Terraform module for load testing a Scylla cluster with scylla-bench.

Profiling Scylla cluster is often limited by network throughput of a host running scylla-bench utility (or cassandra-stress one). Also to minimalize network latency effect it is best to colocate scylla-bench nodes near datacenters the cluster is running in. To perform an efficient load the following factors need to be considered:

• run scylla-bench from host being as close to the cluster as possible
• ensure the host is capable of withstanding sufficient bandwidth
• speed up by sharding the load among a number of concurrent scylla-bench hosts

In addition the procedure usually involves a number of simple steps, which this Terraform module helps to automate:

• concurrently spawns a number of EC2 instances
  • amount of the instances is configured by instances variable (4 being a default)
  • size of a single instance is controlled by aws_instance_type variable (c4.large being a default)
• provisions the instances to prepare for the load
• loads the schema from a single instance and waits until it's propagated
  • schema can be configured by schema variable (replication(factor=3) being a default)
  • for more complex usecases schema script can also be configured with create_schema_script variable (scripts/create-schema.sh being a default)
• once the environment is ready, the load begins:
  • there are a number of variables, that can alter the default behaviour: consult Variables section for more information
  • the load script itself can also be overwritten with write_script variables (scripts/write.sh being a default)

Once the load if finished (the fact needs to be observed externally, e.g. using Scylla Monitoring stack), the infrastracture created by this module can be safely teared down.

$ cat main.tf 

module "scylla-bench" {
	source  = "github.com/rjeczalik/terraform-aws-scylla-bench"
	username = "scylla_admin"
	password = "qwerty123"
	seeds = ["66.249.65.11", "66.249.66.11", "66.249.67.11"]
	instances = 4
	keys = 1000000000
	limit = 10000
}

Once you configure the module, start the profiling with:

$ terraform apply

When it finishes, tear it down with:

$ terraform destroy

• add support for profile-driven loads (uploading profile file)
• add support for spot instances
• add support for external VPC (private cluster + vpc peering)