SuryaRakshak

Smart heatwave surveillance and advisory system for Indian regions.

Project Structure

• /client - React frontend (Vite)
• /server - Express API backend (Open-Meteo & NASA POWER integration)

Data sources

Everything shown as a measurement comes from a live public API (none needs a key). There are no machine-learning models: risk is computed with the NWS heat-index equations (Rothfusz regression, Steadman formula below 26.7°C) and fixed risk bands.

Source	Used for
Open-Meteo Forecast API	Current conditions, hourly and 7-day forecast, UV, tamper-check baseline
Open-Meteo Historical Weather API (ERA5)	30-day anomaly chart and the 10-year heat-exposure ranking
NASA POWER	Latest available daily maximum temperature (lags real time by a few days)
Nominatim / OpenStreetMap	Geocoding when adding a region

If a source is unreachable the API reports it (live: false, or an HTTP 502/503) rather than substituting numbers.

Not live data: the health advisories are general guidance, the cooling centers are labelled sample placeholders until real ones are registered via POST /api/resources, the AWS stations are a simulation for demonstrating tamper detection (they hold no readings until telemetry is posted), and the logins are demo credentials.

Render Deployment Settings (Backend)

When creating a Web Service on Render.com:

• Environment: Node
• Root Directory: server
• Build Command: npm install
• Start Command: npm start (or node src/index.js)

Or simply use the included Blueprint by selecting New -> Blueprint and connecting your repository (render.yaml will configure everything automatically).

Vercel Deployment Settings (Frontend)

1. Deploy the /client directory on Vercel or point the Vercel Root Directory to client.
2. Add Environment Variable in Vercel settings:

   VITE_API_BASE_URL=https://<your-render-backend-name>.onrender.com
   

Microservice Architecture (Docker Desktop + Minikube)

The Express monolith in /server is also split into independent services under /services (the monolith is kept for the Render/Vercel deployments above).

Service	Port	Owns	Routes
frontend (nginx)	80	React build + API gateway	/ and routing of /api/*
weather-service	3001	regions, forecast, climate history	/api/dashboard, /api/regions, /api/forecast, /api/analytics, /api/exposure, /api/health
alert-service	3002	alerts	/api/alerts
advisory-service	3003	advisories, cooling centers	/api/health-advisories, /api/resources
aws-service	3004	AWS stations, security logs	/api/aws/*
auth-service	3005	demo login	/api/auth/*
mongo	27017	one database per service (suryarakshak_<service>)

weather-service (dashboard) and aws-service (tamper alerts) call alert-service over HTTP; if it is down they degrade gracefully. Every service exposes /healthz (liveness) and /readyz (readiness, waits for MongoDB).

Docker Desktop (Compose)

docker compose up --build        # app at http://localhost:8080
docker compose down              # add -v to also delete the Mongo volume

Each backend service is also published on its own host port (3001-3005) for direct testing.

Kubernetes on Docker Desktop

Enable Kubernetes in Docker Desktop (Settings → Kubernetes, kubeadm), then:

./scripts/docker-desktop-deploy.sh                               # builds images, applies k8s/, waits for rollout
kubectl -n suryarakshak get deploy,pods                          # one Deployment per microservice
kubectl -n suryarakshak scale deploy --all --replicas=0          # stop everything (kubectl apply -k k8s restores it)
kubectl delete namespace suryarakshak                            # tear down

The app is served at http://localhost:30080. The cluster runs on Docker Desktop's own engine, so locally built images are used directly without a registry.

Optional: the Kubernetes Dashboard shows per-deployment status charts:

kubectl apply -f https://raw.githubusercontent.com/kubernetes/dashboard/v2.7.0/aio/deploy/recommended.yaml
kubectl -n kubernetes-dashboard create serviceaccount admin-user
kubectl create clusterrolebinding admin-user --clusterrole=view --serviceaccount=kubernetes-dashboard:admin-user
kubectl -n kubernetes-dashboard create token admin-user          # login token
kubectl -n kubernetes-dashboard port-forward svc/kubernetes-dashboard 8443:443

Open https://localhost:8443, sign in with the token, choose the suryarakshak namespace and open Workloads.

Kubernetes on kind (multi-node)

./scripts/kind-deploy.sh         # creates a 4-node kind cluster, loads the images, deploys; app at http://localhost:8081

Kubernetes on Minikube

./scripts/minikube-deploy.sh     # starts Minikube (docker driver), builds images in its daemon, applies k8s/, prints the URL
kubectl -n suryarakshak get pods
minikube service frontend -n suryarakshak          # open in browser
kubectl delete namespace suryarakshak               # tear down

Images are built directly into Minikube's Docker daemon (eval $(minikube docker-env)), so no registry is needed. Each service runs 2 replicas behind a ClusterIP Service; the frontend is exposed as a NodePort. Re-run the script after code changes to rebuild and roll out.

Monitoring (Prometheus + Grafana)

Every service exposes Prometheus metrics at /metrics (via prom-client in services/common):

Metric	Meaning
http_requests_total{method,route,status}	Requests handled, by route pattern and status code
http_request_duration_seconds	Request latency histogram
service_ready	1 when the service and its database are ready
service_calls_total{target,outcome}	Calls to other microservices, success or failure
process_*, nodejs_*	CPU, memory, event-loop lag and other Node.js defaults

Prometheus discovers service pods through their prometheus.io/scrape and prometheus.io/port annotations (it only has read access to pods in the suryarakshak namespace). Grafana is provisioned with the Prometheus data source and a SuryaRakshak Microservices dashboard, viewable without signing in.

	Kubernetes	Compose
Grafana	http://localhost:30030	http://localhost:3000
Prometheus	http://localhost:30090	http://localhost:9090

Config lives in k8s/monitoring/ (shared by both setups) and docker/prometheus.yml (Compose scrape targets). Example queries:

sum by (service) (up{job="suryarakshak-services"})
sum by (service) (rate(http_requests_total[5m]))
histogram_quantile(0.95, sum by (service, le) (rate(http_request_duration_seconds_bucket[5m])))
sum by (service, target, outcome) (service_calls_total)

Running a single service locally (no Docker, in-memory data)

cd services/common && npm install
cd ../alert-service && PORT=3002 npm start