> This repository is currently an example of how jobs may be structured. It is > not complete and requires a lot more work. However, given the other paths > that we have been exploring this is a similar amount of work.
smoke_rook is a testing platform for rook.io. The intention is to provide developers with a way to simulate various environments and scenarios that may occur within them.
For example, smoke_rook can perform tests such as adding new nodes to your kubernetes cluster and ensuring that they are correctly enrolled and handled by rook.io.
Additionally smoke_rook can handle disaster testing such as kernel panics, physically removed nodes, and so forth.
Because a test may need to interact with the underlying hardware the unit tests will set up and configure the nodes, distros, kubernetes, and rook itself. These are then exposed to the test writer to interact with further or to verify the environment.
smoke_rook requires VM's from libcloud to set up and perform the tests against.
Installing requirements:
# zypper in python-pip docker
# systemctl start docker
# pip install tox
$ PROFILE=libvirt
$ tox -e bindep ${PROFILE}
# zypper in <indicated missing package names>
You will need to configure the platform that the tests are ran against:
cp configuration.env my.env vim my.env # Make any changes needed source my.env
If you are using OpenStack you can use your openrc for most of the configuratoin.
Running tests:
$ tox -e py37
A OpenStack network needs to be available for usage. The network name needs to be exported as:
export OS_INTERNAL_NETWORK=my-test-net
If the network is not available, one can be created via:
_OS_SUBNET=`echo $OS_INTERNAL_NETWORK|sed -e 's/-net/-subnet/'`
_OS_ROUTER=`echo $OS_INTERNAL_NETWORK|sed -e 's/-net/-router/'`
openstack network create ${OS_INTERNAL_NETWORK}
openstack subnet create --network ${OS_INTERNAL_NETWORK} --subnet-range 192.168.100.0/24 ${_OS_SUBNET}
openstack router create ${_OS_ROUTER}
openstack router set --external-gateway floating ${_OS_ROUTER}
where floating is the name of the external network.
- smoke_rook will remove and manage known host keys on the test runner, which may include removing legitimate entries.
Currently there are [at least] 4 abstractions that need to be available:
- Hardware (VM's, etc),
- Operating Systems (packages/configuration etc),
- Kubernetes (deployment/packages etc),
- Rook (packaging etc).
To begin with, each of these is being implemented targeting OpenStack, openSUSE, Upstream Kubernetes, and Upstream rook.io respectfully. It is intended that each of these are easy to swap out for other platforms depending on the testing environment. Therefore the code is being written in a generic/pluggable way.
- Uses pytest
- Each aforementioned abstraction is set up as a pytest fixture
- tests/conftest.py sets up the required fixtures
- The fixtures are generally scoped to the module
- This means a file such as test/test_my_grouped_tests.py can do serial tests against the same cluster
- When the fixtures are 'exited' they clean up their resources
- Tests are thread-safe at a module level. Each test module will have its own deployment created to perform tests against.
TODO some examples
Use either libcloud or kcli to abstract away the hardware. Extend either library for our needs.
- kcli: Will likely need fixes for how auth against OpenStack works to be
- compatible with ECP's domain/project_domain. I also ran into a bug testing against libvirt networks that will require more exploration.
- libcloud: Needs a new release to get
- apache/libcloud#1365. The libvirt driver needs extending to be able to create and destroy vms, images, volumes, and networks. This is a large amount of work, but can also be pulled from existing projects such as kcli or python-libvirt.
- Create a library for deploying kubernetes on provided nodes.
- This would be an ABC with implementations for Vanilla Kubenetes, CaaSP and so on. The hardest part will be the underlying operating systems a various deployment will support. It may be enough to raise an error if the OS is not compatible.
- Create a library for deploying rook.io on said kubernetes cluster.
- This will likely need some plugability to change things such as container registries.
Each Hardware, Kubenetes, and Rook deployments are py.test fixtures. As we can scope those to a module we can write tests that reuse the same deployments rather than setting up new nodes for each individual test.
We can also eventually break things out of py.test to allow devs to build and debug clusters etc. as well as providing tools for checking any rogue resources left behind by tests and so on.