swpp-2023-project-team-1

Development Rule

1. Git

1.1 Branch Naming

main

The branch that can be released as product.

main

develop

The branch being developed for next release.

develop

feature

The branch where features are developed.

feat/{feature-name}

release

The branch preparing for this release

release-{version}

hotfix

The branch that fixes bugs encountered in the released version

hotfix-{version}

1.2 Commit Rule

type	description
feat	New features
fix	Bug fix
build	Modifying build-related files, installing or deleting modules
chore	Other minor changes
docs	Document modifications
style	Code style and Format
refactor	Code refactoring
test	Modifying for test code
perf	Performance improvements

[feat] {new feature} [chore] {minor changes}

2. Merge and Resolve Conflicts

2.1 Rebase and Merge

Rebase does not leave any commit history, so the commit history becomes cleaner.

2.2 Resolve Conflicts in Local

To resolve conflicts that occur when trying to merge your branch into the main branch

### Now you are in {your branch}

### Move to main branch and pull
git checkout main
git pull

### Move to {your branch} and merge main
git checkout {your branch}
git merge main

### Resolve Conflicts on your code editor.

### Push your code
git push

2.3 Pull Request

Merge Pull request

Frontend

Backend

1. Initial Setup

Install Python env and dependencies

Setting up the FastAPI development environment.

Install poetry

curl -sSL https://install.python-poetry.org | python3 -

Move from the project root path to the backend directory, create a virtual environment, and install the package.

cd backend
poetry install

Activate Virtual Environment

poetry shell

For other commands, refer to poetry official documentation

e.g) Add packages to pyproject.toml and installs them.

poetry add pytest --group local

e.g) Manage Environments.

poetry env use 3.11

IDE setting

VSCode

Create a .vscode/settings.json file and paste the contents below.

{
  "python.linting.mypyEnabled": true,
  "python.linting.pylintEnabled": false,
  "python.linting.enabled": true,
  "python.linting.flake8Enabled": false,
  "[python]": {
    "editor.defaultFormatter": "ms-python.black-formatter"
  }
}

2. Docker

The actual development environment and deployment environment are Dockerized. Therefore, you must install Docker and run the docker compose command. For Docker compose v1, use the docker-compose command.

Local

In your local environment, Build and run using the docker-compose.local.yaml file. (only DB uses Docker)

docker compose -f docker-compose.local.yaml up -d --build 
uvicorn app.main:spire_app --reload --log-level debug --host 0.0.0.0 --port 8000

Production

### In ec2 environment. Build and run using the docker-compose.prod.yaml file.

docker compose -f docker-compose.prod.yaml up -d --build

Build

docker compose up --build: If there are code changes, rebuild and run the container.

Run

docker compose up: Runs a Docker container. If it is not built, build it once at first.

Stop

docker compose down: Terminates the Docker container.
docker compose down -v: Terminates the Docker container and deletes the volume as well.

For detailed options, see link.

3. Local Development

With the Docker db container running, run the command below to proceed with development. Local Dockerfile

Install Dependencies

poetry install

Run

cd backend
poetry shell

uvicorn app.main:spire_app --reload --host 0.0.0.0

Test

pytest

Lint

flake8 app
mypy app

Format

black app

4. Migration (Alembic)

Make Migration

alembic revision --autogenerate -m "type your commit message"

Afterwards, the migration history is saved in ./backend/migrations/versions.

Migrate

This command has no practical use. It is set to run automatically when the docker compose up --build command is executed.

alembic upgrade head

Downgrade

alembic downgrade -1

Inference server

1. Build Triton server image

Build Triton server image

Download Dockerfile in /inference_server. Following command will build a docker image for our Triton inference server.

docker build -t spire_ai - < Dockerfile

You may need to run only one model in one container if you don't have sufficient memory.

docker run -it --name spire_stable_diffusion spire_ai /bin/sh
cd models
rm -rf open_seed
exit
docker commit spire_stable_diffusion spire_ai_stable_diffusion

docker run -it --name spire_open_seed spire_ai /bin/sh
cd models
rm -rf stable_diffusion
exit
docker commit spire_open_seed spire_ai_open_seed

2. Deploy on k8s on-premise environment

Deploy on k8s on-premise environment

Triton server can be deployed on various environment, but below is how we deployed this on our on-premise server. Please refer to https://github.com/triton-inference-server/server/tree/main for greater details. Our method is essentially adhoc since our permission on SNU GPU server is limited and we don't have any prior experience with k8s.

Download .yaml files in /inference_server/k8s.

First, following command will download all pretrained weights of our Triton inference server and initialize the server.

kubectl apply -f pod.yaml

If everything is okay and ready, it should be something like this.

Delete the pod before moving on.

kubectl delete pods --all

Second, these two command will create a deployment and a service. Whenever these are ready, server can get requests and send responses. You must fill in the NodePort numbers in service.yaml in order to create a service.

kubectl apply -f deployment.yaml
kubectl apply -f service.yaml

You can check the status of deployments and services via following commands.

kubectl get deployments
kubectl get services

The address for inference server will be as below. <public-node-ip> is node IP of SNU GPU server and <node-port> will be the number you selected in service.yaml for NodePort that corresponses to http.

http://<public-node-ip>:<node-port>

Don't forget to delete all services, deployments and pods once you are done.

kubectl delete services --all
kubectl delete deployments --all

Files

DEVELOPMENT.md

Latest commit

History