hashicorp / c2m

Nomad, Terraform, and Packer configurations for the 2 Million Container Challenge (C2M)

C2M: HashiCorp Nomad 2 Million Container Challenge

This repository contains all of the code to run scalability tests on Nomad for the second iteration of the C1M challenge:

C2M

Overview

1. Use Packer to build images
2. Use Terraform to provision images
3. Use command and control instance to run journey

C2M Parameters

The following commands were used to provision and run C2M:

# From an operators local machine
terraform apply -var cluster_capacity=18000 -var 'nomad_num_schedulers=6' -var 'dd_key=<REDACTED>' -parallelism=20

# From the command and control instance output by terraform
# scp nomad/bench.nomad to command and control instance
# compile journey/ and scp to command and control instance
WORKERS=75 JOBS=2000 JOBSPEC=./bench.nomad PREFIX=c2m- ./journey start

Packer

Required: Packer 1.6+

Packer's HCL2 support is being used, so refer to the Packer 1.5 docs for reference.

Building Images

cd packer

# Run once per region
packer build -var-file regions/us-west-2.pkrvars.hcl .

Run once per region file to build an image per region.

Terraform

Required: Terraform v0.12+

Provisioning Infrastructure

To create or update infrastructure:

cd terraform
terraform init  # On first run only

# The retry_join_tag allows everyone to have their own cluster.
terraform plan -var retry_join_tag="$USER"  
terraform apply -var retry_join_tag="$USER"  

# When you're done
terraform destroy -var retry_join_tag="$USER"

Accessing Cluster

After provisioning infrastructure with Terraform above it may be accessed via a helper script:

./nomad.sh status

ssh

Add your ssh key to terraform/variables.tf and login as the admin user:

ssh admin@...