Pandemic-Knowledge

A fully-featured multi-source data pipeline for continuously extracting knowledge from COVID-19 data.

• Contamination figures
• Vaccination figures
• Death figures
• COVID-19-related news (Google News, Twitter)

What you can achieve

Live contaminations map + Latest news	Last 7 days news

France 3-weeks live map (Kibana Canvas)	Live vaccinations map

Context

This project was realized over 4 days as part of a MSc hackathon from ETNA, a french computer science school.

The incentives were both to experiment/prototype a big data pipeline and contribute to an open source project.

Install

Below, you'll find the procedure to process COVID-related file and news into the Pandemic Knowledge database (elasticsearch).

The process is scheduled to run every 24 hours so you can update the files and obtain the latest news

• Pandemic-Knowledge
  • What you can achieve
  • Context
  • Install
    • Env file
    • Initialize elasticsearch
    • Initialize Prefect
    • Run Prefect workers
    • COVID-19 data
    • News data
    • News web app

Env file

Running this project on your local computer ? Just copy the .env.example file :

cp .env.example .env

Open this .env file and edit password-related variables.

Initialize elasticsearch

Raise your host's ulimits for ElasticSearch to handle high I/O :

sudo sysctl -w vm.max_map_count=500000

Then :

docker-compose -f create-certs.yml run --rm create_certs
docker-compose up -d es01 es02 es03 kibana

Initialize Prefect

Create a ~/.prefect/config.toml file with the following content :

# debug mode
debug = true

# base configuration directory (typically you won't change this!)
home_dir = "~/.prefect"

backend = "server"

[server]
host = "http://172.17.0.1"
port = "4200"
host_port = "4200"
endpoint = "${server.host}:${server.port}"

Run Prefect :

docker-compose up -d prefect_postgres prefect_hasura prefect_graphql prefect_towel prefect_apollo prefect_ui

We need to create a tenant. Execute on your host :

pip3 install prefect
prefect backend server
prefect server create-tenant --name default --slug default

Access the web UI at localhost:8081

Run Prefect workers

Agents are services that run your scheduled flows.

1. Open and optionally edit the agent/config.toml file.

2. Let's instanciate 3 workers :

docker-compose -f agent/docker-compose.yml up -d --build --scale agent=3 agent

ℹ️ You can run the agent on another machine than the one with the Prefect server. Edit the agent/config.toml file for that.

COVID-19 data

Injection scripts should are scheduled in Prefect so they automatically inject data with the latest news (delete + inject).

There are several data source supported by Pandemic Knowledge

• Our World In Data; used by Google
  • docker-compose slug : insert_owid
  • MinIO bucket : contamination-owid
  • Format : CSV
• OpenCovid19-Fr
  • docker-compose slug : insert_france
  • Format : CSV (download from Internet)
• Public Health France - Virological test results (official source)
  • docker-compose slug : insert_france_virtests
  • Format : CSV (download from Internet)

1. Start MinIO and import your files according to the buckets evoked upper.

   For Our World In Data, create the contamination-owid bucket and import the CSV file inside.

   docker-compose up -d minio

   MinIO is available at localhost:9000

2. Download dependencies and start the injection service of your choice. For instance :

   pip3 install -r ./flow/requirements.txt
   docker-compose -f insert.docker-compose.yml up --build insert_owid

3. In Kibana, create an index pattern contamination_owid_*

4. Once injected, we recommend to adjust the number of replicas in the DevTool :

   PUT /contamination_owid_*/_settings
   {
       "index" : {
           "number_of_replicas" : "2"
       }
   }

5. Start making your dashboards in Kibana !

News data

There are two sources for news :

• Google News (elasticsearch index: news_googlenews)
• Twitter (elasticsearch index: news_tweets)

1. Run the Google News crawler :

docker-compose -f crawl.docker-compose.yml up --build crawl_google_news # and/or crawl_tweets

2. In Kibana, create a news_* index pattern

3. Edit the index pattern fields :

Name	Type	Format
img	string	Url
link	string with Type: Image with empty URL template	Url

4. Create your visualisation

News web app

Browse through the news with our web application.

1. Make sure you've accepted the self-signed certificate of Elasticsearch at https://localhost:9200

2. Start-up the app

   docker-compose -f news_app/docker-compose.yml up --build -d

3. Discover the app at localhost:8080

---

TODOs

Possible improvements :

• Using Dask for parallelizing process of CSV lines by batch of 1000
• Removing indices only when source process is successful (adding new index, then remove old index)
• Removing indices only when crawling is successful (adding new index, then remove old index)

Useful commands

To stop everything :

docker-compose down
docker-compose -f agent/docker-compose.yml down
docker-compose -f insert.docker-compose.yml down
docker-compose -f crawl.docker-compose.yml down

To start each service, step by step :

docker-compose up -d es01 es02 es03 kibana
docker-compose up -d minio
docker-compose up -d prefect_postgres prefect_hasura prefect_graphql prefect_towel prefect_apollo prefect_ui
docker-compose -f agent/docker-compose.yml up -d --build --scale agent=3 agent