Terraform project to create and benchmark aurora on AWS
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.
terraform initial_commit Sep 1, 2016


Work in progress.

This project is meant to be used with the express purpose of standing up an RDS cluster, running Aurora, in AWS for benchmarking.

This project requires using S3 as the remote storage backend.

terraform remote config \
    -backend=s3 \
    -backend-config="bucket=terraform-state-here" \
    -backend-config="key=project/terraform.tfstate" \

The project will provide:

  • VPC w/ 4 Subnets (2 public, 2 private)
  • Managed Nat Gateway(s)
  • Bastion box (jumpbox)
  • 2 Client instances to initiate the test
  • Security Groups (these are wide open, take caution)
  • Basic IAM roles&profiles
  • A Multi-zone RDS Aurora cluster w/ 2 instances @ r3.4xlarge
  • A Multi-zone RDS Aurora cluster w/ 2 instances @ r3.8xlarge

Intially stand up the 4x environment first. When connecting to RDS via the client instance, your endpoint may not be the writer. If this is the case, just initiate a failover in RDS for Aurora and you can begin testing in that AZ, or alternatively modify the "runtest.sh" file in the client instance with the appropriate hostname for the RDS instance you wish to connect to.

Run the gendata.sh script to populate the database. This script will add 100 tables with 20m rows each.

Prior to running the tests you must make a small change to the userlimits for your environment.

$ ulimit -n 4086

This will change the max open files to 4086 files as opposed to the initial setting of 1024, allowing us to run more than 1000 threads. The original test data only went up to 512, but we wanted to go further Aurora for our own curiousity.

After that is set you are almost ready to run the test.

I recommend firing off a snapshot now, so you can then spin-up the 8x environment from snapshot and avoid loading the data twice.

After you have a snapshot, go ahead and start testing your 4x cluster, while spinning up the 8x cluster.

The 8x cluster has it's own client node for benchmarking as well.