Creating a StatefulSet

Scenario

StatefulSets are a type of workloads whose data or status is stored while they are running. For example, MySQL is a StatefulSet because it needs to store new data.

A container can be migrated between different hosts, but data is not stored on the hosts. To store StatefulSet data persistently, attach HA storage volumes provided by CCE to the container.

Constraints

When you delete or scale a StatefulSet, the system does not delete the storage volumes associated with the StatefulSet to ensure data security.
When you delete a StatefulSet, reduce the number of replicas to 0 before deleting the StatefulSet so that pods in the StatefulSet can be stopped in order.
When you create a StatefulSet, a headless Service is required for pod access. For details, see Headless Service.
When a node is unavailable, pods become Unready. In this case, manually delete the pods of the StatefulSet so that the pods can be migrated to a normal node.

Prerequisites

Before creating a workload, you must have an available cluster. For details on how to create a cluster, see Buying a CCE Cluster.
To enable public access to a workload, ensure that an EIP or load balancer has been bound to at least one node in the cluster.

If a pod has multiple containers, ensure that the ports used by the containers do not conflict with each other. Otherwise, creating the StatefulSet will fail.

Using the CCE Console

Log in to the CCE console.
Click the cluster name to go to the cluster console, choose Workloads in the navigation pane, and click the Create Workload in the upper right corner.

Set basic information about the workload.

Basic Info

Workload Type: Select StatefulSet. For details about workload types, see Overview.
Workload Name: Enter the name of the workload. Enter 1 to 63 characters starting with a lowercase letter and ending with a lowercase letter or digit. Only lowercase letters, digits, and hyphens (-) are allowed.
Namespace: Select the namespace of the workload. The default value is default. You can also click Create Namespace to create one. For details, see Creating a Namespace.
Pods: Enter the number of pods of the workload.
Container Runtime: A CCE standard cluster uses runC by default, whereas a CCE Turbo cluster supports both runC and Kata. For details about the differences, see Kata Runtime and Common Runtime.
Time Zone Synchronization: Specify whether to enable time zone synchronization. After time zone synchronization is enabled, the container and node use the same time zone. The time zone synchronization function depends on the local disk mounted to the container. Do not modify or delete the time zone. For details, see Configuring Time Zone Synchronization.

Container Settings

Container Information

Multiple containers can be configured in a pod. You can click Add Container on the right to configure multiple containers for the pod.

Basic Info: Configure basic information about the container.

Parameter	Description
Container Name	Name the container.
Pull Policy	Image update or pull policy. If you select Always, the image is pulled from the image repository each time. If you do not select Always, the existing image of the node is preferentially used. If the image does not exist, the image is pulled from the image repository.
Image Name	Click Select Image and select the image used by the container. To use a third-party image, see Using Third-Party Images.
Image Tag	Select the image tag to be deployed.
CPU Quota	Request: minimum number of CPU cores required by a container. The default value is 0.25 cores. Limit: maximum number of CPU cores available for a container. Do not leave Limit unspecified. Otherwise, intensive use of container resources will occur and your workload may exhibit unexpected behavior. If Request and Limit are not specified, the quota is not limited. For more information and suggestions about Request and Limit, see Configuring Container Specifications.
Memory Quota	Request: minimum amount of memory required by a container. The default value is 512 MiB. Limit: maximum amount of memory available for a container. When memory usage exceeds the specified memory limit, the container will be terminated. If Request and Limit are not specified, the quota is not limited. For more information and suggestions about Request and Limit, see Configuring Container Specifications.
(Optional) GPU Quota	Configurable only when the cluster contains GPU nodes and the CCE AI Suite (NVIDIA GPU) add-on is installed. All: No GPU will be used. Dedicated: GPU resources are dedicated for the container. Shared: percentage of GPU resources used by the container. For example, if this parameter is set to 10%, the container uses 10% of GPU resources. For details about how to use GPUs in the cluster, see Default GPU Scheduling in Kubernetes.
(Optional) NPU Quota	Number of required Ascend chips. The value must be an integer and the CCE AI Suite (Ascend NPU) add-on must be installed. For details about how to use NPUs in the cluster, see NPU Scheduling.
(Optional) Privileged Container	Programs in a privileged container have certain privileges. If Privileged Container is enabled, the container is assigned privileges. For example, privileged containers can manipulate network devices on the host machine and modify kernel parameters.
(Optional) Init Container	Whether to use the container as an init container. An init container does not support health check. An init container is a special container that runs before other app containers in a pod are started. Each pod can contain multiple containers. In addition, a pod can contain one or more init containers. Application containers in a pod are started and run only after the running of all init containers completes. For details, see Init Containers.

(Optional) Lifecycle: Configure operations to be performed in a specific phase of the container lifecycle, such as Startup Command, Post-Start, and Pre-Stop. For details, see Configuring Container Lifecycle Parameters.
(Optional) Health Check: Set the liveness probe, ready probe, and startup probe as required. For details, see Configuring Container Health Check.
(Optional) Environment Variables: Configure variables for the container running environment using key-value pairs. These variables transfer external information to containers running in pods and can be flexibly modified after application deployment. For details, see Configuring Environment Variables.
(Optional) Data Storage: Mount local storage or cloud storage to the container. The application scenarios and mounting modes vary with the storage type. For details, see Storage.
- StatefulSets support dynamic attachment of EVS disks. For details, see Dynamically Mounting an EVS Disk to a StatefulSet and Dynamically Mounting a Local PV to a StatefulSet.
  Dynamic mounting is achieved by using the volumeClaimTemplates field and depends on the dynamic creation capability of StorageClass. A StatefulSet associates each pod with a PVC using the volumeClaimTemplates field, and the PVC is bound to the corresponding PV. Therefore, after the pod is rescheduled, the original data can still be mounted based on the PVC name.
- After a workload is created, the storage that is dynamically mounted cannot be updated.
(Optional) Security Context: Assign container permissions to protect the system and other containers from being affected. Enter the user ID to assign container permissions and prevent systems and other containers from being affected.
(Optional) Logging: Report standard container output logs to AOM by default, without requiring manual settings. You can manually configure the log collection path. For details, see Connecting CCE to AOM.
To disable the standard output of the current workload, add the annotation kubernetes.AOM.log.stdout: [] in Labels and Annotations. For details about how to use this annotation, see Table 1.

Image Access Credential: Select the credential used for accessing the image repository. The default value is default-secret. You can use default-secret to access images in SWR. For details about default-secret, see default-secret.
(Optional) GPU: All is selected by default. The workload instance will be scheduled to the node of the specified GPU type.

Headless Service Parameters

A headless Service is used to solve the problem of mutual access between pods in a StatefulSet. The headless Service provides a fixed access domain name for each pod. For details, see Headless Service.

(Optional) Service Settings

A Service provides external access for pods. With a static IP address, a Service forwards access traffic to pods and automatically balances load for these pods.

You can also create a Service after creating a workload. For details about Services of different types, see Overview.

(Optional) Advanced Settings

Upgrade: Specify the upgrade mode and parameters of the workload. Rolling upgrade and Replace upgrade are available. For details, see Workload Upgrade Policies.
Pod Management Policies
For some distributed systems, the StatefulSet sequence is unnecessary and/or should not occur. These systems require only uniqueness and identifiers.
- OrderedReady: The StatefulSet will deploy, delete, or scale pods in order and one by one. (The StatefulSet continues only after the previous pod is ready or deleted.) This is the default policy.
- Parallel: The StatefulSet will create pods in parallel to match the desired scale without waiting, and will delete all pods at once.
Scheduling: Configure affinity and anti-affinity policies for flexible workload scheduling. Load affinity and node affinity are provided.
- Load Affinity: Common load affinity policies are offered for quick load affinity deployment.
  - Multi-AZ deployment preferred: Workload pods are preferentially scheduled to nodes in different AZs through pod anti-affinity (podAntiAffinity). If all the nodes in the cluster are deployed in the same AZ, the pods will be scheduled to that AZ but onto different nodes for high availability. If there are fewer nodes than pods, the extra pods will fail to run.
  - Forcible multi-AZ deployment: Workload pods are forcibly scheduled to nodes in different AZs through pod anti-affinity (podAntiAffinity). If there are fewer AZs than pods, the extra pods will fail to run.
  - Custom policies: Affinity and anti-affinity policies can be customized as needed. For details, see Scheduling Policies (Affinity/Anti-affinity).
- Node Affinity: Common load affinity policies are offered for quick load affinity deployment.
  - Specified node scheduling: Workload pods can be deployed on specified nodes through node affinity (nodeAffinity). If no node is specified, the pods will be randomly scheduled based on the default scheduling policy of the cluster.
  - Specified node pool scheduling: Workload pods can be deployed in a specified node pool through node affinity (nodeAffinity). If no node pool is specified, the pods will be randomly scheduled based on the default scheduling policy of the cluster.
  - Custom policies: Affinity and anti-affinity policies can be customized as needed. For details, see Scheduling Policies (Affinity/Anti-affinity).
Toleration: Using both taints and tolerations allows (not forcibly) the pod to be scheduled to a node with the matching taints, and controls the pod eviction policies after the node where the pod is located is tainted. For details, see Taints and Tolerations.
Labels and Annotations: Add labels or annotations for pods using key-value pairs. After entering the key and value, click Confirm. For details about how to use and configure labels and annotations, see Labels and Annotations.
DNS: Configure a separate DNS policy for the workload. For details, see DNS Configuration.
APM Settings: Use Application Performance Management (APM) to provide more accurate problem analysis and location for Java programs. For details, see Configuring APM Settings for Performance Bottleneck Analysis.
Network Settings
- Pod ingress/egress bandwidth limitation: You can set ingress/egress bandwidth limitation for pods. For details, see Configuring QoS for a Pod.

Click Create Workload in the lower right corner.

Using kubectl

In this example, a Nginx workload is used and the EVS volume is dynamically mounted to it using the volumeClaimTemplates field.

Use kubectl to connect to the cluster. For details, see Connecting to a Cluster Using kubectl.

Create and edit the nginx-statefulset.yaml file.

nginx-statefulset.yaml is an example file name, and you can change it as required.

vi nginx-statefulset.yaml

The following provides an example of the file contents. For more information on StatefulSet, see the Kubernetes documentation.

apiVersion: apps/v1
kind: StatefulSet
metadata:
  name: nginx
spec:
  selector:
    matchLabels:
      app: nginx
  template:
    metadata:
      labels:
        app: nginx
    spec:
      containers:
        - name: container-1
          image: nginx:latest
          imagePullPolicy: IfNotPresent
          resources:
            requests:
              cpu: 250m
              memory: 512Mi
            limits:
              cpu: 250m
              memory: 512Mi
          volumeMounts:
            - name: test
              readOnly: false
              mountPath: /usr/share/nginx/html
              subPath: ''
      imagePullSecrets:
        - name: default-secret
      dnsPolicy: ClusterFirst
      volumes: []
  serviceName: nginx-svc
  replicas: 2
  volumeClaimTemplates:  # Dynamically mounts the EVS volume to the workload.
    - apiVersion: v1
      kind: PersistentVolumeClaim
      metadata:
        name: test
        namespace: default
        annotations:
          everest.io/disk-volume-type: SAS  # SAS EVS volume type.
        labels:
          failure-domain.beta.kubernetes.io/region: ae-ad-1  # region where the EVS volume is created.
          failure-domain.beta.kubernetes.io/zone:    # AZ where the EVS volume is created. It must be the same as the AZ of the node.
      spec:
        accessModes:
          - ReadWriteOnce  # The value must be ReadWriteOnce for the EVS volume.
        resources:
          requests:
            storage: 10Gi
        storageClassName: csi-disk # Storage class name. The value is csi-disk for the EVS volume.
  updateStrategy:
    type: RollingUpdate

vi nginx-headless.yaml

apiVersion: v1
kind: Service
metadata:
  name: nginx-svc
  namespace: default
  labels:
    app: nginx
spec:
  selector:
    app: nginx
    version: v1
  clusterIP: None
  ports:
    - name: nginx
      targetPort: 80
      nodePort: 0
      port: 80
      protocol: TCP
  type: ClusterIP

Create a workload and the corresponding headless service.

kubectl create -f nginx-statefulset.yaml

If the following information is displayed, the StatefulSet has been successfully created.
```
statefulset.apps/nginx created
```
kubectl create -f nginx-headless.yaml

If the following information is displayed, the headless service has been successfully created.
```
service/nginx-svc created
```
If the workload will be accessed through a ClusterIP or NodePort Service, configure the access mode. For details, see Network.