LinuxNetworking):
    • get_dev_features
    • has_netfail_standby_feature
      • calls get_dev_features
    • is_netfailover
    • is_netfail_master
      • this is called from generate_fallback_config
    • is_netfail_primary
    • is_netfail_standby
    • N.B. all of these take an optional driver argument which is used to pass around a value to avoid having to look it up by calling device_driver every time. This is something of a leaky abstraction, and is better served by caching on device_driver or storing the cached value on self, so we can drop the parameter from the new API.
  • those that use /sys (via helpers) and have non-exhaustive BSD logic:
    • get_devicelist
  • those that already have separate Linux/BSD implementations:
    • find_fallback_nic
    • get_interfaces_by_mac
  • those that have no OS-specific functionality (so do not need to be refactored):
    • ParserError
    • RendererNotFoundError
    • has_url_connectivity
    • is_ip_address
    • is_ipv4_address
    • natural_sort_key

Note that the functions in cloudinit.net use inconsistent parameter names for “string that contains a device name”; we can standardise on devname (the most common one) in the refactor.

Managing Work/Tracking Progress

To ensure that we won’t have multiple people working on the same part of the refactor at the same time, there is a bug for each function. You can see the current status by looking at the list of bugs tagged net-refactor.

When you’re working on refactoring a particular method, ensure that you have assigned yourself to the corresponding bug, to avoid duplicate work.

Generally, when considering what to pick up to refactor, it is best to start with functions in cloudinit.net which are not called by anything else in cloudinit.net. This allows you to focus only on refactoring that function and its callsites, rather than having to update the other cloudinit.net function also.

References