Integration Testing

Overview

This page describes the execution, development, and architecture of the cloud-init integration tests:

  • Execution explains the options available and running of tests
  • Development shows how to write test cases
  • Architecture explains the internal processes

Execution

Overview

In order to avoid the need for dependencies and ease the setup and configuration users can run the integration tests via tox:

$ git clone https://github.com/canonical/cloud-init
$ cd cloud-init
$ tox -e citest -- -h

Everything after the double dash will be passed to the integration tests. Executing tests has several options:

  • run an alias to run both collect and verify. The tree_run command does the same thing, except uses a deb built from the current working tree.
  • collect deploys on the specified platform and distro, patches with the requested deb or rpm, and finally collects output of the arbitrary commands. Similarly, `tree_collect will collect output using a deb built from the current working tree.
  • verify given a directory of test data, run the Python unit tests on it to generate results.
  • bddeb will build a deb of the current working tree.

Run

The first example will provide a complete end-to-end run of data collection and verification. There are additional examples below explaining how to run one or the other independently.

$ git clone https://github.com/canonical/cloud-init
$ cd cloud-init
$ tox -e citest -- run --verbose \
    --os-name stretch --os-name xenial \
    --deb cloud-init_0.7.8~my_patch_all.deb \
    --preserve-data --data-dir ~/collection \
    --preserve-instance

The above command will do the following:

  • run both collect output and run tests the output
  • --verbose verbose output
  • --os-name stretch on the Debian Stretch release
  • --os-name xenial on the Ubuntu Xenial release
  • --deb cloud-init_0.7.8~patch_all.deb use this deb as the version of cloud-init to run with
  • --preserve-data always preserve collected data, do not remove data after successful test run
  • --preserve-instance do not destroy the instance after test to allow for debugging the stopped instance during integration test development. By default, test instances are destroyed after the test completes.
  • --data-dir ~/collection write collected data into ~/collection, rather than using a temporary directory

For a more detailed explanation of each option see below.

Note

By default, data collected by the run command will be written into a temporary directory and deleted after a successful. If you would like to preserve this data, please use the option --preserve-data.

Collect

If developing tests it may be necessary to see if cloud-config works as expected and the correct files are pulled down. In this case only a collect can be ran by running:

$ tox -e citest -- collect -n xenial --data-dir /tmp/collection

The above command will run the collection tests on xenial and place all results into /tmp/collection.

Verify

When developing tests it is much easier to simply rerun the verify scripts without the more lengthy collect process. This can be done by running:

$ tox -e citest -- verify --data-dir /tmp/collection

The above command will run the verify scripts on the data discovered in /tmp/collection.

TreeRun and TreeCollect

If working on a cloud-init feature or resolving a bug, it may be useful to run the current copy of cloud-init in the integration testing environment. The integration testing suite can automatically build a deb based on the current working tree of cloud-init and run the test suite using this deb.

The tree_run and tree_collect commands take the same arguments as the run and collect commands. These commands will build a deb and write it into a temporary file, then start the test suite and pass that deb in. To build a deb only, and not run the test suite, the bddeb command can be used.

Note that code in the cloud-init working tree that has not been committed when the cloud-init deb is built will still be included. To build a cloud-init deb from or use the tree_run command using a copy of cloud-init located in a different directory, use the option --cloud-init /path/to/cloud-init.

$ tox -e citest -- tree_run --verbose \
    --os-name xenial --os-name stretch \
    --test modules/final_message --test modules/write_files \
    --result /tmp/result.yaml

Bddeb

The bddeb command can be used to generate a deb file. This is used by the tree_run and tree_collect commands to build a deb of the current working tree. It can also be used a user to generate a deb for use in other situations and avoid needing to have all the build and test dependencies installed locally.

  • --bddeb-args: arguments to pass through to bddeb
  • --build-os: distribution to use as build system (default is xenial)
  • --build-platform: platform to use for build system (default is lxd)
  • --cloud-init: path to base of cloud-init tree (default is ‘.’)
  • --deb: path to write output deb to (default is ‘.’)

Setup Image

By default an image that is used will remain unmodified, but certain scenarios may require image modification. For example, many images may use a much older cloud-init. As a result tests looking at newer functionality will fail because a newer version of cloud-init may be required. The following options can be used for further customization:

  • --deb: install the specified deb into the image
  • --rpm: install the specified rpm into the image
  • --repo: enable a repository and upgrade cloud-init afterwards
  • --ppa: enable a ppa and upgrade cloud-init afterwards
  • --upgrade: upgrade cloud-init from repos
  • --upgrade-full: run a full system upgrade
  • --script: execute a script in the image. This can perform any setup required that is not covered by the other options

Test Case Development

Overview

As a test writer you need to develop a test configuration and a verification file:

  • The test configuration specifies a specific cloud-config to be used by cloud-init and a list of arbitrary commands to capture the output of (e.g my_test.yaml)
  • The verification file runs tests on the collected output to determine the result of the test (e.g. my_test.py)

The names must match, however the extensions will of course be different, yaml vs py.

Configuration

The test configuration is a YAML file such as ntp_server.yaml below:

#
# Empty NTP config to setup using defaults
#
# NOTE: this should not require apt feature, use 'which' rather than 'dpkg -l'
# NOTE: this should not require no_ntpdate feature, use 'which' to check for
#       installation rather than 'dpkg -l', as 'grep ntp' matches 'ntpdate'
# NOTE: the verifier should check for any ntp server not 'ubuntu.pool.ntp.org'
cloud_config: |
  #cloud-config
  ntp:
    servers:
      - pool.ntp.org
required_features:
  - apt
  - no_ntpdate
  - ubuntu_ntp
collect_scripts:
  ntp_installed_servers: |
    #!/bin/bash
    dpkg -l | grep ntp | wc -l
  ntp_conf_dist_servers: |
    #!/bin/bash
    ls /etc/ntp.conf.dist | wc -l
  ntp_conf_servers: |
    #!/bin/bash
    cat /etc/ntp.conf | grep '^server'

There are several keys, 1 required and some optional, in the YAML file:

  1. The required key is cloud_config. This should be a string of valid YAML that is exactly what would normally be placed in a cloud-config file, including the cloud-config header. This essentially sets up the scenario under test.
  2. One optional key is collect_scripts. This key has one or more sub-keys containing strings of arbitrary commands to execute (e.g. `cat /var/log/cloud-config-output.log`). In the example above the output of dpkg is captured, grep for ntp, and the number of lines reported. The name of the sub-key is important. The sub-key is used by the verification script to recall the output of the commands ran.
  3. The optional enabled key enables or disables the test case. By default the test case will be enabled.
  4. The optional required_features key may be used to specify a list of features flags that an image must have to be able to run the test case. For example, if a test case relies on an image supporting apt, then the config for the test case should include required_features: [ apt ].

Default Collect Scripts

By default the following files will be collected for every test. There is no need to specify these items:

  • /var/log/cloud-init.log
  • /var/log/cloud-init-output.log
  • /run/cloud-init/.instance-id
  • /run/cloud-init/result.json
  • /run/cloud-init/status.json
  • `dpkg-query -W -f='${Version}' cloud-init`

Verification

The verification script is a Python file with unit tests like the one, ntp_server.py, below:

# This file is part of cloud-init. See LICENSE file for license information.

"""cloud-init Integration Test Verify Script"""
from tests.cloud_tests.testcases import base


class TestNtp(base.CloudTestCase):
    """Test ntp module"""

    def test_ntp_installed(self):
        """Test ntp installed"""
        out = self.get_data_file('ntp_installed_empty')
        self.assertEqual(1, int(out))

    def test_ntp_dist_entries(self):
        """Test dist config file has one entry"""
        out = self.get_data_file('ntp_conf_dist_empty')
        self.assertEqual(1, int(out))

    def test_ntp_entires(self):
        """Test config entries"""
        out = self.get_data_file('ntp_conf_empty')
        self.assertIn('pool 0.ubuntu.pool.ntp.org iburst', out)
        self.assertIn('pool 1.ubuntu.pool.ntp.org iburst', out)
        self.assertIn('pool 2.ubuntu.pool.ntp.org iburst', out)
        self.assertIn('pool 3.ubuntu.pool.ntp.org iburst', out)

# vi: ts=4 expandtab

Here is a breakdown of the unit test file:

  • The import statement allows access to the output files.
  • The class can be named anything, but must import the base.CloudTestCase, either directly or via another test class.
  • There can be 1 to N number of functions with any name, however only functions starting with test_* will be executed.
  • There can be 1 to N number of classes in a test module, however only classes inheriting from base.CloudTestCase will be loaded.
  • Output from the commands can be accessed via self.get_data_file('key') where key is the sub-key of collect_scripts above.
  • The cloud config that the test ran with can be accessed via self.cloud_config, or any entry from the cloud config can be accessed via self.get_config_entry('key').
  • See the base CloudTestCase for additional helper functions.

Layout

Integration tests are located under the tests/cloud_tests directory. Test configurations are placed under configs and the test verification scripts under testcases:

cloud-init$ tree -d tests/cloud_tests/
tests/cloud_tests/
├── configs
│   ├── bugs
│   ├── examples
│   ├── main
│   └── modules
└── testcases
    ├── bugs
    ├── examples
    ├── main
    └── modules

The sub-folders of bugs, examples, main, and modules help organize the tests. View the README.md in each to understand in more detail each directory.

Test Creation Helper

The integration testing suite has a built in helper to aid in test development. Help can be invoked via tox -e citest -- create --help. It can create a template test case config file with user data passed in from the command line, as well as a template test case verifier module.

The following would create a test case named example under the modules category with the given description, and cloud config data read in from /tmp/user_data.

$ tox -e citest -- create modules/example \
    -d "a simple example test case" -c "$(< /tmp/user_data)"

Development Checklist

  • Configuration File
    • Named ‘your_test.yaml’
    • Contains at least a valid cloud-config
    • Optionally, commands to capture additional output
    • Valid YAML
    • Placed in the appropriate sub-folder in the configs directory
    • Any image features required for the test are specified
  • Verification File
    • Named ‘your_test.py’
    • Valid unit tests validating output collected
    • Passes pylint & pep8 checks
    • Placed in the appropriate sub-folder in the test cases directory
  • Tested by running the test:

    $ tox -e citest -- run -verbose \
        --os-name <release target> \
        --test modules/your_test.yaml \
        [--deb <build of cloud-init>]
    

Platforms

EC2

To run on the EC2 platform it is required that the user has an AWS credentials configuration file specifying his or her access keys and a default region. These configuration files are the standard that the AWS cli and other AWS tools utilize for interacting directly with AWS itself and are normally generated when running aws configure:

$ cat $HOME/.aws/credentials
[default]
aws_access_key_id = <KEY HERE>
aws_secret_access_key = <KEY HERE>
$ cat $HOME/.aws/config
[default]
region = us-west-2

Azure Cloud

To run on Azure Cloud platform users login with Service Principal and export credentials file. Region is defaulted and can be set in tests/cloud_tests/platforms.yaml. The Service Principal credentials are the standard authentication for Azure SDK to interact with Azure Services:

Create Service Principal account or login

$ az ad sp create-for-rbac --name "APP_ID" --password "STRONG-SECRET-PASSWORD"
$ az login --service-principal --username "APP_ID" --password "STRONG-SECRET-PASSWORD"

Export credentials

$ az ad sp create-for-rbac --sdk-auth > $HOME/.azure/credentials.json
{
    "clientId": "<Service principal ID>",
    "clientSecret": "<Service principal secret/password>",
    "subscriptionId": "<Subscription associated with the service principal>",
    "tenantId": "<The service principal's tenant>",
    "activeDirectoryEndpointUrl": "https://login.microsoftonline.com",
    "resourceManagerEndpointUrl": "https://management.azure.com/",
    "activeDirectoryGraphResourceId": "https://graph.windows.net/",
    "sqlManagementEndpointUrl": "https://management.core.windows.net:8443/",
    "galleryEndpointUrl": "https://gallery.azure.com/",
    "managementEndpointUrl": "https://management.core.windows.net/"
}

Set region in platforms.yaml

azurecloud:
    enabled: true
    region: West US 2
    vm_size: Standard_DS1_v2
    storage_sku: standard_lrs
    tag: ci

Architecture

The following section outlines the high-level architecture of the integration process.

Overview

The process flow during a complete end-to-end LXD-backed test.

  1. Configuration
    • The back end and specific distro releases are verified as supported
    • The test or tests that need to be run are determined either by directory or by individual yaml
  2. Image Creation
    • Acquire the request LXD image
    • Install the specified cloud-init package
    • Clean the image so that it does not appear to have been booted
    • A snapshot of the image is created and reused by all tests
  3. Configuration
    • For each test, the cloud-config is injected into a copy of the snapshot and booted
    • The framework waits for /var/lib/cloud/instance/boot-finished (up to 120 seconds)
    • All default commands are ran and output collected
    • Any commands the user specified are executed and output collected
  4. Verification
    • The default commands are checked for any failures, errors, and warnings to validate basic functionality of cloud-init completed successfully
    • The user generated unit tests are then ran validating against the collected output
  5. Results
    • If any failures were detected the test suite returns a failure
    • Results can be dumped in yaml format to a specified file using the -r <result_file_name>.yaml option

Configuring the Test Suite

Most of the behavior of the test suite is configurable through several yaml files. These control the behavior of the test suite’s platforms, images, and tests. The main config files for platforms, images and test cases are platforms.yaml, releases.yaml and testcases.yaml.

Config handling

All configurable parts of the test suite use a defaults + overrides system for managing config entries. All base config items are dictionaries.

Merging is done on a key-by-key basis, with all keys in the default and override represented in the final result. If a key exists both in the defaults and the overrides, then the behavior depends on the type of data the key refers to. If it is atomic data or a list, then the overrides will replace the default. If the data is a dictionary then the value will be the result of merging that dictionary from the default config and that dictionary from the overrides.

Merging is done using the function tests.cloud_tests.config.merge_config, which can be examined for more detail on config merging behavior.

The following demonstrates merge behavior:

defaults:
    list_item:
     - list_entry_1
     - list_entry_2
    int_item_1: 123
    int_item_2: 234
    dict_item:
        subkey_1: 1
        subkey_2: 2
        subkey_dict:
            subsubkey_1: a
            subsubkey_2: b

overrides:
    list_item:
     - overridden_list_entry
    int_item_1: 0
    dict_item:
        subkey_2: false
        subkey_dict:
            subsubkey_2: 'new value'

result:
    list_item:
     - overridden_list_entry
    int_item_1: 0
    int_item_2: 234
    dict_item:
        subkey_1: 1
        subkey_2: false
        subkey_dict:
            subsubkey_1: a
            subsubkey_2: 'new value'

Image Config

Image configuration is handled in releases.yaml. The image configuration controls how platforms locate and acquire images, how the platforms should interact with the images, how platforms should detect when an image has fully booted, any options that are required to set the image up, and features that the image supports.

Since settings for locating an image and interacting with it differ from platform to platform, there are 4 levels of settings available for images on top of the default image settings. The structure of the image config file is:

default_release_config:
    default:
        ...
    <platform>:
        ...
    <platform>:
        ...

releases:
    <release name>:
        <default>:
            ...
        <platform>:
            ...
        <platform>:
            ...

The base config is created from the overall defaults and the overrides for the platform. The overrides are created from the default config for the image and the platform specific overrides for the image.

System Boot

The test suite must be able to test if a system has fully booted and if cloud-init has finished running, so that running collect scripts does not race against the target image booting. This is done using the system_ready_script and cloud_init_ready_script image config keys.

Each of these keys accepts a small bash test statement as a string that must return 0 or 1. Since this test statement will be added into a larger bash statement it must be a single statement using the [ test syntax.

The default image config provides a system ready script that works for any systemd based image. If the image is not systemd based, then a different test statement must be provided. The default config also provides a test for whether or not cloud-init has finished which checks for the file /run/cloud-init/result.json. This should be sufficient for most systems as writing this file is one of the last things cloud-init does.

The setting boot_timeout controls how long, in seconds, the platform should wait for an image to boot. If the system ready script has not indicated that the system is fully booted within this time an error will be raised.

Feature Flags

Not all test cases can work on all images due to features the test case requires not being present on that image. If a test case requires features in an image that are not likely to be present across all distros and platforms that the test suite supports, then the test can be skipped everywhere it is not supported.

Feature flags, which are names for features supported on some images, but not all that may be required by test cases. Configuration for feature flags is provided in releases.yaml under the features top level key. The features config includes a list of all currently defined feature flags, their meanings, and a list of feature groups.

Feature groups are groups of features that many images have in common. For example, the Ubuntu_specific feature group includes features that should be present across most Ubuntu releases, but may or may not be for other distros. Feature groups are specified for an image as a list under the key feature_groups.

An image’s feature flags are derived from the features groups that that image has and any feature overrides provided. Feature overrides can be specified under the features key which accepts a dictionary of {<feature_name>: true/false} mappings. If a feature is omitted from an image’s feature flags or set to false in the overrides then the test suite will skip any tests that require that feature when using that image.

Feature flags may be overridden at run time using the --feature-override command line argument. It accepts a feature flag and value to set in the format <feature name>=true/false. Multiple --feature-override flags can be used, and will all be applied to all feature flags for images used during a test.

Setup Overrides

If an image requires some of the options for image setup to be used, then it may specify overrides for the command line arguments passed into setup image. These may be specified as a dictionary under the setup_overrides key. When an image is set up, the arguments that control how it is set up will be the arguments from the command line, with any entries in setup_overrides used to override these arguments.

For example, images that do not come with cloud-init already installed should have setup_overrides: {upgrade: true} specified so that in the event that no additional setup options are given, cloud-init will be installed from the image’s repos before running tests. Note that if other options such as --deb are passed in on the command line, these will still work as expected, since apt’s policy for cloud-init would prefer the locally installed deb over an older version from the repos.

Platform Specific Options

There are many platform specific options in image configuration that allow platforms to locate images and that control additional setup that the platform may have to do to make the image usable. For information on how these work, please consult the documentation for that platform in the integration testing suite and the releases.yaml file for examples.

Error Handling

The test suite makes an attempt to run as many tests as possible even in the event of some failing so that automated runs collect as much data as possible. In the event that something goes wrong while setting up for or running a test, the test suite will attempt to continue running any tests which have not been affected by the error.

For example, if the test suite was told to run tests on one platform for two releases and an error occurred setting up the first image, all tests for that image would be skipped, and the test suite would continue to set up the second image and run tests on it. Or, if the system does not start properly for one test case out of many to run on that image, that test case will be skipped and the next one will be run.

Note that if any errors occur, the test suite will record the failure and where it occurred in the result data and write it out to the specified result file.

Results

The test suite generates result data that includes how long each stage of the test suite took and which parts were and were not successful. This data is dumped to the log after the collect and verify stages, and may also be written out in yaml format to a file. If part of the setup failed, the traceback for the failure and the error message will be included in the result file. If a test verifier finds a problem with the collected data from a test run, the class, test function and test will be recorded in the result data.

Exit Codes

The test suite counts how many errors occur throughout a run. The exit code after a run is the number of errors that occurred. If the exit code is non-zero then something is wrong either with the test suite, the configuration for an image, a test case, or cloud-init itself.

Note that the exit code does not always directly correspond to the number of failed test cases, since in some cases, a single error during image setup can mean that several test cases are not run. If run is used, then the exit code will be the sum of the number of errors in the collect and verify stages.

Data Dir

When using run, the collected data is written into a temporary directory. In the event that all tests pass, this directory is deleted, but if a test fails or an error occurs, this data will be left in place, and a message will be written to the log giving the location of the data.