Skip to main content
Version: v1.7

Initialize Env Infra Resources

This section will introduce how to initialize and destroy infrastructure of environment with KubeVela easily.

What can be infrastructure of environment

An Application development team usually needs to initialize some shared environment for users. An environment is a logical concept that represents a set of common infrastructure resources for Applications.

For example, a team usually wants two environments: one for development, and one for production.

In general, the infra resource types that can be initialized include the following types:

  1. One or more Kubernetes clusters. Different environments may need different sizes and versions of Kubernetes clusters. Environment initialization can also manage multiple clusters .

  2. Any type of Kubernetes custom resources (CRDs) and system plug-ins can be set up in environment initialization.

  3. All kinds of shared resources and services. For example. shared resources in microservices. These shared resources can be a microservice component, cloud database, cache, load balancer, API gateway, and so on.

  4. Various management policies and processes. An environment may have different global policies. The policy can be chaos test, security scan, SLO and son on; the process can be initializing a database table, registering an automatic discovery configuration, and so on.

KubeVela allows you to use different resources to initialize the environment.

You can use the Policy and Workflow in your Application. Note that there may be dependencies between initializations, we can use depends-on-app in workflow to do it.

The initialization of different environments has dependencies. Common resources can be separated as dependencies. In this way, reusable initialization modules can be formed.

For example, if both the test and develop environments rely on the same controllers, these controllers can be pulled out and initialized as separate environments, specifying dependency initialization in both the development and test environments.

How to use

Directly use Application for initialization

If we want to use some CRD controller like OpenKruise in cluster, we can use Helm to initialize kruise.

We can directly use Application to initialize a kruise environment. The application below will deploy a kruise controller in cluster.

We have to enable FluxCD addon in cluster since we use Helm to deploy kruise. We can use depends-on-app to make sure addon-fluxcd is deployed before kruise, we also use apply-once policy here to make sure we'll only apply the application once for initialization.

depends-on-app will check if the cluster has the application with name and namespace defines in properties. If the application exists, the next step will be executed after the application is running. If the application do not exists, KubeVela will check the ConfigMap with the same name, and read the config of the Application and apply to cluster. For more information, please refer to depends-on-app.

cat <<EOF | vela up -f -
apiVersion: core.oam.dev/v1beta1
kind: Application
metadata:
name: kruise
namespace: vela-system
spec:
components:
- name: kruise
type: helm
properties:
repoType: helm
url: https://openkruise.github.io/charts/
chart: kruise
version: 1.2.0
git:
branch: master
values:
featureGates: PreDownloadImageForInPlaceUpdate=true
policies:
- name: apply-once
type: apply-once
properties:
enable: true
workflow:
steps:
- name: check-flux
type: depends-on-app
properties:
name: addon-fluxcd
namespace: vela-system
- name: apply-kruise
type: apply-component
properties:
component: kruise
EOF

Check the application in cluster:

vela status kruise -n vela-system
Expected Outcome
About:

Name: kruise
Namespace: vela-system
Created at: 2022-10-31 18:10:27 +0800 CST
Status: running

Workflow:

mode: StepByStep-DAG
finished: true
Suspend: false
Terminated: false
Steps
- id: l6apfsi5c2
name: check-flux
type: depends-on-app
phase: succeeded
- id: p2nqell47w
name: apply-kruise
type: apply-component
phase: succeeded

Services:

- Name: kruise
Cluster: local Namespace: vela-system
Type: helm
Healthy Fetch repository successfully, Create helm release successfully
No trait applied

Kruise is running successfully! Then you can use kruise in your cluster. If you need to set up a new environment, the only thing you need to do is to apply the files like above.

Add initialize workflow in application

Some Kubernetes native resources like ConfigMap/PVC are commonly used in the environment.

If you want to apply those resources before deploying your application, you can add an initialization workflow to your application.

KubeVela provides a built-in workflow step apply-object to fill in native Kubernetes resources. In this way, by filling in Kubernetes native resources, we can avoid writing redundant component definitions.

Apply the following application, it will initialize an environment with ConfigMap/PVC. There is two components in this application, the first one will write data to PVC, the second on will read the data from PVC:

apiVersion: core.oam.dev/v1beta1
kind: Application
metadata:
name: server-with-pvc-and-cm
namespace: default
spec:
components:
- name: log-gen-worker
type: worker
properties:
image: busybox
cmd:
- /bin/sh
- -c
- >
i=0;
while true;
do
echo "$i: $(date)" >> /test-pvc/date.log;
i=$((i+1));
sleep 1;
done
volumes:
- name: "my-pvc"
type: "pvc"
mountPath: "/test-pvc"
claimName: "my-claim"
- name: "my-configmap"
type: "configMap"
mountPath: "/test-cm"
cmName: "my-cm"
items:
- key: test-key
path: test-key
- name: log-read-worker
type: worker
properties:
name: count-log
image: busybox
cmd:
- /bin/sh
- -c
- 'tail -n+1 -f /test-pvc/date.log'
volumes:
- name: "my-pvc"
type: "pvc"
mountPath: "/test-pvc"
claimName: "my-claim"
- name: "my-configmap"
type: "configMap"
mountPath: "/test-cm"
cmName: "my-cm"
items:
- key: test-key
path: test-key

policies:
- name: my-policy
properties:
clusters:
- local
type: topology
- name: apply-once
type: apply-once
properties:
enable: true

workflow:
steps:
- name: apply-pvc
type: apply-object
properties:
value:
apiVersion: v1
kind: PersistentVolumeClaim
metadata:
name: my-claim
namespace: default
spec:
accessModes:
- ReadWriteOnce
resources:
requests:
storage: 8Gi
storageClassName: standard
- name: apply-cm
type: apply-object
properties:
value:
apiVersion: v1
kind: ConfigMap
metadata:
name: my-cm
namespace: default
data:
test-key: test-value
- name: deploy-comp
properties:
policies:
- my-policy
type: deploy

Check the PVC and ConfigMap in cluster:

vela status server-with-pvc-and-cm --detail --tree
Expected Outcome
CLUSTER       NAMESPACE     RESOURCE                       STATUS    APPLY_TIME          DETAIL
local ─── default ─┬─ ConfigMap/my-cm updated 2022-10-31 18:00:52 Data: 1 Age: 57s
├─ PersistentVolumeClaim/my-claim updated 2022-10-31 10:00:52 Status: Bound Volume: pvc-b6f88ada-af98-468d-8cdd-31ca110c5e1a Capacity: 8Gi Access Modes: RWO StorageClass: standard Age: 57s
├─ Deployment/log-gen-worker updated 2022-10-31 10:00:52 Ready: 1/1 Up-to-date: 1 Available: 1 Age: 57s
└─ Deployment/log-read-worker updated 2022-10-31 10:00:52 Ready: 1/1 Up-to-date: 1 Available: 1 Age: 57s

Check the application in cluster:

vela status server-with-pvc-and-cm
Expected Outcome
$ vela status server-with-pvc-and-cm
About:

Name: server-with-pvc-and-cm
Namespace: default
Created at: 2022-10-31 18:00:51 +0800 CST
Status: running

Workflow:

mode: StepByStep-DAG
finished: true
Suspend: false
Terminated: false
Steps
- id: xboizfjo28
name: apply-pvc
type: apply-object
phase: succeeded
- id: 4ngx25mrx8
name: apply-cm
type: apply-object
phase: succeeded
- id: 1gzzt3mfw1
name: deploy-comp
type: deploy
phase: succeeded

Services:

- Name: log-gen-worker
Cluster: local Namespace: default
Type: worker
Healthy Ready:1/1
No trait applied

- Name: log-read-worker
Cluster: local Namespace: default
Type: worker
Healthy Ready:1/1
No trait applied

Check the logs of the second component:

vela logs server-with-pvc-and-cm --component log-read-worker
Expected Outcome
+ log-read-worker-7f4bc9d9b5-kb5l6 › log-read-worker
log-read-worker 2022-10-31T10:01:15.606903716Z 0: Mon Oct 31 10:01:13 UTC 2022
log-read-worker 2022-10-31T10:01:15.606939383Z 1: Mon Oct 31 10:01:14 UTC 2022
log-read-worker 2022-10-31T10:01:15.606941883Z 2: Mon Oct 31 10:01:15 UTC 2022
log-read-worker 2022-10-31T10:01:16.607006425Z 3: Mon Oct 31 10:01:16 UTC 2022
log-read-worker 2022-10-31T10:01:17.607184925Z 4: Mon Oct 31 10:01:17 UTC 2022
log-read-worker 2022-10-31T10:01:18.607304426Z 5: Mon Oct 31 10:01:18 UTC 2022
...

We can see that both components is running. The two components share the same PVC and use the same ConfigMap.

Destroy the infrastructure of environment

As we have already modeled the environment as a KubeVela Application, we can destroy the environment easily by deleting the application.

vela delete server-with-pvc-and-cm

Then the KubeVela controller will clean up all these resources.