label-studio-deployment

A simple terraform + packer configuration to deploy Label Studio to a single machine with a custom subdomain.

User Guide

Requirements

1. Google Cloud CLI.
2. Application Default Credentials.
3. A domain name where you have configured its name servers to a Cloud DNS zone.
4. terraform and packer.
5. make (probably comes with your distro).
6. Any public ssh key in ~/.ssh that matches id_*.pub (and its private counterpart).
7. [optional] docker + docker compose if you want to try locally.

Steps to deploy

First you need to define some configurations. You need to set a username and password for the default account in Label Studio, and provide your project in GCP and a Cloud DNS zone to create the final subdomain. These configurations are written in HCL, and so strings should be quoted.

# image/packer.auto.pkrvars.hcl
label_studio_username = <your username> # must be an email like [email protected]
label_studio_password = <your password>

# infra/terraform.tfvars
project   = <you gcp project>
dns_zone  = <cloud dns zone from point 3. in requirements>
subdomain = <the name you chose for the subdomain> # optional, defaults to "labelstudio"

For convenience, you can see your default project with gcloud config get-value project and your dns zones with gcloud dns managed-zones list.

Next, run the following

$ make init
$ make

Done! This has created a machine image and deployed it to a compute engine instance. The url of the instace is

$ echo https://$(terraform -chdir=infra output -raw label_studio_domain)

where you can login using the username and password that you configured in image/packer.auto.pkrvars.hcl.

For simple management you can ssh to it via the command

$ ssh $(terraform -chdir=infra output -raw ssh_connect)

Clean Up

Run

$ make clean

This will destroy the cloud infrastructure (only what was created, you dns zone is left intact), and remove the images created, so that they don't incur in storage charges.

Warning

This project uses Let's Encrypt for SSL certificates. Each time a new machine image is created by Packer a new certificate is requested, and so you must be wary of the rate limits of let's encrypt. In practice, this means you shouldn't run make more than 50 times a week using the same Cloud DNS zone and subdomain.

To simplifly turning off and back on the deployed machine (without running make and building a new image each time) i've provided two simple commands make terminated and make running which should be equivalent to terminating and turning back on the compute instance via de GCP Console.