5. Developer’s Guide

Note

ISC acknowledges that users and developers have different needs, so the user and developer documents should eventually be separated. However, since the project is still in its early stages, this section is kept in the Stork ARM for convenience.

5.1. Rakefile

Rakefile is a script for performing many development tasks, like building source code, running linters, running unit tests, and running Stork services directly or in Docker containers.

There are several other Rake targets. For a complete list of available tasks, use rake -T. Also see the Stork wiki for detailed instructions.

5.2. Generating Documentation

To generate documentation, simply type rake doc. Sphinx and rtd-theme must be installed. The generated documentation will be available in the doc/singlehtml directory.

5.3. Setting Up the Development Environment

The following steps install Stork and its dependencies natively, i.e., on the host machine, rather than using Docker images.

First, PostgreSQL must be installed. This is OS-specific, so please follow the instructions from the Installation chapter.

Once the database environment is set up, the next step is to build all the tools. Note that the first command below downloads some missing dependencies and installs them in a local directory. This is done only once and is not needed for future rebuilds, although it is safe to rerun the command.

$ rake build_backend
$ rake build_ui

The environment should be ready to run. Open three consoles and run the following three commands, one in each console:

$ rake run_server
$ rake serve_ui
$ rake run_agent

Once all three processes are running, connect to http://localhost:8080 via a web browser. See Using Stork for information on initial password creation or addition of new machines to the server.

The run_agent runs the agent directly on the current operating system, natively; the exposed port of the agent is 8888.

There are other Rake tasks for running preconfigured agents in Docker containers. They are exposed to the host on specific ports.

When these agents are added as machines in the Stork Server UI, both a localhost address and a port specific to a given container must be specified. The list of containers can be found in the Docker Containers for Development section.

5.3.1. Installing Git Hooks

There is a simple git hook that inserts the issue number in the commit message automatically; to use it, go to the utils directory and run the git-hooks-install script. It copies the necessary file to the .git/hooks directory.

5.4. Agent API

The connection between the Stork server and the agents is established using gRPC over http/2. The agent API definition is kept in the backend/api/agent.proto file. For debugging purposes, it is possible to connect to the agent using the grpcurl tool. For example, a list of currently provided gRPC calls may be retrieved with this command:

$ grpcurl -plaintext -proto backend/api/agent.proto localhost:8888 describe
agentapi.Agent is a service:
service Agent {
  rpc detectServices ( .agentapi.DetectServicesReq ) returns ( .agentapi.DetectServicesRsp );
  rpc getState ( .agentapi.GetStateReq ) returns ( .agentapi.GetStateRsp );
  rpc restartKea ( .agentapi.RestartKeaReq ) returns ( .agentapi.RestartKeaRsp );
}

Specific gRPC calls can also be made. For example, to get the machine state, use the following command:

$ grpcurl -plaintext -proto backend/api/agent.proto localhost:8888 agentapi.Agent.getState
{
  "agentVersion": "0.1.0",
  "hostname": "copernicus",
  "cpus": "8",
  "cpusLoad": "1.68 1.46 1.28",
  "memory": "16",
  "usedMemory": "59",
  "uptime": "2",
  "os": "darwin",
  "platform": "darwin",
  "platformFamily": "Standalone Workstation",
  "platformVersion": "10.14.6",
  "kernelVersion": "18.7.0",
  "kernelArch": "x86_64",
  "hostID": "c41337a1-0ec3-3896-a954-a1f85e849d53"
}

5.5. REST API

The primary user of the REST API is the Stork UI in a web browser. The definition of the REST API is located in the api folder and is described in Swagger 2.0 format.

The description in Swagger is split into multiple files. Two files comprise a tag group:

  • *-paths.yaml - defines URLs
  • *-defs.yaml - contains entity definitions

All these files are combined by the yamlinc tool into a single Swagger file, swagger.yaml. Then swagger.yaml generates code for:

  • the UI fronted by swagger-codegen
  • the backend in Go lang by go-swagger

All these steps are accomplished by Rakefile.

5.6. Backend Unit Tests

There are unit tests for the Stork agent and server backends, written in Go. They can be run using Rake:

$ rake unittest_backend

This requires preparing a database in PostgreSQL. One way to avoid doing this manually is by using a Docker container with PostgreSQL, which is automatically created when running the following Rake task:

$ rake unittest_backend_db

This task spawns a container with PostgreSQL in the background which then runs unit tests. When the tests are completed, the database is shut down and removed.

5.6.1. Unit Tests Database

When a Docker container with a database is not used for unit tests, the PostgreSQL server must be started and the following role must be created:

postgres=# CREATE USER storktest WITH PASSWORD 'storktest';
CREATE ROLE
postgres=# ALTER ROLE storktest SUPERUSER;
ALTER ROLE

To point unit tests to a specific Stork database, set the POSTGRES_ADDR environment variable, e.g.:

$ rake unittest_backend POSTGRES_ADDR=host:port

By default it points to localhost:5432.

Similarly, if the database setup requires a password other than the default storktest, the STORK_DATABASE_PASSWORD variable can be used by issuing the following command:

$ rake unittest_backend STORK_DATABASE_PASSWORD=secret123

Note that there is no need to create the storktest database itself; it is created and destroyed by the Rakefile task.

5.6.2. Unit Tests Coverage

A coverage report is presented once the tests have executed. If coverage of any module is below a threshold of 35%, an error is raised.

5.6.3. Benchmarks

Benchmarks are part of the backend unit tests. They are implemented using the golang “testing” library and they test performance-sensitive parts of the backend. Unlike unit tests, the benchmarks do not return pass/fail status. They measure average execution time of functions and print the results to the console.

In order to run unit tests with benchmarks, the benchmark environment variable must be specified as follows:

$ rake unittest_backend benchmark=true

This command runs all unit tests and all benchmarks. Running benchmarks without unit tests is possible using the combination of the benchmark and test environment variables:

$ rake unittest_backend benchmark=true test=Bench

Benchmarks are useful to test the performance of complex functions and find bottlenecks. When working on improving the performance of a function, examining a benchmark result before and after the changes is a good practice to ensure that the goals of the changes are achieved.

Similarly, adding a new logic to a function often causes performance degradation, and careful examination of the benchmark result drop for that function may be a driver for improving efficiency of the new code.

5.6.4. Short Testing Mode

It is possible to filter out long running unit tests. Set the short variable to true on the command line:

$ rake unittest_backend short=true

5.7. Web UI Unit Tests

Stork offers web UI tests, to take advantage of the unit-tests generated automatically by Angular. The simplest way to run these tests is by using Rake tasks:

rake build_ui
rake ng_test

The tests require the Chromium (on Linux) or Chrome (on Mac) browser. The rake ng_test task attempts to locate the browser binary and launch it automatically. If the browser binary is not found in the default location, the Rake task returns an error. It is possible to set the location manually by setting the CHROME_BIN environment variable; for example:

export CHROME_BIN=/usr/local/bin/chromium-browser
rake ng_test

By default, the tests launch the browser in headless mode, in which test results and any possible errors are printed in the console. However, in some situations it is useful to run the browser in non-headless mode because it provides debugging features in Chrome’s graphical interface. It also allows for selectively running the tests. Run the tests in non-headless mode using the debug variable appended to the rake command:

rake ng_test debug=true

That command causes a new browser window to open; the tests run there automatically.

The tests are run in random order by default, which can make it difficult to chase the individual errors. To make debugging easier by always running the tests in the same order, click Debug in the new Chrome window, then click Options and unset the “run tests in random order” button. A specific test can be run by clicking on its name.

When adding a new component or service with ng generate component|service …, the Angular framework adds a .spec.ts file with boilerplate code. In most cases, the first step in running those tests is to add the necessary Stork imports. If in doubt, refer to the commits on https://gitlab.isc.org/isc-projects/stork/-/merge_requests/97. There are many examples of ways to fix failing tests.

5.8. System Tests

System tests for Stork are designed to test the software in a distributed environment. They allow for testing several Stork servers and agents running at the same time in one test case, inside LXD containers. It is possible to set up Kea (and eventually, BIND 9) services along with Stork agents. The framework enables experimenting in containers so custom Kea configurations can be deployed or specific Kea daemons can be stopped.

The tests can use the Stork server REST API directly or the Stork web UI via Selenium.

5.8.1. Dependencies

System tests require:

5.8.2. LXD Installation

The easiest way to install LXD is to use snap. First, install snap.

On Fedora:

$ sudo dnf install snapd

On Ubuntu:

$ sudo apt install snapd

Then install LXD:

$ sudo snap install lxd

And then add the user to lxd group:

$ sudo usermod -a -G lxd $USER

Now log in again to make the user’s presence in lxd group visible in the shell session.

After installing LXD, it requires initialization. Run:

$ lxd init

and then for each question press Enter, i.e., use the default values:

Would you like to use LXD clustering? (yes/no) [default=no]: **Enter**
Do you want to configure a new storage pool? (yes/no) [default=yes]: **Enter**
Name of the new storage pool [default=default]: **Enter**
Name of the storage backend to use (dir, btrfs) [default=btrfs]: **Enter**
Would you like to create a new btrfs subvolume under /var/snap/lxd/common/lxd? (yes/no) [default=yes]: **Enter**
Would you like to connect to a MAAS server? (yes/no) [default=no]:  **Enter**
Would you like to create a new local network bridge? (yes/no) [default=yes]:  **Enter**
What should the new bridge be called? [default=lxdbr0]:  **Enter**
What IPv4 address should be used? (CIDR subnet notation, "auto" or "none") [default=auto]:  **Enter**
What IPv6 address should be used? (CIDR subnet notation, "auto" or "none") [default=auto]:  **Enter**
Would you like LXD to be available over the network? (yes/no) [default=no]:  **Enter**
Would you like stale cached images to be updated automatically? (yes/no) [default=yes]  **Enter**
Would you like a YAML "lxd init" preseed to be printed? (yes/no) [default=no]:  **Enter**

More details can be found at: https://linuxcontainers.org/lxd/getting-started-cli/

The subvolume is stored in /var/snap/lxd/common/lxd, and is used to store images and containers. If the space is exhausted, it is not possible to create new containers. This is not connected with total disk space but rather with the space in this subvolume. To free space, remove stale images or stopped containers. Basic usage of LXD is presented at: https://linuxcontainers.org/lxd/getting-started-cli/#lxd-client

5.8.2.1. LXD troubleshooting on Arch

Problem: After running lxd init you get this message

Error: Failed to connect to local LXD: Get "http://unix.socket/1.0": dial unix /var/lib/lxd/unix.socket: connect: no such file or directory

Solution: You should restart lxd daemon:

sudo systemctl restart lxd

Problem: After running rake system_tests you get these message (tail):

************ START   tests.py::test_users_management[ubuntu/18.04-centos/7] **************************************************************

stork-agent-ubuntu-18-04-gw0: {'fg': 'yellow', 'style': ''}
stork-server-centos-7-gw0: {'fg': 'red', 'style': 'bold'}

and nothing more happens, CPU and RAM usage by lxd are ~0%.

Solution: original post

  1. Create /etc/subuid file with content:
root:1000000:65536
  1. Create /etc/subuid with the same content
  2. Add these lines to /etc/default/lxc:
lxc.idmap = u 0 100000 65536
lxc.idmap = g 0 100000 65536

5.8.3. Running System Tests

After preparing all the dependencies, it is possible to start tests. But first, the RPM and deb Stork packages need to be prepared. This can be done with this Rake task:

$ rake build_pkgs_in_docker

When using packages, the tests can be invoked by the following Rake task:

$ rake system_tests

This command first prepares the Python virtual environment (venv) where pytest and other Python dependencies are installed. pytest is a Python testing framework that is used in Stork system tests.

At the end of the logs are listed test cases with their result status.

The tests can be invoked directly using pytest, but first the directory must be changed to tests/system:

$ cd tests/system
$ ./venv/bin/pytest --tb=long -l -r ap -s tests.py

The switches passed to pytest are:

  • --tb=long: in case of failures, present long format of traceback
  • -l: show values of local variables in tracebacks
  • -r ap: at the end of execution, print a report that includes (p)assed and (a)ll except passed (p)

To run a particular test case, add it just after test.py:

$ ./venv/bin/pytest --tb=long -l -r ap -s tests.py::test_users_management[centos/7-ubuntu/18.04]

To get a list of tests without actually running them, the following command can be used:

$ ./venv/bin/pytest --collect-only tests.py

The names of all available tests are printed as <Function name_of_the_test>.

A single test case can be run using a rake task with the test variable set to the test name:

$ rake system_tests test=tests.py::test_users_management[centos/7-ubuntu/18.04]

5.8.4. Developing System Tests

System tests are defined in tests.py and other files that start with test_. There are two other files that define the framework for Stork system tests:

  • conftest.py - defines hooks for pytests
  • containers.py - handles LXD containers: starting/stopping; communication, such as invoking commands; uploading/downloading files; installing and preparing Stork Agent/Server and Kea; and other dependencies that they require.

Most tests are constructed as follows:

@pytest.mark.parametrize("agent, server", SUPPORTED_DISTROS)
def test_machines(agent, server):
    # login to stork server
    r = server.api_post('/sessions',
                        json=dict(useremail='admin', userpassword='admin'),
                        expected_status=200)
    assert r.json()['login'] == 'admin'

    # add machine
    machine = dict(
        address=agent.mgmt_ip,
        agentPort=8080)
    r = server.api_post('/machines', json=machine, expected_status=200)
    assert r.json()['address'] == agent.mgmt_ip

    # wait for application discovery by Stork Agent
    for i in range(20):
        r = server.api_get('/machines')
        data = r.json()
        if len(data['items']) == 1 and \
           len(data['items'][0]['apps'][0]['details']['daemons']) > 1:
            break
        time.sleep(2)

    # check discovered application by Stork Agent
    m = data['items'][0]
    assert m['apps'][0]['version'] == '1.7.3'

It may be useful to explain each part of this code.

@pytest.mark.parametrize("agent, server", SUPPORTED_DISTROS)

This indicates that the test is parameterized: there will be one or more instances of this test in execution for each set of parameters.

The constant SUPPORTED_DISTROS defines two sets of operating systems for testing:

SUPPORTED_DISTROS = [
    ('ubuntu/18.04', 'centos/7'),
    ('centos/7', 'ubuntu/18.04')
]

The first set indicates that for the Stork agent Ubuntu 18.04 should be used in the LXD container, and for the Stork server CentOS 7. The second set is the opposite of the first one.

The next line:

def test_machines(agent, server):

defines the test function. Normally, the agent and server argument would get the text values 'ubuntu/18.04' and 'centos/7', but a hook exists in the pytest_pyfunc_call() function of conftest.py that intercepts these arguments and uses them to spin up LXD containers with the indicated operating systems. This hook also collects Stork logs from these containers at the end of the test and stores them in the test-results folder for later analysis if needed.

Instead of text values, the hook replaces the arguments with references to actual LXC container objects, so that the test can interact directly with them. Besides substituting the agent and server arguments, the hook intercepts any argument that starts with agent or server. This allows multiple agents in the test, e.g. agent1, agent_kea, or agent_bind9.

Next, log into the Stork server using its REST API:

# login to stork server
r = server.api_post('/sessions',
                    json=dict(useremail='admin', userpassword='admin'),
                    expected_status=200)
assert r.json()['login'] == 'admin'

Then, add a machine with a Stork agent to the Stork server:

# add machine
machine = dict(
    address=agent.mgmt_ip,
    agentPort=8080)
r = server.api_post('/machines', json=machine, expected_status=200)
assert r.json()['address'] == agent.mgmt_ip

A check then verifies the returned address of the machine.

After a few seconds, the Stork agent detects the Kea application and reports it to the Stork server. The server is periodically polled for updated information about the Kea application.

# wait for application discovery by Stork Agent
for i in range(20):
    r = server.api_get('/machines')
    data = r.json()
    if len(data['items']) == 1 and \
       len(data['items'][0]['apps'][0]['details']['daemons']) > 1:
        break
    time.sleep(2)

Finally, the returned data about Kea can be verified:

# check discovered application by Stork Agent
m = data['items'][0]
assert m['apps'][0]['version'] == '1.7.3'

5.9. Docker Containers for Development

To ease development, there are several Docker containers available. These containers are used in the Stork demo and are fully described in the Demo chapter.

The following Rake tasks start these containers.

Rake tasks for managing development containers.
Rake Task Description
rake build_kea_container Build a container agent-kea with a Stork agent and Kea with DHCPv4.
rake run_kea_container Start an agent-kea container. Published port is 8888.
rake build_kea6_container Build an agent-kea6 container with a Stork agent and Kea with DHCPv6.
rake run_kea6_container Start an agent-kea6 container. Published port is 8886.
rake build_kea_ha_containers Build two containers, agent-kea-ha1 and agent-kea-ha2, that are configured to work together in High Availability mode, with Stork agents, and Kea with DHCPv4.
rake run_kea_ha_containers Start the agent-kea-ha1 and agent-kea-ha2 containers. Published ports are 8881 and 8882.
rake build_kea_hosts_container Build an agent-kea-hosts container with a Stork agent and Kea with DHCPv4 with host reservations stored in a database. This requires premium features.
rake run_kea_hosts_container Start the agent-kea-hosts container. This requires premium features.
rake build_bind9_container Build an agent-bind9 container with a Stork agent and BIND 9.
rake run_bind9_container Start an agent-bind9 container. Published port is 9999.

5.10. Packaging

There are scripts for packaging the binary form of Stork. There are two supported formats: RPM and deb.

The RPM package is built on the latest CentOS version. The deb package is built on the latest Ubuntu LTS.

There are two packages built for each system: a server and an agent.

Rake tasks can perform the entire build procedure in a Docker container: build_rpms_in_docker and build_debs_in_docker. It is also possible to build packages directly in the current operating system; this is provided by the deb_agent, rpm_agent, deb_server, and rpm_server Rake tasks.

Internally, these packages are built by FPM (https://fpm.readthedocs.io/). The containers that are used to build packages are prebuilt with all dependencies required, using the build_fpm_containers Rake task. The definitions of these containers are placed in docker/pkgs/centos-8.txt and docker/pkgs/ubuntu-18-04.txt.

5.11. Implementation details

5.11.1. Agent Registration Process

The diagram below shows a flowchart of the agent registration process in Stork. It merely demonstrates the successful registration path. First Certificate Signing Request (CSR) is generated using an existing or new private key and agent token. The CSR, server token (optional), and agent token are sent to the Stork server. A successful server response contains a signed agent certificate, a server CA certificate, and an assigned Machine ID. If the agent was already registered with the provided agent token, only the assigned machine ID is returned without new certificates. The agent uses the returned machine ID to verify that the registration was successful.

_images/registration-agent.svg

Agent registration