​

Enable RDF (Knowledge Graph)

1. Deploy Apache Jena Fuseki into your cluster.
2. Configure the Collate Helm release to point at the Fuseki endpoint.

​

Prerequisites

• A running Collate BYOC deployment installed with the open-metadata/openmetadata Helm chart. See EKS, AKS, GKE, or On-Prem.
• OpenSearch as the search engine—already required for Collate BYOC.
• kubectl and helm access to the namespace where Collate is deployed (the examples below use collate).

​

Step 1: Deploy Apache Jena Fuseki

1. Create the Fuseki admin credentials. Fuseki needs an admin password, and the Collate server uses the same credentials to connect to it. Create a Kubernetes secret holding the password—the Fuseki chart reads fuseki-admin-credentials / admin-password by default:
   kubectl create secret generic fuseki-admin-credentials \
     --from-literal=admin-password=<STRONG_PASSWORD> \
     --namespace collate
   
2. Deploy Fuseki from the dependencies chart. Add the Helm repository (skip if you already added it for the Collate install):
   helm repo add open-metadata https://helm.open-metadata.org/
   helm repo update
   
   Create a values file that enables only Fuseki. The fuseki sizing and persistence values shown are the chart defaults (the production BYOC sizing). On charts before 2.0.3, the image block overrides the stock chart default; on 2.0.3 and later it pins the version the chart already uses. The minimum required is fuseki.enabled: true, disabling the other dependencies, and openmetadata-dependencies 2.0.3 or later. Tune the rest for your catalog size:
   # fuseki.values.yml — deploy only Fuseki from the openmetadata-dependencies chart
   mysql:
     enabled: false
   opensearch:
     enabled: false
   airflow:
     enabled: false
   
   fuseki:
     enabled: true
     # Explicit pin of the OpenMetadata Fuseki image (already the chart default
     # on openmetadata-dependencies 2.0.3+). See the notes below.
     image:
       repository: openmetadata/fuseki
       tag: "6.2.0"
     # Persist the graph across pod restarts (enabled by default).
     persistence:
       enabled: true
       size: 100Gi
       storageClass: gp3   # set to a StorageClass in your cluster; leave "" for the cluster default
     jvmArgs: "-Xmx8g -Xms8g"
     resources:
       requests:
         cpu: "1500m"
         memory: "10Gi"
       limits:
         cpu: "2500m"
         memory: "12Gi"
   
   On openmetadata-dependencies 2.0.3 and later, the chart defaults to the openmetadata/fuseki:6.2.0 image. It sets FUSEKI_DATASETS to create the datasets, sets FUSEKI_UNION_DEFAULT_GRAPH=true and OPENMETADATA_EXTENSION_ENABLED=true, and runs the pod with fsGroup: 1000. SPARQL queries then read across all named graphs instead of only the default graph. The image block above is an explicit version pin, not a requirement. Charts before 2.0.3 default to stock Apache Jena Fuseki (daschswiss/apache-jena-fuseki) and don’t inject this environment, so RDF indexing finds no openmetadata dataset and the pod can crash-loop. Upgrade the chart rather than only overriding the image. Install it as a dedicated release in the same namespace as Collate:
   helm upgrade --install fuseki open-metadata/openmetadata-dependencies \
     --values fuseki.values.yml \
     --namespace collate
   
   This deploys a fuseki Deployment, a ClusterIP Service, and a PersistentVolumeClaim. The Collate server reaches Fuseki in-cluster at http://fuseki:3030 (or http://fuseki.collate.svc.cluster.local:3030 from other namespaces).
   Note: You don’t need to create the Fuseki datasets manually. On openmetadata-dependencies 2.0.3 and later, the chart sets FUSEKI_DATASETS to openmetadata,openmetadata_a,openmetadata_b and FUSEKI_UNION_DEFAULT_GRAPH=true, so the image creates all three datasets with the union default graph enabled on first start. The _a and _b datasets let RDF indexing rebuild the graph in one dataset while the other stays live. Don’t create datasets by hand through the Fuseki admin API: datasets created that way have unionDefaultGraph off and fail the readiness check.
   Important: Size Fuseki for your catalog. Fuseki holds the whole graph in memory and can be OOMKilled during the initial RDF indexing of a large catalog. Keep jvmArgs (-Xmx) below the container memory limit and raise both together for larger deployments. The defaults above (8 GB heap / 12 GiB limit, 100 GiB storage) follow the production BYOC recommendation; reduce them for smaller catalogs.

​

Step 2: Configure Collate to Use Fuseki

openmetadata:
  config:
    rdf:
      enabled: true
      storageType: "FUSEKI"
      remoteEndpoint: "http://fuseki:3030/openmetadata"
      username: "admin"
      password:
        secretRef: fuseki-admin-credentials
        secretKey: admin-password
      dataset: "openmetadata"
      baseUri: "https://open-metadata.org/"

helm upgrade --install openmetadata open-metadata/openmetadata \
  --values openmetadata.values.yml \
  --namespace collate

kubectl logs -n collate deploy/openmetadata | grep -i rdf

​

Step 3: Populate the Knowledge Graph

1. In Collate, click the Profile icon and go to Settings > Applications.
2. Open the RDF Indexing app.
3. Run the app.

​

Troubleshooting

• The Knowledge Graph tab doesn’t appear on data assets RDF is enabled, but the graph hasn’t been populated yet. Go to Settings > Applications, open the RDF Indexing app, and run it. The tab appears once indexing completes.
• The RDF Indexing app fails or the Collate server shows a connection error This usually means the Collate server can’t connect to Fuseki, or the credentials are wrong. Check the following:
  • Confirm the Fuseki pod is running:
    kubectl get pods -n collate -l app=fuseki
    
    The pod status should be Running. If it’s Pending or CrashLoopBackOff, check its logs:
    kubectl logs -n collate deploy/fuseki
    
  • Confirm the remoteEndpoint in your Helm values points to the correct in-cluster address (http://fuseki:3030/<dataset-name>) and that the dataset name in the URL matches the dataset field.
  • Confirm the username and password values match the secret you created in Step 1: Deploy Apache Jena Fuseki.
• RDF indexing fails with a 405 or a read-only Graph Store error Either the dataset is served read-only, or Fuseki is a stock Apache Jena Fuseki image. On openmetadata-dependencies 2.0.3 and later the openmetadata/fuseki:6.2.0 image is the default and the chart sets FUSEKI_DATASETS to create the datasets, so a 405 or read-only store usually means Fuseki serves the dataset read-only. If you’re on an older chart, upgrade to 2.0.3 or later, then re-run the RDF Indexing app. To inspect the dataset, port-forward the Fuseki service and send an OPTIONS request to its data endpoint. The Collate server image has no curl, so run curl from your own machine:
  1. Port-forward the Fuseki service:
     kubectl port-forward -n collate svc/fuseki 3030:3030
     
  2. From your machine, send the OPTIONS request:
     curl -is -X OPTIONS -u "admin:<password>" "http://localhost:3030/openmetadata/data"
     
  Read the response headers. The X-OpenMetadata-* headers mean the extension is active. An Allow header without POST means the Graph Store is read-only. Stock Fuseki also returns an Allow header. Only the missing X-OpenMetadata-* headers identify a stock image.
• Indexing starts but stops partway through, and the Fuseki pod restarts Fuseki ran out of memory during indexing. This happens when the catalog is large. Raise both fuseki.jvmArgs (-Xmx) and fuseki.resources.limits.memory together in fuseki.values.yml, upgrade the Fuseki release, then re-run the RDF Indexing app:
  helm upgrade --install fuseki open-metadata/openmetadata-dependencies \
    --values fuseki.values.yml \
    --namespace collate
  
  The chart defaults (8 GB heap / 12 GiB limit) follow the production BYOC recommendation—larger catalogs may need more.