Multi-Level Service Synchronized Horizontal Scaling (Workload Triggers)

Recent Pages

Multi-Level Service Synchronized Horizontal Scaling (Workload Triggers)

Last updated: 2024-12-24 15:55:32

Workload Triggers
Kubernetes-based Event-Driven Autoscaler (KEDA) supports Kubernetes Workload triggers, enabling scaling based on the number of Pods in one or more workloads. This is very useful in multi-level service call scenarios. For details, please refer to KEDA Scalers: Kubernetes Workload.
Use Cases
Multi-level Service Simultaneous Scaling
The picture shows multi-level microservice call:
﻿
The services A, B, and C usually have a fixed proportional quantity.
If the pressure on A suddenly increases, forcing a scale-out, B and C can also scale out almost simultaneously with A by using KEDA's Kubernetes Workload triggers, without waiting for pressure to propagate slowly.
First, configure the scale-out for A, which can be based on CPU and memory pressure. For example:
apiVersion: keda.sh/v1alpha1
kind: ScaledObject
metadata:
  name: a
spec:
  scaleTargetRef:
    apiVersion: apps/v1
    kind: Deployment
    name: a
  pollingInterval: 15
  minReplicaCount: 10
  maxReplicaCount: 1000
  triggers:
    - type: memory
      metricType: Utilization
      metadata:
        value: "60"
    - type: cpu
      metricType: Utilization
      metadata:
        value: "60"
Then, configure the scale-out for B and C, assuming a fixed ratio of A:B:C = 3:3:2. For example:
apiVersion: keda.sh/v1alpha1
kind: ScaledObject
metadata:
  name: b
spec:
  scaleTargetRef:
    apiVersion: apps/v1
    kind: Deployment
    name: b
  pollingInterval: 15
  minReplicaCount: 10
  maxReplicaCount: 1000
  triggers:
    - type: kubernetes-workload
      metadata:
        podSelector: 'app=a' # Select service A
        value: '1' # A/B=3/3=1
apiVersion: keda.sh/v1alpha1
kind: ScaledObject
metadata:
  name: c
spec:
  scaleTargetRef:
    apiVersion: apps/v1
    kind: Deployment
    name: c
  pollingInterval: 15
  minReplicaCount: 3
  maxReplicaCount: 340
  triggers:
    - type: kubernetes-workload
      metadata:
        podSelector: 'app=a' # Select service A
        value: '3' # A/C=3/2=1.5
With the above configuration, when the pressure on A increases, A, B, and C will scale out almost simultaneously without waiting for the pressure to propagate step by step. This allows for faster adaptation to pressure changes, improving system elasticity and performance.
﻿