Skip to content

kedacore/sample-go-rabbitmq

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

28 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

RabbitMQ consumer and sender

A simple docker container that will receive messages from a RabbitMQ queue and scale via KEDA. The receiver will receive a single message at a time (per instance), and sleep for 1 second to simulate performing work. When adding a massive amount of queue messages, KEDA will drive the container to scale out according to the event source (RabbitMQ).

Pre-requisites

Setup

This setup will go through creating a RabbitMQ queue on the cluster and deploying this consumer with the ScaledObject to scale via KEDA. If you already have RabbitMQ you can use your existing queues.

First you should clone the project:

git clone https://github.com/kedacore/sample-go-rabbitmq
cd sample-go-rabbitmq

Creating a RabbitMQ queue

Install RabbitMQ via Helm

Since the Helm stable repository was migrated to the Bitnami Repository, add the Bitnami repo and use it during the installation:

helm repo add bitnami https://charts.bitnami.com/bitnami
Helm 3

RabbitMQ Helm Chart version 7.0.0 or later

helm install rabbitmq --set auth.username=user --set auth.password=PASSWORD bitnami/rabbitmq --wait

Notes:

  • If you are running the rabbitMQ image on KinD, you will run into permission issues unless you set volumePermissions.enabled=true. Use the following command if you are using KinD:

    helm install rabbitmq --set auth.username=user --set auth.password=PASSWORD --set volumePermissions.enabled=true bitnami/rabbitmq --wait
    
  • With RabbitMQ Helm Chart version 6.x.x or earlier, username and password should be specified with rabbitmq.username and rabbitmq.password parameters https://hub.helm.sh/charts/bitnami/rabbitmq

Helm 2

RabbitMQ Helm Chart version 7.0.0 or later

helm install --name rabbitmq --set auth.username=user --set auth.password=PASSWORD bitnami/rabbitmq --wait

Notes:

  • If running this demo on a computer with a ARM Processor, refer to the earlier note
  • If using KinD refer to the earlier note
  • For RabbitMQ Helm Chart version 6.x.x or earlier, refer to the earlier note

Wait for RabbitMQ to Deploy

⚠️ Be sure to wait until the deployment has completed before continuing. ⚠️

kubectl get po

NAME         READY   STATUS    RESTARTS   AGE
rabbitmq-0   1/1     Running   0          3m3s

Deploying a RabbitMQ consumer

Deploy a consumer

kubectl apply -f deploy/deploy-consumer.yaml

Validate the consumer has deployed

kubectl get deploy

You should see rabbitmq-consumer deployment with 0 pods as there currently aren't any queue messages and for that reason it is scaled to zero.

NAME                DESIRED   CURRENT   UP-TO-DATE   AVAILABLE   AGE
rabbitmq-consumer   0         0         0            0           3s

This consumer is set to consume one message per instance, sleep for 1 second, and then acknowledge completion of the message. This is used to simulate work. The ScaledObject included in the above deployment is set to scale to a minimum of 0 replicas on no events, and up to a maximum of 30 replicas on heavy events (optimizing for a queue length of 5 message per replica). After 30 seconds of no events the replicas will be scaled down (cooldown period). These settings can be changed on the ScaledObject as needed.

Publishing messages to the queue

Deploy the publisher job

The following job will publish 300 messages to the "hello" queue the deployment is listening to. As the queue builds up, KEDA will help the horizontal pod autoscaler add more and more pods until the queue is drained after about 2 minutes and up to 30 concurrent pods. You can modify the exact number of published messages in the deploy-publisher-job.yaml file.

kubectl apply -f deploy/deploy-publisher-job.yaml

Validate the deployment scales

kubectl get deploy -w

You can watch the pods spin up and start to process queue messages. As the message length continues to increase, more pods will be pro-actively added.

You can see the number of messages vs the target per pod as well:

kubectl get hpa

After the queue is empty and the specified cooldown period (a property of the ScaledObject, default of 300 seconds) the last replica will scale back down to zero.

Cleanup resources

kubectl delete job rabbitmq-publish
kubectl delete ScaledObject rabbitmq-consumer
kubectl delete deploy rabbitmq-consumer
helm delete rabbitmq

About

Sample KEDA deployment using a Go RabbitMQ consumer

Resources

License

Code of conduct

Stars

Watchers

Forks

Releases

No releases published

Packages