A GitHub Action to run lightweight ephemeral Kubernetes clusters during workflow. Fundamental advantage of this action is a full customization of embedded k3s clusters. In addition, it provides multi-cluster support.
Applications running on Kubernetes clusters (microservices, controllers,...) come with their own set of complexities and concerns. In particular, E2E testing k8s based applications requires new approaches to confirm proper operation and continued availability under heavy load or in the face of resource failure. AbsaOSS/k3d-action allows to test the overall application functionality. For instance, the E2E use-case is operator testing in AbsaOSS/k8gb.
The full CI/CD pipeline tutorial with k3d-action by Viktor Farcic.
AbsaOSS/k3d-action runs k3d which is a lightweight wrapper to run k3s (Rancher Lab’s minimal Kubernetes distribution) in containers. Thanks to that, we could spin up the test environment quickly with minimal memory requirements, which is especially important in multi-cluster environments.
AbsaOSS/k3d-action defines several input attributes and two outputs:
-
cluster-name(Required) Cluster name. -
args(Optional) list of k3d arguments defined by k3d command tree -
k3d-version(Optional) version of k3d. If not set, will be used version from mapping below.
Implementation of additional features brings complexity and sometimes it may happen that extra feature is broken in special cases.
To prevent potential issues, the k3d version is fixed according to the mapping below:
| k3d-action | k3d | k3s |
|---|---|---|
| v1.1.0 | v3.4.0 | rancher/k3s:v1.20.2-k3s1 |
| v1.2.0 | v4.2.0 | rancher/k3s:v1.20.2-k3s1 |
| v1.3.0 | v4.2.0 | rancher/k3s:v1.20.4-k3s1 |
| v1.4.0 | v4.4.1 | rancher/k3s:v1.20.8-k3s1 or set image explicitly |
| v1.5.0 | v4.4.7 | rancher/k3s:v1.21.2-k3s1 or set image explicitly |
| v2.0.0 | v5.1.0 | rancher/k3s:v1.22.3+k3s1 or set image explicitly |
| v2.1.0 | v5.2.2 | rancher/k3s:v1.21.7-k3s1 or set image explicitly |
Starting from k3d-action v2.1.0 users can explicitly set k3d version via k3d-version input e.g. k3d-version: v5.2.2 otherwise k3d uses default version according to the mapping above.
Starting from k3d-action v1.4.0 users can also explicitly set k3s image version via configuration or
argument e.g.--image docker.io/rancher/k3s:v1.20.4-k3s1 otherwise k3d uses default version according to the mapping above.
For further k3s details see:
- docker rancher/k3s
- github k3s/releases
Although AbsaOSS/k3d-action strongly supports multi-cluster. Single cluster scenarios are very popular. The minimum single-cluster configuration looks like this :
- uses: AbsaOSS/k3d-action@v2
name: "Create Single Cluster"
with:
cluster-name: "test-cluster-1"
args: --agents 1k3d creates a cluster with one worker node (with traefik and metrics services), one agent and one default load-balancer node. In real scenarios you might prefer to do some port mapping and disable default load balancer. Such an action would look like this:
- uses: AbsaOSS/k3d-action@v2
name: "Create Single Cluster"
with:
cluster-name: "test-cluster-1"
args: >-
-p "8083:80@agent:0:direct"
-p "8443:443@agent:0:direct"
-p "5053:53/udp@agent:0:direct"
--agents 3
--no-lb
--image docker.io/rancher/k3s:v1.20.4-k3s1
--k3s-arg "--no-deploy=traefik,servicelb,metrics-server@server:*"The created cluster exposes two TCP (:8083,:8443) and one UDP (:5053) ports. The cluster comprises one server, three
agents and no load balancers. k3s-server-argument
disable default traefik and metrics.
For more details see: Demo, Source
From v1.2.0 you can configure action via config files or mix arguments together with config files. This setup is useful when you want to share the configuration for local testing and testing within k3d-action.
- uses: ./
name: "Create single k3d Cluster"
with:
cluster-name: "test-cluster-1"
args: >-
--agents 1
--config=<path to config yaml>All you need to do is to place configuration file somewhere into your project. However, keep in mind, that command line arguments will always take precedence over configuration, so the previous example will result in only one agent, not three as configured.
apiVersion: k3d.io/v1alpha3
kind: Simple
image: docker.io/rancher/k3s:v1.20.5-k3s1
servers: 1
agents: 3 # The action will overwrite this by 1
ports:
- port: 0.0.0.0:80:80
nodeFilters:
- agent:0:direct
- port: 0.0.0.0:443:443
nodeFilters:
- agent:0:direct
- port: 0.0.0.0:5053:53/udp
nodeFilters:
- agent:0:direct
options:
k3d:
wait: true
timeout: "60s"
disableLoadbalancer: true
k3s:
extraArgs:
- arg: --no-deploy=traefik,servicelb,metrics-server
nodeFilters:
- server:*
kubeconfig:
updateDefaultKubeconfig: true
switchCurrentContext: trueFor more details see: Demo, Source action, Source config
k3d creates a bridge-network for each separate cluster or attaches the created cluster to an existing network.
When you create a cluster named test-cluster-1, k3d will automatically create a network
named k3d-test-cluster-1 with the range 172.18.0.0/16. When you create a second cluster
test-cluster-2, k3d automatically creates a network named k3d-test-cluster-2 with a
range of 172.19.0.0/16. Other clusters will have ranges 172.20.0.0/16,172.21.0.0/16 etc.
The following example creates a total of four clusters, the first two are created on
the network nw01, 172.18.0.0/16, the next two clusters are created on the network
nw02, 172.19.0.0/16.
- uses: AbsaOSS/k3d-action@v2
name: "Create 1st Cluster in 172.18.0.0/16"
with:
cluster-name: "test-cluster-1"
args: >-
-p "80:80@agent:0:direct"
-p "443:443@agent:0:direct"
-p "5053:53/udp@agent:0:direct"
--agents 3
--no-lb
--k3s-arg "--no-deploy=traefik,servicelb,metrics-server@server:*"
--network "nw01"
- uses: AbsaOSS/k3d-action@v2
name: "Create 2nd Cluster in 172.18.0.0/16"
with:
cluster-name: "test-cluster-2"
args: >-
-p "81:80@agent:0:direct"
-p "444:443@agent:0:direct"
-p "5054:53/udp@agent:0:direct"
--agents 3
--no-lb
--k3s-arg "--no-deploy=traefik,servicelb,metrics-server@server:*"
--network "nw01"
- uses: AbsaOSS/k3d-action@v2
name: "Create 1st Cluster in 172.19.0.0/16"
with:
cluster-name: "test-cluster-3"
args: >-
-p "82:80@agent:0:direct"
-p "445:443@agent:0:direct"
-p "5055:53/udp@agent:0:direct"
--agents 3
--no-lb
--k3s-arg "--no-deploy=traefik,servicelb,metrics-server@server:*"
--network "nw02"
- uses: AbsaOSS/k3d-action@v2
name: "Create 2nd Cluster in 172.19.0.0/16"
with:
cluster-name: "test-cluster-4"
args: >-
-p "83:80@agent:0:direct"
-p "446:443@agent:0:direct"
-p "5056:53/udp@agent:0:direct"
--agents 3
--no-lb
--k3s-arg "--no-deploy=traefik,servicelb,metrics-server@server:*"
--network "nw02"AbsaOSS/k3d-action creates four identical clusters in two different bridge networks.
For more details see: Demo, Source
Before test starts, you need to build your app and install into the cluster. This requires interaction with the image registry. Usually you don't want to push a new image into the remote registry for each test. Instead, you can import the image directly into the created cluster:
docker build . -t <repository>:<semver>
k3d image import <repository>:<semver> -c <cluster-name>Example below demonstrates how to interact with imported docker registry:
steps:
- uses: actions/checkout@v2
- uses: AbsaOSS/k3d-action@v2
name: "Create single k3d Cluster with imported Registry"
with:
cluster-name: test-cluster-1
args: >-
--agents 3
--no-lb
--k3s-arg "--no-deploy=traefik,servicelb,metrics-server@server:*"
- name: "Docker repo demo"
run: |
docker build . -t myproj/demo:v1.0.0
k3d image import myproj/demo:v1.0.0 -c test-cluster-1 --verbose
kubectl apply -f pod.yaml
# pod.yaml
#
# apiVersion: v1
# kind: Pod
# metadata:
# name: test-pod
# spec:
# containers:
# - name: demo-app
# image: myproj/demo:v1.0.0For further details see: