Deploy the CPU render-farm worker pool template with OpenTofu

Stand up a dispatcher VM plus CPU workers that pull render jobs and upload results to Object Storage using the validated OpenTofu template render-farm-worker. Quake AI has no GPU rendering; this template provisions a CPU-only render pool for hobbyist previews, overnight batch jobs, and audio-render workloads.

Prerequisites#

You need:

• A Quake AI account with application credentials
• OpenTofu 1.6.0 or later (installation guide)
• The AWS CLI and curl installed
• OpenStack credentials sourced into the shell. See the OpenStack CLI guide.
• A copy of the render-farm-worker template from the template reference page
• Enough project quota for one dispatcher and two c2a.large workers by default, plus two Object Storage buckets

Step 1: Mint credentials and configure variables#

Mint EC2-compatible credentials and export the standard AWS variables:

openstack ec2 credentials create
export AWS_ACCESS_KEY_ID=YOUR_ACCESS_KEY
export AWS_SECRET_ACCESS_KEY=YOUR_SECRET_KEY
export AWS_ENDPOINT_URL_S3=https://object.us-east-1.rumble.cloud

Copy terraform.tfvars.example to terraform.tfvars and set:

key_name              = "YOUR_KEY_NAME"
jobs_bucket_name      = "YOUR_PROJECT_RENDER_JOBS"
output_bucket_name    = "YOUR_PROJECT_RENDER_OUT"
s3_access_key         = "YOUR_ACCESS_KEY"
s3_secret_key         = "YOUR_SECRET_KEY"
enable_dispatcher_fip = true

worker_count = 2

Workers stay on the private subnet with no floating IPs. Setting enable_dispatcher_fip = true allocates one floating IP on the dispatcher so your workstation can reach the job API.

Step 2: Apply the template#

From the template directory, run:

tofu init
tofu plan
tofu apply

Type yes when prompted. Cloud-init installs the dispatcher API and worker agents on each node. First boot can take 10 to 15 minutes while FFmpeg and Blender install on the workers.

When the run finishes, note job_submit_url, jobs_bucket, and output_bucket from the outputs.

Step 3: Submit a sample job and verify output#

Upload a short test asset:

ffmpeg -f lavfi -i "color=c=blue:s=320x240:d=2" -pix_fmt yuv420p /tmp/frame-source.mp4
aws s3 cp /tmp/frame-source.mp4 \
  "s3://$(tofu output -raw jobs_bucket)/jobs/tutorial-frame/source.mp4" \
  --endpoint-url "$AWS_ENDPOINT_URL_S3"

Create and submit a job manifest:

cat > /tmp/tutorial-job.json <<'EOF'
{
  "job_id": "tutorial-frame",
  "input_key": "jobs/tutorial-frame/source.mp4",
  "output_prefix": "jobs/tutorial-frame/out/",
  "profile": "ffmpeg-transcode",
  "params": { "format": "mp4" }
}
EOF

curl -sS -X POST \
  -H "Content-Type: application/json" \
  --data-binary @/tmp/tutorial-job.json \
  "$(tofu output -raw job_submit_url)"

Workers poll the queue on the private subnet. Allow 10 to 15 minutes after tofu apply for cloud-init to finish before you expect rendered output. Poll the output bucket:

aws s3 ls "s3://$(tofu output -raw output_bucket)/tutorial-frame/" \
  --endpoint-url "$AWS_ENDPOINT_URL_S3"

You should see render.mp4 under the tutorial-frame/ prefix. If nothing appears after 15 minutes, confirm dispatcher and worker instances are Active in the Console.

Next steps#

• CPU render-farm worker pool template
• Media streaming
• Self-Hosted AI on Cloud Infrastructure
• Monitoring stack template when you set enable_monitoring = true

Clean up#

Empty both buckets, then run tofu destroy from the project directory:

aws s3 rm "s3://$(tofu output -raw jobs_bucket)" --recursive --endpoint-url "$AWS_ENDPOINT_URL_S3"
aws s3 rm "s3://$(tofu output -raw output_bucket)" --recursive --endpoint-url "$AWS_ENDPOINT_URL_S3"
tofu destroy

Delete the EC2-compatible credential with openstack ec2 credentials delete YOUR_ACCESS_KEY if you created one only for this deployment.