Horizontal Scaling Based on CLB Monitoring Metrics

Business Scenario

Business traffic on TKE is often accessed through CLB (Tencent Cloud Load Balancer). Sometimes, it's desirable for workloads to scale directly based on CLB monitoring metrics, for example:

1. For long connection scenarios like game rooms and online meetings, one connection corresponds to one user. Each Pod in the workload handles a relatively fixed upper limit of connections, so scaling can be based on CLB connection count metrics.
2. For online services using the HTTP protocol, where a single Pod in the workload can support a relatively fixed QPS, scaling can be based on CLB's QPS (requests per second) metric.

Introduction to keda-tencentcloud-clb-scaler

KEDA has many built-in triggers, but not for Tencent Cloud CLB. However, KEDA supports external-type triggers to extend triggers. keda-tencentcloud-clb-scaler is a KEDA External Scaler based on Tencent Cloud CLB monitoring metrics, enabling elastic scaling based on CLB metrics such as connections, QPS, and bandwidth.

Prepare Access Keys

You need to prepare access keys (SecretID, SecretKey) for a Tencent Cloud account. Refer to Sub-account Access Key Management. The account must have at least the following permissions:

{
    "version": "2.0",
    "statement": [
        {
            "effect": "allow",
            "action": [
                "clb.DescribeLoadBalancers",
                "monitor.DescribeProductList",
                "monitor.GetMonitorData",
                "monitor.DescribeBaseMetrics"
            ],
            "resource": [
                "*"
            ]
        }
    ]
}

Install keda-tencentcloud-clb-scaler

helm repo add clb-scaler https://imroc.github.io/keda-tencentcloud-clb-scaler
helm upgrade --install clb-scaler clb-scaler/clb-scaler -n keda \
  --set region="ap-chengdu" \
  --set credentials.secretId="xxx" \
  --set credentials.secretKey="xxx"

• Modify region to the region where the CLB is located (usually the cluster's region). Region list: https://cloud.tencent.com/document/product/213/6091
• credentials.secretId and credentials.secretKey are the Tencent Cloud account access keys, used to call related cloud APIs to query CLB monitoring data.

Deploy Workload

Below is a workload YAML example for testing:

apiVersion: v1
kind: Service
metadata:
  labels:
    app: httpbin
  name: httpbin
spec:
  ports:
    - port: 8080
      protocol: TCP
      targetPort: 80
  selector:
    app: httpbin
  type: LoadBalancer

---
apiVersion: apps/v1
kind: Deployment
metadata:
  name: httpbin
spec:
  replicas: 1
  selector:
    matchLabels:
      app: httpbin
  template:
    metadata:
      labels:
        app: httpbin
    spec:
      containers:
        - image: kennethreitz/httpbin:latest
          name: httpbin

After deployment, a corresponding public network CLB will be automatically created to receive traffic. Get the corresponding CLB ID with the following command:

$ kubectl get svc httpbin -o jsonpath='{.metadata.annotations.service\.kubernetes\.io/loadbalance-id}'
lb-********

Record the obtained CLB ID, as it will be needed for the subsequent KEDA scaling configuration.

Configure Elastic Scaling Based on CLB Monitoring Metrics Using ScaledObject

Configuration Method

Scaling based on CLB monitoring metrics is typically used for online services and usually configured with KEDA's ScaledObject for elastic scaling, configuring an external type trigger and passing in the required metadata, which mainly includes the following fields:

• scalerAddress is the address used by keda-operator when calling keda-tencentcloud-clb-scaler.
• loadBalancerId is the CLB instance ID.
• metricName is the CLB monitoring metric name. Most metrics for public and private networks are the same. Refer to the official documentation for specific metric lists: Public Network Load Balancer Monitoring Metrics and Private Network Load Balancer Monitoring Metrics.
• threshold is the metric threshold for scaling, i.e., it decides whether to scale by comparing metricValue / Pod count with the threshold value.
• listener is the only optional configuration, specifying the CLB listener for monitoring metrics, format: protocol/port.

Configuration Example 1: Elastic Scaling Based on CLB Connection Count Metric

apiVersion: keda.sh/v1alpha1
kind: ScaledObject
metadata:
  name: httpbin
spec:
  scaleTargetRef:
    apiVersion: apps/v1
    kind: Deployment
    name: httpbin
  pollingInterval: 15
  minReplicaCount: 1
  maxReplicaCount: 100
  triggers:
    - type: external
      metadata:
        scalerAddress: clb-scaler.keda.svc.cluster.local:9000
        loadBalancerId: lb-xxxxxxxx
        metricName: ClientConnum # Connection count metric
        threshold: "100" # Each Pod handles 100 connections
        listener: "TCP/8080" # Optional, specify listener, format: protocol/port

Configuration Example 2: Elastic Scaling Based on CLB QPS Metric

apiVersion: keda.sh/v1alpha1
kind: ScaledObject
metadata:
  name: httpbin
spec:
  scaleTargetRef:
    apiVersion: apps/v1
    kind: Deployment
    name: httpbin
  pollingInterval: 15
  minReplicaCount: 1
  maxReplicaCount: 100
  triggers:
    - type: external
      metadata:
        scalerAddress: clb-scaler.keda.svc.cluster.local:9000
        loadBalancerId: lb-xxxxxxxx
        metricName: TotalReq # Requests per second metric
        threshold: "500" # Average of 500 QPS per Pod
        listener: "TCP/8080" # Optional, specify listener, format: protocol/port