Skip to content

ai-hospital-services/medicine-prescriber-prototype

Folders and files

NameName
Last commit message
Last commit date

Latest commit

Β 

History

6 Commits
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 

Repository files navigation

medicine-prescriber-prototype

build_backend build_frontend codecov code style: black license

Medicine prescriber prototype for AI-HOSPITAL.SERVICES

Table of Contents:

Repository map

 πŸ“Œ -------------------------------> you are here
┬
β”œβ”€β”€ .deploy   ---------------------> deployment related code
β”‚   β”œβ”€β”€ clusters   ----------------> contains fluxcd helm release pipelines as code for gitops based CD
β”‚   └── helm   --------------------> contains Helm charts for Kubernetetes workloads - backend api and frontend app
│── .github   ---------------------> github related workflows
β”‚   └── workflows   ---------------> contains code for running github workflows for CI and docker build and publish to GCR
│── .vscode   ---------------------> visual studio code debug configuration
│── backend   ---------------------> backend
β”‚   └── api   ---------------------> contains Python code for running backend api
β”‚   └── test   --------------------> contains Pytest code for testing application workloads in Google Kubernetes Engine (GKE)
│── frontend/app   ----------------> contains Dart code for Flutter based frontend app
│── machine_learning   ------------> machine learning experiments and training
β”‚   └── experiments   -------------> contains Jupyter notebook for Tensorflow based DNN training experiment
β”‚   └── pipelines/training   ------> contains Kubeflow pipeline code for running machine learning training workflow
│── media   -----------------------> contains images

Infrastructure

gcp-topology-infrastructure

Built with

  • Flutter v3
  • Python v3.9 & Flask v2.2
  • Tensorflow v2
  • MongoDB v6
  • Docker
  • Kubernetes & Helm chart
  • Flux CD v0.35
  • ingress-nginx controller v4.2.5
  • cert-manager for letsencrypt v1.9.1
  • Kubeflow standalone pipelines v1.8.5
  • Github workflows

Getting started

To get a local copy up and running, follow these steps.

Prerequisites

# create a virtual environment
# assuming you have "python3 --version" = "Python 3.9.13" installed in the current terminal session
python3 -m venv ./venv

# activate virtual environment
source ./venv/bin/activate

# upgrade pip
python -m pip install --upgrade pip

# install python dependencies
pip install -r ./backend/api/requirements.txt -r ./backend/api/requirements_dev.txt

# lint python code
pylint ./backend/api

Run machine learning experiments

# change directory
cd machine_learning/experiments/tensorflow

# make sure the data file is copied to the 'data' subdirectory
mkdir data
cp <DATA SOURCE DIRECTORY>/data.psv data/

# run the jupyter notebook - tensorflow_dnn.ipynb
jupyter notebook tensorflow_dnn.ipynb

# check the machine learning training output
ls data
>>> causes_tokeniser.json  data.psv  model.h5  symptoms_tokeniser.json

Run backend api locally

Make sure you've configured oauth2 tenant like Auth0 or any other, to enable access tokens for backend api. Use the postman collection in backend/ai-hospital.services.postman_collection.json to run queries after the following steps:

# change directory
cd backend

# copy the machine learning training output to a new data
mkdir api/data
cp ../machine_learning/experiments/tensorflow/data/* api/data/

# argument --debug-mode = true or false (default) to enable debug mode logging
FLASK_DEBUG=1 \
MONGODB_URL="mongodb://localhost:27017/" \
TENANT_DOMAIN="<TENANT DOMAIN>" \
REDIRECT_URL="<REDIRECT URL>" \
CLIENT_ID="<CLIENT ID>" \
CLIENT_SECRET="<CLIENT SECRET>" \
  python -m api.app --debug-mode true --port 8080

# note:
# FLASK_DEBUG = debug mode for flask - 1 (default) or 0
# MONGODB_URL = mongodb connection url - "mongodb://localhost:27017/" (default)
# TENANT_DOMAIN = oauth2 tenant domain
# REDIRECT_URL = oauth2 redirect url
# CLIENT_ID = oauth2 client id
# CLIENT_SECRET = oauth2 client secret

# curl to hit backend api
curl http://localhost:8080
>>> Welcome to backend api!

Kubernetes workloads

kubernetes

Setup prerequisites in google kubernetes engine

Make sure to setup the backend infrastructure on google cloud by following instructions:

Setup a google cloud dns with the public ip address in google cloud created by terraform - <PREFIX>-<ENVIRONMENT>-app-ip01

Install NGINX ingress controller

# change directory
cd .deploy/helm

# prepare the 'nginx-ingress-values-secret.yaml'
touch nginx-ingress-values-secret.yaml

# update the 'nginx-ingress-values-secret.yaml' file like the following, where,
# controller:
#   service:
#     externalTrafficPolicy: "Local"
#     loadBalancerIP: "<LOAD BALANCER IP>"
#   nodeSelector: {"kubernetes.io/arch": "arm64"}
#   tolerations: [{key: "kubernetes.io/arch", value: "arm64"}]
#   admissionWebhooks:
#     patch:
#       nodeSelector: {"kubernetes.io/arch": "arm64"}
#       tolerations: [{key: "kubernetes.io/arch", value: "arm64"}]
# defaultBackend:
#   nodeSelector: {"kubernetes.io/arch": "arm64"}
#   tolerations: [{key: "kubernetes.io/arch", value: "arm64"}]

# to override arm64 node selection and tolerations, replace with,
# nodeSelector: {}
# tolerations: []

# note:
# - <LOAD BALANCER IP> is the public ip address in google cloud created by terraform - <PREFIX>-<ENVIRONMENT>-app-ip01

# install/upgrade helm chart
helm upgrade -i ingress-nginx ingress-nginx \
  --repo https://kubernetes.github.io/ingress-nginx \
  --namespace ingress-nginx \
  --create-namespace \
  --version 4.2.5 \
  -f nginx-ingress-values.yaml \
  -f nginx-ingress-values-secret.yaml

# if you want to stop and remove helm chart and namespace
helm delete -n ingress-nginx ingress-nginx
kubectl delete namespace ingress-nginx

Install cert-manager for letsencrypt ssl certificate generation

# install/upgrade helm chart
helm upgrade -i cert-manager cert-manager \
    --repo "https://charts.jetstack.io" \
    --namespace cert-manager \
    --create-namespace \
    --version v1.9.1 \
    -f cert-manager-values.yaml --dry-run

# if you want to stop and remove helm chart and namespace
helm delete -n cert-manager cert-manager
kubectl delete namespace cert-manager

Install flux cd

# change directory
cd .deplpoy/clusters/gke01

# install flux system
kubectl apply -f flux-system/flux-system.yaml

# create ssh key for flux to authenticate to github
flux create secret git medicine-prescriber-prototype-auth \
  --url=ssh://git@github.com/ai-hospital-services/medicine-prescriber-prototype \
  --ssh-key-algorithm=ecdsa \
  --ssh-ecdsa-curve=p521 \
  --export > medicine-prescriber-prototype-flux-auth-secret.yaml

# install github ssh key as secret
kubectl apply -f medicine-prescriber-prototype-flux-auth-secret.yaml

# add the above generated ssh key in 'medicine-prescriber-prototype-flux-auth-secret.yaml' to your github.com > settings > SSH keys

# create git source in flux
kubectl apply -f medicine-prescriber-prototype-source.yaml

Setup backend api in google kubernetes engine

Make sure you've configured oauth2 tenant like Auth0 or any other, to enable access tokens for backend api. Use the postman collection in backend/ai-hospital.services.postman_collection.json to run queries after the following steps:

# make sure the docker image for backend api has been built with image tag for GCR in Asia, like,  
# asia.gcr.io/<PREFIX>-<ENVIRONMENT>/backend-api:<VERSION>
# matching <VERSION> from the 'appVersion' value in '.deploy/backend-api/Chart.yaml'
docker build -t asia.gcr.io/<PREFIX>-<ENVIRONMENT>/backend-api:<VERSION> backend/api

# push docker image
gcloud auth configure-docker
docker push asia.gcr.io/<PREFIX>-<ENVIRONMENT>/backend-api:<VERSION>

# change directory
cd .deplpoy/clusters/gke01

# create machine learning output secret and 'backend' namespace
mkdir backend/data
cp ../../../machine_learning/experiments/data/* backend/data/
kubectl apply -k backend

# prepare the '../../helm/backend-api/values-secret.yaml'
touch ../../helm/backend-api/values-secret.yaml

# update the '../../helm/backend-api/values-secret.tfvars' file like the following, where,
# image:
#   repository: "asia.gcr.io/<PREFIX>-<ENVIRONMENT>/backend-api"
# config:
#   mongodbURL: "mongodb://<MONGODB APP USERNAME>:<MONGODB APP PASSWORD>@mongodb.mongodb:27017/ai_hospital_services?authMechanism=DEFAULT&authSource=ai_hospital_services"
#   tenantDomain: "<TENANT DOMAIN>"
#   redirectURL: "<REDIRECT URL>"
#   clientID: "<CLIENT ID>"
#   clientSecret: "<CLIENT SECRET>"
#   hosts:
#     - host: <API DOMAIN NAME>
#       paths:
#         - path: /
#           pathType: Prefix
#   tls:
#    - secretName: api-ai-hospital-services-tls
#      hosts:
#        - <API DOMAIN NAME>
# letsencrypt:
#   enabled: true
#   email: "<DOMAIN EMAIL ADDRESS>"
#   mode: "production"

# to override arm64 node selection and tolerations, add,
# nodeSelector: {}
# tolerations: []

# note:
# - <TENANT DOMAIN> = oauth2 tenant domain
# - <REDIRECT_URL> = oauth2 redirect url
# - <CLIENT ID> = oauth2 client id
# - <CLIENT SECRET> = oauth2 client secret
# - <API DOMAIN NAME> = backend api domain name to be configured in google cloud dns
# - <DOMAIN EMAIL ADDRESS> = domain email address for letsencrypt

# create helm release values file as secret
kubectl -n flux-system create secret generic values-backend-api \
  --from-file=values.yaml=../../helm/backend-api/values-secret.yaml

# create helm release for backend api
kubectl apply -f backend/backend-api-release.yaml

# watch the helm release deploy
flux get helmreleases --watch

# if you want to stop and remove helm release and namespace
kubectl delete -f backend/backend-api-release.yaml
kubectl delete -k backend

# configure the google cloud dns by adding an 'A' record for the <API DOMAIN NAME>

Setup frontend app in google kubernetes engine

Make sure to setup the backend api by following the instructions in previous section

# make sure the docker image for frontend app has been built with image tag for GCR in Asia, like,  
# asia.gcr.io/<PREFIX>-<ENVIRONMENT>/frontend-app:<VERSION>
# matching <VERSION> from the 'appVersion' value in '.deploy/frontend-app/Chart.yaml'
docker build -t asia.gcr.io/<PREFIX>-<ENVIRONMENT>/frontend-app:<VERSION> frontend/app

# push docker image
gcloud auth configure-docker
docker push asia.gcr.io/<PREFIX>-<ENVIRONMENT>/frontend-app:<VERSION>

# change directory
cd .deplpoy/clusters/gke01

# create 'froontend' namespace
kubectl apply -k frontend

# prepare the '../../helm/frontend-app/values-secret.yaml'
touch ../../helm/frontend-app/values-secret.yaml

# update the '../../helm/frontend-app/values-secret.yaml' file like the following, where,
# image:
#   repository: "asia.gcr.io/<PREFIX>-<ENVIRONMENT>/frontend-app"
# config:
#   backendAPIURL: "<BACKEND END API URL>"
#   authoriseURL: "<TENANT AUTHORIZE ENDPOINT URL>"
#   redirectURL: "<REDIRECT URL>"
#   audience: "<TENANT AUDIENCE API>"
#   clientID: "<CLIENT ID>"
#   hosts:
#     - host: <APP DOMAIN NAME>
#       paths:
#         - path: /
#           pathType: Prefix
#   tls:
#    - secretName: app-ai-hospital-services-tls
#      hosts:
#        - <APP DOMAIN NAME>
# letsencrypt:
#   enabled: true
#   email: "<DOMAIN EMAIL ADDRESS>"
#   mode: "production"

# to override arm64 node selection and tolerations, add,
# nodeSelector: {}
# tolerations: []

# note:
# - <BACKEND END API URL> = backend api url configured in the google cloud dns
# - <TENANT AUTHORIZE ENDPOINT URL> = oauth2 tenant authorize endpoint url
# - <REDIRECT_URL> = oauth2 redirect url
# - <CLIENT ID> = oauth2 client id
# - <CLIENT SECRET> = oauth2 client secret
# - <APP DOMAIN NAME> = frontend app domain name to be configured in google cloud dns
# - <DOMAIN EMAIL ADDRESS> = domain email address for letsencrypt

# create helm release values file as secret
kubectl -n flux-system create secret generic values-frontend-app \
  --from-file=values.yaml=../../helm/frontend-app/values-secret.yaml

# create helm release for backend api
kubectl apply -f frontend/frontend-app-release.yaml

# watch the helm release deploy
flux get helmreleases --watch

# if you want to stop and remove helm release and namespace
kubectl delete -f frontend/frontend-app-release.yaml
kubectl delete -k frontend

# configure the google cloud dns by adding an 'A' record for the <APP DOMAIN NAME>

Setup kubeflow machine learning pipeline

kubeflow-pipeline-execution

Reference:

Make sure the data file is uploaded to the google cloud storage bucket - <PREFIX>-<ENVIRONMENT>-storagebucket01 as data.psv.

# enable prerequisite google cloud services
gcloud services enable \
  serviceusage.googleapis.com \
  compute.googleapis.com \
  container.googleapis.com \
  iam.googleapis.com \
  servicemanagement.googleapis.com \
  cloudresourcemanager.googleapis.com \
  ml.googleapis.com \
  iap.googleapis.com \
  sqladmin.googleapis.com \
  meshconfig.googleapis.com \
  krmapihosting.googleapis.com \
  servicecontrol.googleapis.com \
  endpoints.googleapis.com \
  iamcredentials.googleapis.com

# install kubeflow standalone pipeline runtime
export PIPELINE_VERSION=1.8.5
kubectl apply -k "github.com/kubeflow/pipelines/manifests/kustomize/cluster-scoped-resources?ref=$PIPELINE_VERSION"
kubectl wait --for condition=established --timeout=60s crd/applications.app.k8s.io
kubectl apply -k "github.com/kubeflow/pipelines/manifests/kustomize/env/platform-agnostic-pns?ref=$PIPELINE_VERSION"

# build machine learning dependencies docker image
cd machine_learning/pipelines/training
docker build -t asia.gcr.io/<PREFIX>-<ENVIRONMENT>/machine_learning_training:tensorflow_dnn .

# push docker image
gcloud auth configure-docker
docker push asia.gcr.io/<PREFIX>-<ENVIRONMENT>/machine_learning_training:tensorflow_dnn

# install kubeflow pip dependencies
pip install kfp --upgrade

# apply role binding for kubeflow 'pipeline-runner' service account to create secrets in 'backend' namespace
kubectl apply secretsrolebinding.yaml

# annotate kubernetes service account to use google service account (already provisioned by terraform) that has iam role to read/write to google cloud storage bucket
PROJECT_ID="<PREFIX>-<ENVIRONMENT>"
NAMESPACE="kubeflow"
GSA="gke01-kfp-user@${PROJECT_ID}.iam.gserviceaccount.com"
KSA="pipeline-runner"
kubectl annotate serviceaccount \
  --namespace $NAMESPACE \
  --overwrite \
  $KSA \
  iam.gke.io/gcp-service-account=$GSA

# run the above for the following combination as well
GSA="gke01-kfp-system@${PROJECT_ID}.iam.gserviceaccount.com"
KSA="ml-pipeline-ui"

GSA="gke01-kfp-system@${PROJECT_ID}.iam.gserviceaccount.com"
KSA="ml-pipeline-visualizationserver"

# port forward from kubeflow pipeline ui service
kubectl port-forward --namespace kubeflow svc/ml-pipeline-ui 3000:80

# run the training pipeline - 'Machine learning training pipeline - tensorflow dnn'
GCS_STORAGE_BUCKET_NAME="<PREFIX>-<ENVIRONMENT>-storagebucket01" \
DATA_FILE_NAME="data.psv" \
SECRET_NAME="test" \
SECRET_NAMESPACE="backend" \
  python tensorflow_dnn.py

# check the pipeline execution on browser
http://localhost:3000

# check the training output files are available in the google cloud storage bucket - '<PREFIX>-<ENVIRONMENT>-storagebucket01'
# check the secret - 'test' has been created in the 'backend' namespace

# if you want to stop and remove helm release and namespace
export PIPELINE_VERSION=1.8.5
kubectl delete -k "github.com/kubeflow/pipelines/manifests/kustomize/env/platform-agnostic-pns?ref=$PIPELINE_VERSION"
kubectl delete -k "github.com/kubeflow/pipelines/manifests/kustomize/cluster-scoped-resources?ref=$PIPELINE_VERSION"

Run tests

# change directory
cd backend

# run unit tests
pytest -v --cov=api

Authors

πŸ‘€ Ankur Soni

  • Github
  • LinkedIn
  • Twitter

πŸ“ License

This project is Apache licensed.