Build flexible abstraction for any Kubernetes Resources with CUE and KubeVela

May 30, 2022 · 10 min read

Jianbo Sun

KubeVela Team

This blog will introduce how to use CUE and KubeVela to build you own abstraction API to reduce the complexity of Kubernetes resources. As a platform builder, you can dynamically customzie the abstraction, build a path from shallow to deep for your developers per needs, adapt to growing number of different scenarios, and meet the iterative demands of the company's long-term business development.

Convert Kubernetes API Objects Into Custom Components#

Let's start the journey by using the Kubernetes StatefulSet as example, we will convert it to be a customized module and provide capabilities.

Save the YAML example of StatefulSet in the official document locally and name it as my-stateful.yaml, then execute command as below:

 vela def init my-stateful -t component --desc "My StatefulSet component." --template-yaml ./my-stateful.yaml -o my-stateful.cue

View the generated "my-stateful.cue" file:

$ cat my-stateful.cue
"my-stateful": {
    annotations: {}
    attributes: workload: definition: {
        apiVersion: "<change me> apps/v1"
        kind:       "<change me> Deployment"
    }
    description: "My StatefulSet component."
    labels: {}
    type: "component"
}

template: {
    output: {
        apiVersion: "v1"
        kind:       "Service"
            ... // omit non-critical info
    }
    outputs: web: {
        apiVersion: "apps/v1"
        kind:       "StatefulSet"
            ... // omit non-critical info
    }
    parameter: {}
}

Modify the generated file as follows:

The example of the official StatefulSet website is a composite component composed of two objects StatefulSet and Service. According to KubeVela Rules for customize component, in composite components, one of the resources such (StatefulSet in our example) need to be represented by the template.output field as a core workload, and other auxiliary objects are represented by template.outputs, so we make some adjustments and all the automatically generated output and outputs are switched.
Then we fill in the apiVersion and kind data of the core workload into the part marked as <change me>

After modification, you can use vela def vet to do format check and verification.

$ vela def vet my-stateful.cue
Validation succeed.

The file after two steps of changes is as follows:

$ cat my-stateful.cue
"my-stateful": {
    annotations: {}
    attributes: workload: definition: {
        apiVersion: "apps/v1"
        kind:       "StatefulSet"
    }
    description: "My StatefulSet component."
    labels: {}
    type: "component"
}

template: {
    output: {
        apiVersion: "apps/v1"
        kind:       "StatefulSet"
        metadata: name: "web"
        spec: {
            selector: matchLabels: app: "nginx"
            replicas:    3
            serviceName: "nginx"
            template: {
                metadata: labels: app: "nginx"
                spec: {
                    containers: [{
                        name: "nginx"
                        ports: [{
                            name:          "web"
                            containerPort: 80
                        }]
                        image: "k8s.gcr.io/nginx-slim:0.8"
                        volumeMounts: [{
                            name:      "www"
                            mountPath: "/usr/share/nginx/html"
                        }]
                    }]
                    terminationGracePeriodSeconds: 10
                }
            }
            volumeClaimTemplates: [{
                metadata: name: "www"
                spec: {
                    accessModes: ["ReadWriteOnce"]
                    resources: requests: storage: "1Gi"
                    storageClassName: "my-storage-class"
                }
            }]
        }
    }
    outputs: web: {
        apiVersion: "v1"
        kind:       "Service"
        metadata: {
            name: "nginx"
            labels: app: "nginx"
        }
        spec: {
            clusterIP: "None"
            ports: [{
                name: "web"
                port: 80
            }]
            selector: app: "nginx"
        }
    }
    parameter: {}
}

Install ComponentDefinition into the Kubernetes cluster:

$ vela def apply my-stateful.cue
ComponentDefinition my-stateful created in namespace vela-system.

You can see that a my-stateful component via vela components command:

$ vela components
NAME        NAMESPACE   WORKLOAD                                DESCRIPTION
...
my-stateful vela-system statefulsets.apps                       My StatefulSet component.
... 

When you put this customized component into Application, it looks like:

cat <<EOF | vela up -f -
apiVersion: core.oam.dev/v1beta1
kind: Application
metadata:
  name: website
spec:
  components:
    - name: my-component
      type: my-stateful
EOF

Define Customized Parameters For Component#

In previous section we have defined a ComponentDefinition that has no parameter. In this section we will show how to expose parameters.

In this example, we expose the following parameters to the user:

Image name, allowing users to customize the image
Instance name, allowing users to customize the instance name of the generated StatefulSet object and Service object

    ... # Omit other unmodified fields
    template: {
        output: {
            apiVersion: "apps/v1"
            kind:       "StatefulSet"
            metadata: name: parameter.name
            spec: {
                selector: matchLabels: app: "nginx"
                replicas:    3
                serviceName: "nginx"
                template: {
                    metadata: labels: app: "nginx"
                    spec: {
                        containers: [{
                            image: parameter.image
        
                            ... // Omit other unmodified fields
                        }]
                    }
                }
                    ... // Omit other unmodified fields
            }
        }
        outputs: web: {
            apiVersion: "v1"
            kind:       "Service"
            metadata: {
                name: "nginx"
                labels: app: "nginx"
            }
            spec: {
                ... // Omit other unmodified fields 
            }
        }
        parameter: {
            image: string
            name: string
        }
    }

After modification, use vela def apply to install to the cluster:

$ vela def apply my-stateful.cue
ComponentDefinition my-stateful in namespace vela-system updated.

Then as a platform builder, you have finished your setup. Let's see what's the developer experience now.

Developer Experience#

The only thing your developer need to learn is the Open Application Model which always follow a unified format.

Discover the Component#

The developers can discover and check the parameters of my-stateful component as follows:

$ vela def list
NAME                            TYPE                    NAMESPACE   DESCRIPTION
my-stateful                     ComponentDefinition     vela-system My StatefulSet component.
...snip...

$ vela show my-stateful
# Properties
+----------+-------------+--------+----------+---------+
|   NAME   | DESCRIPTION |  TYPE  | REQUIRED | DEFAULT |
+----------+-------------+--------+----------+---------+
| name     |             | string | true     |         |
| image    |             | string | true     |         |
+----------+-------------+--------+----------+---------+

Updating the ComponentDefinition will not affect existing Applications. It will take effect only after updating the Applications next time.

Use the Component in Application#

The developers can easily specify the three new parameters in the application:

    apiVersion: core.oam.dev/v1beta1
    kind: Application
    metadata:
      name: website
    spec:
      components:
        - name: my-component
          type: my-stateful
          properties:
            image: nginx:latest
            name: my-component

The only thing left is to deploy the yaml file ( assume the name app-stateful.yaml) by executing vela up -f app-stateful.yaml.

Then you can see that the name, image, and number of instances of the StatefulSet object have been updated.

Dry-run for diagnose or integration#

In order to ensure that the developer's application can run correctly with the parameters, you can also use the vela dry-run command to verify the trial run of your template.

vela dry-run -f app-stateful.yaml

By viewing the output, you can compare whether the generated object is consistent with the object you actually expect. You can even execute this YAML directly into the Kubernetes cluster and use the results of the operation for verification.

# Application(website) -- Component(my-component)
---

apiVersion: v1
kind: Service
metadata:
  labels:
    app: nginx
    app.oam.dev/appRevision: ""
    app.oam.dev/component: my-component
    app.oam.dev/name: website
    workload.oam.dev/type: my-stateful
  name: nginx
  namespace: default
spec:
  clusterIP: None
  ports:
  - name: web
    port: 80
  selector:
    app: nginx
  template:
    spec:
      containers:
      - image: saravak/fluentd:elastic
        name: my-sidecar

---
apiVersion: apps/v1
kind: StatefulSet
metadata:
  labels:
    app.oam.dev/appRevision: ""
    app.oam.dev/component: my-component
    app.oam.dev/name: website
    trait.oam.dev/resource: web
    trait.oam.dev/type: AuxiliaryWorkload
  name: web
  namespace: default
spec:
  replicas: 3
  selector:
    matchLabels:
      app: nginx
  serviceName: nginx
  template:
    metadata:
      labels:
        app: nginx
    spec:
      containers:
      - image: k8s.gcr.io/nginx-slim:0.8
        name: nginx
        ports:
        - containerPort: 80
          name: web
        volumeMounts:
        - mountPath: /usr/share/nginx/html
          name: www
      terminationGracePeriodSeconds: 10
  volumeClaimTemplates:
  - metadata:
      name: www
    spec:
      accessModes:
      - ReadWriteOnce
      resources:
        requests:
          storage: 1Gi
      storageClassName: my-storage-class

You can also use vela dry-run -h to view more available function parameters.

Use `context` to avoid duplication#

KubeVela allows you to reference the runtime information of your application via context keyword.

In our example above, the field name in the properties and the field name of the Component have the same meaning, so we can use the context to avoid duplication. We cann use the context.name to reference the component name in the runtime, thus the name parameter in parameter is no longer needed.

Just modify the definition file (my-stateful.cue) as the following

... # Omit other unmodified fields
template: {
    output: {
        apiVersion: "apps/v1"
        kind:       "StatefulSet"
-       metadata: name: parameter.name
+       metadata: name: context.name

            ... // Omit other unmodified field

    }
    parameter: {
-       name: string
        image: string
    }
}

Then deploy the changes by the following:

vela def apply my-stateful.cue

After that, the developers can immediately run application as below:

    apiVersion: core.oam.dev/v1beta1
    kind: Application
    metadata:
      name: website
    spec:
      components:
        - name: my-component
          type: my-stateful
          properties:
            image: "nginx:latest"

There's no upgrade or restart for any system, they're all running into affect dynamically per your needs.

Add Operational Traits On Demand#

OAM follows the principle of "separation of concerns", after the developers finished the component part, the operators can add traits into the application to control the rest part configuration of the deployment. For example, the operators can
control replicas, adding labels/annotations, injecting environment variables/sidecars, adding persistent volumes, and so on.

Technically, the trait system works in two ways:

Patch/Override the configurations defined in component.
Generate more configuration.

The customized process works the same with the component, they both use CUE but has some different keywords for path and override, you can refer to customize trait for details.

Operator Experience with traits#

There're already some built-in traits after KubeVela installed. The operator can use vela traits to view, the traits marked with * are general traits, which can operate on common Kubernetes resource objects.

$ vela traits
NAME                        NAMESPACE   APPLIES-TO          CONFLICTS-WITH  POD-DISRUPTIVE  DESCRIPTION
annotations                 vela-system *                                   true            Add annotations on K8s pod for your workload which follows
                                                                                            the pod spec in path 'spec.template'.
configmap                   vela-system *                                   true            Create/Attach configmaps on K8s pod for your workload which
                                                                                            follows the pod spec in path 'spec.template'.
labels                      vela-system *                                   true            Add labels on K8s pod for your workload which follows the
                                                                                            pod spec in path 'spec.template'.
scaler                      vela-system *                                   false           Manually scale K8s pod for your workload which follows the
                                                                                            pod spec in path 'spec.template'.
sidecar                     vela-system *                                   true            Inject a sidecar container to K8s pod for your workload
                                                                                            which follows the pod spec in path 'spec.template'.
...snip...

Taking sidecar as an example, you can check the usage of sidecar:

$ vela show sidecar
# Properties
+---------+-----------------------------------------+-----------------------+----------+---------+
|  NAME   |               DESCRIPTION               |         TYPE          | REQUIRED | DEFAULT |
+---------+-----------------------------------------+-----------------------+----------+---------+
| name    | Specify the name of sidecar container   | string                | true     |         |
| cmd     | Specify the commands run in the sidecar | []string              | false    |         |
| image   | Specify the image of sidecar container  | string                | true     |         |
| volumes | Specify the shared volume path          | [[]volumes](#volumes) | false    |         |
+---------+-----------------------------------------+-----------------------+----------+---------+


## volumes
+------+-------------+--------+----------+---------+
| NAME | DESCRIPTION |  TYPE  | REQUIRED | DEFAULT |
+------+-------------+--------+----------+---------+
| path |             | string | true     |         |
| name |             | string | true     |         |
+------+-------------+--------+----------+---------+

Use the sidecar directly to inject a container, the application description is as follows:

apiVersion: core.oam.dev/v1beta1
kind: Application
metadata:
  name: website
spec:
  components:
    - name: my-component
      type: my-stateful
      properties:
        image: nginx:latest
        name: my-component
      traits:
      - type: sidecar
        properties:
          name: my-sidecar
          image: saravak/fluentd:elastic

Deploy and run the application, and you can see that a fluentd sidecar has been deployed and running in the StatefulSet.

Both components and traits are re-usable on any KubeVela systems, we can package components, traits along with the CRD controllers together as an addon. There're a growing catalog of addons in the community.

Summarize#

This blog introduces how to deliver complete modular capabilities through CUE. The core is that it can dynamically increase configuration capabilities according to user needs, and gradually expose more functions and usages, so as to reduce the overall learning threshold for users and ultimately improve R&D efficient. The out-of-the-box capabilities provided by KubeVela, including components, traits, policy, and workflow, are also designed as plugable and modifiable capabilities.

All posts