Kedify | Prometheus

Use events from Prometheus to trigger autoscaling with Kedify and KEDA

Prometheus is an open-source monitoring and alerting system designed for reliability and scalability

It supports service-oriented architectures through a robust data model with time series data identified by metric name and key/value pairs. It collects and stores metrics as time series data.

Featured Use Cases

Scenario:

An e-commerce platform needs to scale its recommendation service based on the number of user interactions to ensure personalized content is delivered quickly.

Prometheus Scaler Usage:

user_interaction_count - a custom metric tracking the number of user interactions with the recommendation engine.

KEDA Usage:

Adapt to changing user activity levels. By monitoring user_interaction_count with KEDA and Prometheus, the platform can dynamically scale the recommendation service in response to changing user interaction levels, ensuring personalized content is served efficiently.

Get Started

                      apiVersion: keda.sh/v1alpha1
kind: ScaledObject
metadata:
  name: business-metrics-scaledobject
  namespace: default
spec:
  scaleTargetRef:
    name: recommendation-service
  minReplicaCount: 3
  maxReplicaCount: 25
  triggers:
  - type: prometheus
    metadata:
      serverAddress: http://prometheus.default.svc:9090
      threshold: '1000'
      query: sum(rate(user_interaction_count[5m]))

Scenario:

A web application needs to scale based on the number of incoming HTTP requests to ensure it meets response time SLAs.

Prometheus Scaler Usage:

http_requests_total - a metric tracking the total number of HTTP requests received.

KEDA Usage:

Respond to traffic spikes. Use KEDA with the Prometheus scaler to monitor the rate of incoming HTTP requests (rate(http_requests_total[2m])) and scale the number of pods up or down based on thresholds defined in the KEDA ScaledObject.

Get Started

                      apiVersion: keda.sh/v1alpha1
kind: ScaledObject
metadata:
  name: http-load-scaledobject
  namespace: default
spec:
  scaleTargetRef:
    name: web-app-deployment
  minReplicaCount: 1
  maxReplicaCount: 10
  triggers:
  - type: prometheus
    metadata:
      serverAddress: http://prometheus.default.svc:9090
      threshold: '500'
      query: sum(rate(http_requests_total{job="web-app"}[2m]))

Scenario:

An API needs to scale not only based on request volume but also on the error rate to maintain a high-quality service.

Prometheus Scaler Usage:

http_requests_total and http_requests_errors_total - metrics tracking the total number of requests and the number of those requests that resulted in an error, respectively.

KEDA Usage:

Increase capacity to improve error handling. KEDA can use a query that calculates the error rate (rate(http_requests_errors_total[5m]) / rate(http_requests_total[5m])) and triggers scaling actions when the error rate exceeds a certain threshold, potentially adding resources to troubleshoot or handle increased error rates more gracefully.

Get Started

                      apiVersion: keda.sh/v1alpha1
kind: ScaledObject
metadata:
  name: error-rate-scaledobject
  namespace: default
spec:
  scaleTargetRef:
    name: api-deployment
  minReplicaCount: 2
  maxReplicaCount: 10
  triggers:
  - type: prometheus
    metadata:
      serverAddress: http://prometheus.default.svc:9090
      threshold: '0.05' # 5% Error rate
      query: |
        sum(rate(http_requests_errors_total{job="api"}[2m]))
        /
        sum(rate(http_requests_total{job="api"}[2m]))

Scenario:

An application that relies on a database connection pool needs to scale based on the number of active connections to maintain performance and avoid hitting connection limits.

Prometheus Scaler Usage:

db_active_connections - a custom metric that tracks the number of active database connections.

KEDA Usage:

Ensure database performance under varying loads. Configure KEDA to scale the application based on the db_active_connections metric - ensuring that as the number of active connections increases, more application instances are deployed to distribute the load and maintain database performance.

Get Started

                      apiVersion: keda.sh/v1alpha1
kind: ScaledObject
metadata:
  name: db-connection-pool-scaledobject
  namespace: default
spec:
  scaleTargetRef:
    name: database-app
  minReplicaCount: 1
  maxReplicaCount: 10
  triggers:
  - type: prometheus
    metadata:
      serverAddress: http://prometheus.default.svc:9090
      threshold: '200'
      query: avg_over_time(db_active_connections{app="my-database-app"}[5m])