Skip to content

Latest commit

 

History

History
114 lines (69 loc) · 3.79 KB

DEPLOYMENT.md

File metadata and controls

114 lines (69 loc) · 3.79 KB
Raw

Deploying Orca

This guide serves to describe how to deploy Orca and its components in a production environment.

General Requirements

The following tools must be present on the host machine running Orca (where 'x' means any minor/patch sub-version):

  • Docker v27.x.x
  • PostgreSQL v10.x.x

Both the Orchestrator and the Worker(s) will utilize these tools when interacting with grading jobs and grader images.

The orca-grader-base Docker Image

All grader images that Orca may create on demand are built from the orca-grader-base Docker image; it is necessary that this must exist before running the system.

The Dockerfile for the base image is located in the worker/ directory, and can be built by running the following command:

cd worker/
docker build -t orca-grader-base -f images/orca-grader-base.Dockerfile .

In particular, if the Orchestrator and Worker are being run on separate machine instances, please ensure the Orchestrator's machine has this image built by its local Docker service.

The Orchestrator

The Orchestrator expects to be run with Node version 21.0.0.

Environment Variables

The following environment variables need to be set upon deploying the orchestrator:

  • ENVIRONMENT - Used for logging configuration and other conditionals; this should be set to PRODUCTION.
  • POSTGRES_URL - Used for connecting to and interacting with Orca's database; see this Prisma page for the URL format.
  • API_PORT - The port on which the API should listen for incoming connections.

Installing Dependencies

Dependencies can be installed and built from the top level of the orchestrator/ directory.

yarn install
yarn workspaces run build

From there, Prisma will be installed, and migrations for setting up the database can be executed by running the following:

cd packages/db
npx prisma migrate deploy

NOTE: This will not work if the POSTGRES_URL is not set.

This will create the schemas needed for grading jobs and grader image builds to be created.

Running the Orchestrator: A Tale of Two Cities

The Orchestrator is made up of two components: the Web API and the Image Builder Service.

There are two ways to run them: (1) as two separate terminal sessions or (2) with a single command.

Option (1):

In two different terminal sessions, run the following:

  • Session A: yarn server
  • Session B: yarn image-builder

Option (2):

The node tool concurrently can be utilized to run the API and Build Service in one session; this has been aliased to this command:

yarn start

The Worker

The Worker expects to be run with Python version 3.12.5.

Environment Variables

The following environment variables need to be set upon deploying a worker instance:

  • ENVIRONMENT - Used for certain conditional logic on the worker; this should be set to PRODUCTION.
  • POSTGRES_URL - Used for fetching jobs from the queue; this is the same URL for the running PostgreSQL database instance.
  • ORCA_WEB_SERVER_HOST - Needed for fetching grader images that don't exist on the worker's machine; this should be whatever the Orcherstrator's API URL is.

Installing Dependencies

To install all of its dependencies, run the following commands:

python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

The venv Python command ensures installed packages are local to the worker/ directory instead of installing them system-wide.

NOTE: unless actively running the Worker instance, the python virtual environment can be exited by typing deactivate and hitting ENTER in a given terminal session.

Running the Worker

The worker can be run by simply executing the following:

source .venv/bin/activate
python -m orca_grader