The existing Dockerfile includes several services installed at once and running in a single container. This should be improved using a docker-compose file in order to run RobotReviewer as a stack with separate services with the same standards of deployment that were applied to the other related services.
Solution
The changes in this Pull Request include:
RabbitMQ is initialized directly from the official Docker image.
Grobid is initialized directly form the official Docker image.
Separate Dockerfile and configuration files for running Bert-as-a-service server in a separate container.
Replaced run script for entrypoint.sh allowing different commands from base RobotReviewer image.
React app for Robotreviewer's Demo website is built in a preparation stage from an official Node Docker image in main Dockerfile.
Removed following instructions from main Dockerfile:
Node installation & build of static files for React app.
Java and Gradle installation and all Grobid related commands.
Bert installation and all Bert related commands.
Volumes and port exposure commands (moved to docker-compose).
Deployment using docker-compose file with 6 services:
robotreviewer-web:
Flask App for Robotreviewer Demo website, running with Gunicorn WSGI. It also runs a cronjob for uploaded PDFs database cleanup every 12 hours.
robotreviewer-api:
Flask App for Robotreviewer API, running with Gunicorn WSGI.
grobid:
Grobid service for generating arcticles bibliografic data.
Separate server.py and server_api.py scripts for Web & API Flask Apps for proper Celery App calls.
config.json file as a docker-compose config to be able to change the settings or modify RobotReviewer API keys without the need of rebuilding the image.
Template file .envTemplate for necessary environment variables.
Development compose file docker-compose.dev.yml runs web & api services with Flask development server instead of Gunicorn WSGI.
Compose file with GPU support docker-compose.gpu.yml runs celery, bert and grobid with access to GPU resources of the machine with Nvidia CUDA drivers.
How to test
Follow the instructions at README.md for running RobotReviewer with Docker and test the demo website exposed at http://localhost:5050. Upload some PDFs and verify that the reports are generated correctly.
Using cURL or Postman requests with the configured API-Key, test RobotReviewer API endpoints at http://localhost:5051.
(Optional) Run the update script for Pubmed articles from Trialstreamer while running RobotReviewer locally.
Problems
The existing Dockerfile includes several services installed at once and running in a single container. This should be improved using a docker-compose file in order to run RobotReviewer as a stack with separate services with the same standards of deployment that were applied to the other related services.
Solution
The changes in this Pull Request include:
RabbitMQ is initialized directly from the official Docker image.
Grobid is initialized directly form the official Docker image.
Separate Dockerfile and configuration files for running Bert-as-a-service server in a separate container.
Replaced
runscript forentrypoint.shallowing different commands from base RobotReviewer image.React app for Robotreviewer's Demo website is built in a preparation stage from an official Node Docker image in main Dockerfile.
Removed following instructions from main Dockerfile:
Deployment using docker-compose file with 6 services:
robotreviewer-web: Flask App for Robotreviewer Demo website, running with Gunicorn WSGI. It also runs a cronjob for uploaded PDFs database cleanup every 12 hours.
robotreviewer-api: Flask App for Robotreviewer API, running with Gunicorn WSGI.
robotreviewer-celery: Celery worker running annotating tasks involving Machine Learning.
rabbitmq: Messaging broker for Celery tasks.
bert: Bert as a service server.
grobid:
Grobid service for generating arcticles bibliografic data.
Separate
server.pyandserver_api.pyscripts for Web & API Flask Apps for proper Celery App calls.config.jsonfile as a docker-composeconfigto be able to change the settings or modify RobotReviewer API keys without the need of rebuilding the image.Template file
.envTemplatefor necessary environment variables.Development compose file
docker-compose.dev.ymlruns web & api services with Flask development server instead of Gunicorn WSGI.Compose file with GPU support
docker-compose.gpu.ymlruns celery, bert and grobid with access to GPU resources of the machine with Nvidia CUDA drivers.How to test
README.mdfor running RobotReviewer with Docker and test the demo website exposed at http://localhost:5050. Upload some PDFs and verify that the reports are generated correctly.