Skip to content

k8s-benchmark.md

Latest commit

 

History

History
23 lines (17 loc) · 1.85 KB

k8s-benchmark.md

File metadata and controls

23 lines (17 loc) · 1.85 KB
Raw

Setup Benchmarks

The code here allows for quick setup on an EKS cluster for the benchmarks.

The benchmark tests are quite straightforward, we simultaneously publish and consume as many 1Kb messages as possible from the queue (like one would in a real-world scenario). The consumer & producer are run in separate processes.

For PGMB and PGMQ, we delete the messages on "ack", rather than archive them to squeeze out the most performance. Of course, we could try it out with an unlogged table for even more performance, but that would not be fair to RabbitMQ, as we're using durable queues there.

Provisioning

  1. Have an AWS account with an active card. Make sure you have an AWS profile added to your local machine.
  2. Install:
  3. Create the cluster: eksctl create cluster -f k8s-benchmark/cluster.yaml
    • Note: set AWS_PROFILE=your_profile if you have multiple profiles.
  4. Switch to the new cluster: aws eks update-kubeconfig --name pgmb-benchmark-cluster
  5. Give permissions to the cluster to attach to EBS volumes: sh k8s-benchmark/provision-ebs.sh
  6. Run the benchmarks: kubectl apply -f k8s-benchmark/benchmark-pgmb.yaml. Note that the pods may crash initially, but they will restart and run the benchmarks.
  7. You can check the logs with kubectl logs -f <pod-name> -n pgmb-benchmark. The output logs can be used to then graph the results. Terminate the pods when done.

Note: this will cost you some money, so make sure to delete the cluster when you're done via: eksctl delete cluster -f k8s-benchmark/cluster.yaml --disable-nodegroup-eviction. Also if you want to run more rigorous benchmarks, you can increase the size of the node in the cluster.yaml file, and increase the provisioned size of the postgres, consumer, publisher pods.