First you need to get familiar with the Architecture guide, which explains from a high perspective how everything works together.
After our experience building the Enterprise MongoDB Kubernetes Operator, we have realized that it is very important to have a clean environment to work, and as such we have adopted a strategy that makes it easier for everyone to contribute.
This strategy is based on using
envtest
for setting up the tests
and go test
for running the tests, and making the test-runner itself run as a Kubernetes Pod. This
makes it easier to run the tests in environments with access to a Kubernetes
cluster with no go toolchain installed locally, making it easier to reproduce
our local working environments in CI/CD systems.
The operator itself consists of 1 image, that has all the operational logic to deploy and maintain the MongoDBCommunity resources in your cluster.
The operator deploys MongoDB in Pods (via a higher-level resource, a StatefulSet), on each Pod there will be multiple images coexisting during the lifetime of the MongoDB server.
-
Agent image: This image includes a binary provided by MongoDB that handles the local operation of a MongoDB server given a series of configurations provided by the operator. The configuration exists as a ConfigMap that's created by the operator and mounted in the Agent's Pod.
-
MongoDB image: Docker image that includes the MongoDB server.
-
Version upgrade post-start hook image: This image includes a binary that helps orchestrate the restarts of the MongoDB Replica Set members, in particular, when dealing with version upgrades, which requires a very precise set of operations to allow for seamless upgrades and downgrades, with no downtime.
Each Pod holds a member of a Replica Set, and each Pod has different components, each one of them in charge of some part of the lifecycle of the MongoDB database.
-
Please ensure you have signed our Contributor Agreement. You can find it here.
-
Please ensure that all commits are signed.
The operator is built using golang
. We use a simple
json file that describe some local options that you need to set for the testing environment
to be able to run properly. Create a json file with the following content:
{
"namespace": "mongodb",
"repo_url": "localhost:5000",
"operator_image": "mongodb-kubernetes-operator",
"e2e_image": "community-operator-e2e",
"version_upgrade_hook_image": "mongodb-kubernetes-operator-version-upgrade-post-start-hook",
"agent_image": "mongodb-agent-ubi-dev",
"readiness_probe_image": "mongodb-kubernetes-readinessprobe",
"s3_bucket": ""
}
namespace
is the namespace that will be used by scripts/tooling. All the resources will be deployed here.repo_url
the repository that should be used to push/pull all images.operator_image
will be used as the name of the operator deployment, and the name of the operator image when build.e2e_image
the name of e2e test image that will be built.version_upgrade_hook_image
the name of the version upgrade post start hook image.agent_image
the name of the agent image.s3_bucket
the S3 bucket that Dockerfiles will be pushed to as part of the release process. Note: this is only required when running the release tasks locally.
You can set the MONGODB_COMMUNITY_CONFIG
environment variable to be the absolute path of this file.
It will default to ~/.community-operator-dev/config.json
Please see here to see how to build and deploy the operator locally.
The build process consists of multiple Docker images being built. You need to specify where you want the locally built images to be pushed. The Docker registry needs to be accessible from your Kubernetes cluster.
For local testing you can use a local Kind cluster.
You can change the namespace used for tests, if you are using Kind
, for
instance, you can leave this as mongodb
.
The test runner is a Python script, in order to use it a virtualenv needs to be created.
Python 3.9 is not supported yet. Please use Python 3.8.
python -m venv venv
source venv/bin/activate
python -m pip install -r requirements.txt
- create a python environment and install dependencies.
pipenv install -r requirements.txt
- activate the python environment.
pipenv shell
In order to deploy the Operator from source, you can run the following command.
make operator-image deploy
This will build and deploy the operator to namespace specified in your configuration file.
If you are using a local docker registry you should run the following command.
The additional IMG_BUILD_ARGS=--insecure
variable will add the --insecure
flag to the command creating the manifests.
This is necessary if your local registry is not secure. Read more about the flag on the documentatio
IMG_BUILD_ARGS=--insecure make operator-image deploy
kubectl get pods
Follow the steps outlined here to deploy some resources.
To remove the operator and any created resources you can run
make undeploy
Alternatively, you can run the operator locally. Make sure you follow the steps outlined in run-operator-locally.md
make run
Unit tests should be run from the root of the project with:
make test
If this is the first time running E2E tests, you will need to ensure that you have built and pushed
all images required by the E2E tests. You can do this by running the following command,
or with the additional IMG_BUILD_ARGS=--insecure
described above.
make all-images
For subsequent tests you can use
make e2e-k8s test=<test-name>
This will only re-build the e2e test image. Add IMG_BUILD_ARGS=--insecure
if necessary
We have built a simple mechanism to run E2E tests on your cluster using a runner that deploys a series of Kubernetes objects, runs them, and awaits for their completion. If the objects complete with a Success status, it means that the tests were run successfully.
The available tests can be found in the tests/e2e
directory, at the time of this
writing we have:
$ ls -l test/e2e
replica_set
replica_set_change_version
replica_set_readiness_probe
replica_set_scale
...
The tests should run individually using the runner like this, or additionally with IMG_BUILD_ARGS=--insecure
:
make e2e-k8s test=replica_set
This will run the replica_set
E2E test which is a simple test which installs a
MongoDB Replica Set and asserts that the deployed server can be connected to.
make e2e-telepresence test=<test-name>
This method uses telepresence to allow connectivity as if your local machine is in the kubernetes cluster,
there will be full MongoDB connectivity using go test
locally.
Note: you must install telepresence before using this method.
If on MacOS, you can run make install-prerequisites-macos
which will perform the installation.
Run a single test
make e2e-gh test=<test-name>
Run all tests.
- Navigate to the Actions tab on the github repo
Run E2E
>Run Workflow
>Your Branch
Note: the code must be pushed to a remote branch before this will work.
When you run a test locally, if the e2e-test
pod is present, you will have to
first manually delete it; failing to do so will cause the e2e-test
pod to fail.
You can start with the replica_set
test as a starting point to write a new test.
The tests are written using envtest
and they are run using go test
.
Adding a new test is as easy as creating a new directory in test/e2e
with the
new E2E test, and to run them:
make e2e test=<test-name>
To set up the pre-commit hooks, please create symbolic links from the provided hooks:
-
Navigate to your
.git/hooks
directory:cd .git/hooks
-
Create a symlink for every file in the
scripts/git-hooks
directory:ln -s -f ../../scripts/git-hooks/* .