Access the hosted web server for model APIs

Tutorial
Setup
Access the course-provided web server for model deployment by connecting to the server via SSH.
Modified

November 20, 2024

In order to deploy ML models with {vetiver}, you need a cloud server capable of serving the API. You could use any web hosting platform that supports publishing sites via Docker containers. For this course we will provide you with a Cornell-hosted server that you can use to deploy your models.

Access the server via SSH

To access the server, you will need to use the Secure Shell Protocol (SSH). SSH is a cryptographic network protocol for operating network services securely over an unsecured network. It is widely used for secure communication between a client and a server.

Use the Cornell intranet to access the server

In order to access the server, you must either be physically located on-campus using the WiFi network, or logged in using the Cornell VPN. If you are not on-campus or connected to the VPN, you will not be able to access the server.

  • Open a Terminal window on your computer. Depending on your operating system, this may be called a Shell or Bash client.

    A screenshot of a terminal window in MacOS.

    A terminal window in MacOS

  • Run the following command from the terminal, replacing YOUR_NETID with your Cornell NetID:

    ssh YOUR_NETID@appliedml.infosci.cornell.edu

    You will be prompted to enter your password.

    A screenshot of a terminal window prompting the user to enter their password.

    Prompt to enter your password

    Enter your Cornell NetID password and press Enter.1

    A screenshot of a terminal window connected to the Cornell-hosted server via SSH.

    Successfully logged in to the server

1 If you’ve never used the terminal before, you will not see the characters of your password as you type them.

You are now logged into the server. Any commands you run in the terminal will be run on the server, not your local computer.

Create and store SSH key for Git/GitHub

You will build your Docker artifacts using R/RStudio either from your local computer or RStudio Workbench. In order to get those files on to the server, you will need to use Git to clone your repository. Unlike our previous communications with GitHub where we used a Personal Access Token (PAT), we will use SSH keys to authenticate with the server.

The Secure Shell Protocol (SSH) is another method for authenticating your identity when communicating with GitHub. Unlike how we just used SSH to login to the server using our NetID and password, we will generate a key pair that provides you with two long strings of characters: a public and a private key. You can place the public key on any server (like GitHub), and then unlock it by connecting to it with a client that already has the private key (the Cornell web server). When the two match up, the system unlocks without the need for a password.

The URL for SSH remotes looks like git@github.coecis.cornell.edu:<OWNER>/<REPO>.git. Make sure you use this URL to clone a repository. If you accidentally use the HTTPS version, the operation will not work.

Set up your SSH key

Note

You only need to do this authentication process one time on a single system.

  • Run the following command from the terminal to start an R session.

    R

A screenshot of a terminal window running an R session.

R session in the terminal

  • Run the R function credentials::ssh_keygen().

  • If you are asked to generate a key, use the console to select “yes”.

  • You will generate a key. It will begin with “ecdsa….” and look something like this:

    A screenshot of an SSH key.

    SSH key successfully generated
$key
[1] "/home/bcs88/.ssh/id_ecdsa"

$pubkey
[1] "ecdsa-sha2-nistp521 <REDACTED>"
  • Copy the entire string of characters (not including the quotation marks) and paste them into the settings page on GitHub. Give the key an informative title such as “INFO 4940/5940 API server”. Click “Add SSH key.”

Clone your Git repository

Now that you have set up your SSH key, you can clone your repository to the server. This will allow you to access the files you need to deploy your model.

In the terminal, clone the repository containing your Docker artifacts using the SSH URL. Replace <OWNER> and <REPO> with the owner and name of your repository. For example, if your repository is info4940-fa24/project-amazing-iguana, you would use:

git clone git@github.coecis.cornell.edu:info4940-fa24/project-amazing-iguana.git

Build your Docker container

Now that you have cloned your repository, you can build your Docker container. Navigate into the directory containing your Dockerfile using the cd command:

cd project-amazing-iguana

Follow the instructions in More about APIs and Docker to build your Docker container and deploy your model.

Selecting a port

When you deploy your model, you will need to specify a port for the server to listen on. The default port is 8000 but the firewall on the server blocks access to this port from the public. Instead, you should select any port between 2000-3000.

For example, say we will publish the api on port 2500. You need to specify this port in the following places:

  • When you build your Docker artifacts using vetiver_write_docker() or vetiver_prepare_docker(). You can specify the port using the port argument. in vetiver_write_docker() or in docker_args in vetiver_prepare_docker(). This ensures the Dockerfile exposes the correct port and runs the API on it.

    vetiver_write_docker(
  vetiver_model = v,
  port = 2500
)

vetiver_prepare_docker(
  board = board,
  name = "my-model",
  docker_args = list(
    port = 2500
  )
)

  • When you run your Docker image from the terminal using docker run. You can change the port using the -p argument. For example,

    docker run -p 2500:2500 my-image

Your API would then be accessible at http://appliedml.infosci.cornell.edu:2500.