You can visit the live project here.

Since the AI jobs processing the images to remove their background are long running tasks (5 to 10 seconds), we have to run them asynchronously. To achieve that I used a Redis instance that will be used as a queue to store the AI Jobs submitted by the Laravel Web App. The Web App will return return a 200 http code when a user upload an image and a Job will be queued. A mysql db is also used to track the jobs (failed jobs, job status). Once the Job is sumbitted, a Worker service is listening to the queue and start processing it asynchronously. Once the AI job is done, the Worker service broadcasts an Event with the image url and the event is notified to the user thanks to the Websocket server.

I also wanted to build an app that can easily scale, especially an app that can scale horizontally. To achieve that I decoupled the storage layer from the Web app and from the Worker service. I used a separate Minio Cloud Storage service for that. Thanks to that the Worker Service and the Web app are stateless and can be replicated as needed and can the instances can live in separate servers.

In the production server there are 4 Workers services and the AI tasks are distributed equally in a round robbin manner among the different workers.

1. Start the services used by the app: (mysql, redis, minio, reverb)

docker compose -f dev.docker-compose.yaml up -d 

2. Run the migrations

php artisan migrate

This command will spin up a Websocket Reverb Server, a Redis for the queued Jobs a Minio server for storing the images and a Mysql server.

3. Run the php server in a terminal

php artisan serve

4. In another terminal run the Vite server for hot reloading of Front assets (usefull for frontend dev)

npm run dev

5. Run a php queue worker to process Jobs

php artisan queue:work

6. Browse https://localhost:8000

All the configuration variables used in the app are in the file .env at the root of the project.

sudo ./install_or_update.sh

1. Choose local when environment is prompted.
2. Choose a domain name it will automatically add appropirate lines in /etc/hosts in order to route the requests to your domain name to localhost.
3. Choose the number of workers that process the Background Removal tasks.
4. You can choose either or not to build the worker image. It does not depend on domain name and it takes a long time to build (it needs to download AI models) so you can skip the build if you did not change the worker code after the last worker build.
5. Browse http://yourdomain.com.

1. Buy a VPS server with sufficiant RAM and CPU (at least 4GO RAM and 2 cores CPU)
2. Buy a domain name yourdomain
3. Manage DNS records of yourdomain to point to your VPS ip address :
   • yourdomain -> ip address
   • reverb.yourdomain -> ip address
   • minio.yourdomain -> ip address
   • adminio.yourdomain -> ip address

sudo ./install_or_update.sh

4. Choose production when environment is prompted.
5. You can choose either or not to build the worker image. It does not depend on domain name and it takes a long time to build (it needs to download AI models) so you can skip the build if you did not change the worker code after the last worker build.
6. Browse https://yourdomain.com.

A CI/CD pipeline has been created thanks tot Github Actions. On new github release the 3 images: rembg, webserver and worker are build and pusehd to my Docker Hub registery. Then It connects to my VPS through ssh pull the new docker images and tells docker compose to recreate the containers with the new images. The CI/CD yaml file is located at .github/workflows/deploy.yaml.

gh release create <tag_name> --title "<release_title>" --notes "<release_notes>"

Example:

gh release create v1.0.0 --title "First Release" --notes "a Handy app to automtically detect and remove background from images"

• Use act command. It will use the secret file: .secrets to fill the secrets in the deploy.yaml manifest.

act

• Test a specific Job

act -j build

• SEO optimization with some Laravel Package maybe ?
• Disable Session cookie for unauthenticated users (the app does not track user) so they are useless. Then it will allow to cache the home page at nginx level or Cache it using a CDN. It will improve home page response time which is good for SEO.
• Video support
• Create Kubernetes manifests and a Helm package to allow autoscaling horizontally in a cluster of nodes or maybe use Swarm ?