Provides a collection of API's to support deploying and serving of serverless applications and functions.
For the scripts in this repository to work properly, the following requirements apply.
podman
aliased todocker
ordocker
(17.05 or newer) (only if you need/want to build images from code)bash
(4.0.0 or newer)make
helm
To install an unreleased version (either nightly from HEAD or the respective version
from an already cut branch), you can follow these steps. You can replace the crc start
step with whatever cluster you might want to use. Also make sure to check out the respective
branch of the version you want to install first.
$ crc start --cpus=6 --memory 16384
$ unset DOCKER_REPO_OVERRIDE # This makes sure that pre-built images are used
$ make install
If you want to quickly run the most relevant tests locally (those that are required by the CI environment), use these commands:
$ crc start --cpus=6 --memory 16384
$ export DOCKER_REPO_OVERRIDE=quay.io/username
$ make images test-operator
If you want to use CRC to run tests locally, the following configuration has been tested to work with the operator E2E tests.
crc start --cpus=6 --memory 16384
To test the Serverless Operator against your private Openshift cluster you first
need to push the necessary images to a publicly available location. To do that,
make sure the DOCKER_REPO_OVERRIDE
environment variable is set to a docker
repository you can push to, for example quay.io/markusthoemmes
. You might
need to run docker login
to be able to push images. Now run
make images
and all images in this repository will now be built and
pushed to your docker repository.
make images
requires docker or podman to build images, by setting
ON_CLUSTER_BUILDS=true
env variable, make images
will build images
using OpenShift Build and they will be available at the in-cluster
registry image-registry.openshift-image-registry.svc:5000/openshift-marketplace/<image_name>
.
To install the system using those images, use
DOCKER_REPO_OVERRIDE=image-registry.openshift-image-registry.svc:5000/openshift-marketplace
Use the appropriate make targets or scripts in hack
:
make dev
: Deploys the serverless-operator without deploying Knative Serving, Eventing and Kafka components.make install
: Scales the cluster appropriately, deploys serverless-operator, Knative Serving and Eventing.make install-all
: Scales the cluster appropriately, deploys serverless-operator, Knative Serving, Eventing and Kafka.make install-serving
: Scales the cluster appropriately, deploys serverless-operator and Knative Serving.make install-eventing
: Scales the cluster appropriately, deploys serverless-operator and Knative Eventing.make install-kafka
: Scales the cluster appropriately, deploys serverless-operator, Knative Eventing and Knative Kafka inknative-eventing
namespace by default. Requires to install a Strimzi cluster withmake install-strimzi
.make install-previous
: same asmake install
but deploy previous serverless-operator version.make install-strimzi
: Install the latest Strimzi operator and a kafka cluster instance inkafka
namespace by default.make unistall-strimzi
: Uninstall the Strimzi operator and any existing kafka cluster instance.make install-mesh
: Install service mesh operator, Istio Gateway and PeerAuthentication to use Knative Serving + Eventing for secure traffic.make uninstall-mesh
: Uninstall service mesh operator, Istio Gateway and PeerAuthentication.
Note: Don't forget you can chain make
targets. make images dev
is handy
for example.
To update release files with variables stored in project.yaml we use generate scripts and a templates. Update those based on your needs, and generate the files with:
make release-files
make test-unit
: Runs unit tests.make test-e2e
: Scales, installs and runs E2E tests (except for Knative Kafka components).make test-e2e-with-kafka
: Scales, installs and runs E2E tests (also tests Knative Kafka components).make install-mesh test-e2e
: Scales, installs and runs E2E tests, including ServiceMesh integration testsmake test-operator
: Runs unit and E2E tests.
make test-upstream-upgrade
: Installs aprevious
version of Serverless and runs Knative Serving upgrade tests. Requirements:- Running OCP cluster.
- Knative Serving images that the current Serverless operator depends on are published in CI registry. This requirement is automatically met when the respective branch in Knative Serving is created and its pre-submit CI checks run at least once.
- The path
${GOPATH}/src/knative.dev/serving
containing Knative Serving sources from the desired branch. This should be checked out before running tests.
make test-upstream-e2e-no-upgrade
: Installs the latest version of Serverless and runs Knative Serving and Knative Eventing E2E tests (without upgrades). Requirements:- Running OCP cluster.
- Knative Serving and Knative Eventing images that the current Serverless operator depends on are published in CI registry. This requirement is automatically met when the respective branches in Knative Serving and Knative Eventing are created, and their pre-submit CI checks run at least once.
- The path
${GOPATH}/src/knative.dev/serving
containing Knative Serving sources from the desired branch. The path${GOPATH}/src/knative.dev/eventing
containing Knative Eventing sources from the desired branch. This should be checked out before running tests.
There are targets for running individual tests in both Knative Serving Makefile and Knative Eventing Makefile.
Example targets that can be run from the respective repositories (these targets all requires a running OCP cluster and pre-installed Serverless):
make TEST=<name_of_test> BRANCH=<ci_promotion_name> test-e2e-local
: Runs the given test using the latest test images that were published under the given promotion name. This doesn't require building any images manually. Example:make BRANCH=knative-v0.15.2 TEST=TestDestroyPodInflight test-e2e-local
.make IMAGE=<name_of_image> DOCKER_REPO_OVERRIDE=<dockerhub_registry> test-image-single
: Builds an image fromtest/test_images/$(IMAGE)
and pushes it into the Dockerhub registry. This requires ko 0.2.0 or newer.make TEST=<name_of_test> DOCKER_REPO_OVERRIDE=<dockerhub_registry> test-e2e-local
: Runs the given test using the previously built image.
This repository contains the metadata required by the Operator Lifecycle Manager
To install the operator, create a subscription:
cat <<-EOF | oc apply -f -
apiVersion: operators.coreos.com/v1alpha1
kind: Subscription
metadata:
name: serverless-operator
generateName: serverless-operator-
namespace: openshift-operators
spec:
source: serverless-operator
sourceNamespace: openshift-marketplace
name: serverless-operator
channel: techpreview
EOF
To test upgrade from previous version, deploy serverless operator by make install-previous
make install-previous
Then, you can see the installplans that the latest version has APPROVED false
.
$ oc get installplan -n openshift-operators
NAME CSV APPROVAL APPROVED
install-ck6jl serverless-operator.v1.4.1 Manual true
install-hrzzn serverless-operator.v1.5.0 Manual false
For example, v1.5.0 is the latest version in this case. To upgrade v1.5.0, you can edit
spec.approved
to true manually.
spec:
approval: Manual
approved: true
After a few minutes, operators will be upgraded automatically.
To create a new version of the serverless-operator (usually after a release branch has been cut), there are a few steps we have to do. These steps can be done one after the other and do not have to be sent as one PR, to avoid clogging up the respective PR.
The first thing we usually do is to update the version metadata of the operator. That
mostly includes bumping the version to the next desired version (i.e. 1.12 -> 1.13).
This is done by adjusting the respective settings in the project
and olm
part of
project.yaml
. The settings to be
changed usually are project.version
, olm.replaces
and olm.skipRange
.
Next, add the now outdated version of serverless-operator to the CatalogSource deployment
in catalogsource.bash. The image to be added usually has
the following format: registry.ci.openshift.org/openshift/openshift-serverless-$OLD_VERSION:serverless-bundle
.
Add it before the "current" image, which is image-registry.openshift-image-registry.svc:5000/$OLM_NAMESPACE/serverless-bundle
.
After the changes are done, commit them and run make generated-files
. All manifests
will now be updated accordingly. It's encouraged to commit the generated changes
separately, to ease review.
To update the image coordinates of the component you want to bump, adjust its version in
the dependencies
section of project.yaml
.
It should be a rare occasion, but between releases, the manifest files we want to pull
might have changed. If that is the case, adjust the files downloaded in
update-manifests.sh
.
Likewise a rare occasion should be patches to the manifest files. update-manifests.sh
might be applying patches that can be removed in the new release or have to be adjusted.
Make sure to review them and act accordingly.
After the changes are done, commit them and run make generated-files
. All manifests
will now be updated accordingly. It's encouraged to commit the generated changes
separately, to ease review.
The repository itself depends on the respective upstream releases. The
openshift-knative-operator
for example is straightly build from vendoring the upstream
operator and the tests heavily rely on upstream clients, APIs and helpers.
To update the dependencies, update the KN_VERSION
variable in
update-deps.sh
. Then run ./hack/update-deps.sh --upgrade
(like we do upstream) to pull in all the correct versions. It's encouraged to commit
the generated changes separately, to ease review.
If we're bumping our minimum supported Openshift version, bump the OCP_VERSION
variable
in the same file and follow the same process.
To run the linters that CI is running, you can use make lint
. You can run make fix-lint
to fix as many linting warnings automatically as possible. This will for example format YAML
rather than just warn about unformatted YAML.
The required linters for that are:
woke
to detect non-inclusive languagegolangci-lint
to lint golang codeshellcheck
to lint shell filesoperator-sdk
to lint the bundle filesmisspell
to lint typosprettier
to format YAML