Kubernetes Deployment (k3s)

Deploy the OpenAIRE-AI-Research-Evaluator with Helm on k3s.

Prerequisites

k3s cluster running
helm installed
Access to the Docker image ethicalabs/openaire-research-evaluator:latest
The Echo-DSRN model (public on HuggingFace, baked into the image at build)

1. Install

helm install evaluator ./helm/openaire-research-evaluator \
  --set secrets.hfToken=hf_... \
  --set secrets.secretKey=$(openssl rand -hex 32) \
  --set ingress.host=openaire-eval.local \
  --set config.llmBaseUrl=http://<llm-server-host>:<llm-server-port>/v1

Or create a values.local.yaml:

secrets:
  hfToken: hf_...
  secretKey: abc123...
  hfClientId: ...
  hfClientSecret: ...
ingress:
  host: openaire-eval.local
config:
  llmBaseUrl: http://<llm-server-host>:<llm-server-port>/v1
  # Active collab annotation round (stamped on new papers/annotations).
  modelVersion: "v0.1.4"
  # Set true to close the hub (browse-only). Default false.
  hubReadOnly: false

helm install evaluator ./helm/openaire-research-evaluator -f values.local.yaml

The web container runs alembic upgrade head on startup — existing rows are backfilled to the legacy round (v0.1.3) and stay browsable at /collab/archive/v0.1.3/.

2. Access

kubectl port-forward svc/evaluator-openaire-research-evaluator-web 7860:7860

App at http://localhost:7860. With Ingress configured, at http://openaire-eval.local.

Local DNS setup: add to /etc/hosts:

echo "127.0.0.1 openaire-eval.local" | sudo tee -a /etc/hosts

Then access at http://openaire-eval.local (k3s Traefik listens on port 80).

Components

Component	Service	Port
Web (FastAPI + Vue)	`<release>-web`	7860
PostgreSQL	`<release>-postgres`	5432
Redis	`<release>-redis`	6379
Worker (Celery)	— (internal)	—

Troubleshooting

Check web pod logs for alembic migration status:

kubectl logs deployment/evaluator-openaire-research-evaluator-web