Nominatim on kubernetes?

[tamis-laan]

So I want to deploy Nominatim on a Kubernetes cluster. I created the following yaml spec to do this:

apiVersion: apps/v1
kind: Deployment
metadata:
  name: nominatim
spec:
  replicas: 1
  selector:
    matchLabels:
      app: nominatim
  template:
    metadata:
      labels:
        app: nominatim
    spec:
      containers:
        - name: nominatim
          image: mediagis/nominatim:4.2
          ports:
            - containerPort: 8080
          env:
            - name: PBF_URL
              value: https://download.geofabrik.de/europe/netherlands-latest.osm.pbf
            - name: REPLICATION_URL
              value: https://download.geofabrik.de/europe/netherlands-updates/

This works, it creates a replica that starts downloading the pbf file and then starts populating the postgresdb. But this is far from optimal because the pbf file downloaded and the db is populated everytime the cluster is restarted or a new replica is created.

What I would like to do:

1. Create a postgresql deployment (or use the operator) to create a database cluster with read replica's.
2. Have a Nominatim deployment with a side car init job that downloads the data to a persistent volume.
3. Have that same side car start the indexing on the postgress cluster.
4. Being able to scale up the number of Nominatim replicas for dealing with higher loads.

Has anyone done this before? Is there something out there that does this?

[leonardehrenfried]

Some people deploy to kubernetes but what you're planning sounds like the most advanced version i've heard yet.

We would love to add it to the documentation when you've completed it.

[tamis-laan]

@leonardehrenfried

Maybe what I'm trying to do is overkill, but I think it can be done:

1. This is a postgresql operator that scales https://github.com/zalando/postgres-operator.
2. We can create a deployment with a single replica called the "updater" that sets PBF_URL and REPLICATION_URL simply there to download, seed and keep the database up to date. It also needs to set NOMINATIM_DATABASE_DSN, PGHOSTADDR, PGDATABASE, PGUSER and PGPASSWORD for connecting to the postresql database.
3. Then we create a second deployment that is scalable, that only references the postgresql database not responsible for downloading and updating with variables NOMINATIM_DATABASE_DSN, PGHOSTADDR, PGDATABASE, PGUSER and PGPASSWORD.

[winsento]

@dlucia i think it's a cool idea. When I have more time I will write a spec.

[tamis-laan]

I made a simple statefullSet that has a init container for indexing and then starts serving. This way you can set different resource limits for the initial process. Would be nice if there was a init.sh and server.sh where the init.sh takes care of building the database and the start.sh would start serving. That would make things a lot cleaner.

apiVersion: v1
kind: ConfigMap
metadata:
  name: nominatim-config
data:
  PBF_URL: "https://download.geofabrik.de/europe/monaco-latest.osm.pbf"
  REPLICATION_URL: "https://download.geofabrik.de/europe/monaco-updates/"

---
apiVersion: apps/v1
kind: StatefulSet
metadata:
  name: nominatim
spec:
  serviceName: "nominatim"
  replicas: 2
  selector:
    matchLabels:
      app: nominatim
  template:
    metadata:
      labels:
        app: nominatim
    spec:
      initContainers:
        - name: nominatim-init
          image: mediagis/nominatim:4.2
          envFrom:
            - configMapRef:
                name: nominatim-config
          volumeMounts:
            - name: nominatim-db
              mountPath: /var/lib/postgresql/14/main
            - name: nominatim-other
              mountPath: /nominatim
          command:
            - "/bin/bash"
            - "-ex"
            - "-c"
          args:
            - |
              tailpid=0
              replicationpid=0

              stopServices() {
                service apache2 stop
                service postgresql stop
                kill $replicationpid
                kill $tailpid
              }
              trap stopServices SIGTERM TERM INT

              /app/config.sh

              if id nominatim >/dev/null 2>&1; then
                echo "user nominatim already exists"
              else
                useradd -m -p ${NOMINATIM_PASSWORD} nominatim
              fi

              IMPORT_FINISHED=/var/lib/postgresql/14/main/import-finished

              if [ ! -f ${IMPORT_FINISHED} ]; then
                /app/init.sh
                touch ${IMPORT_FINISHED}
              else
                chown -R nominatim:nominatim ${PROJECT_DIR}
              fi
      containers:
        - name: nominatim
          image: mediagis/nominatim:4.2
          ports:
            - containerPort: 8080
          envFrom:
            - configMapRef:
                name: nominatim-config
          volumeMounts:
            - name: nominatim-db
              mountPath: /var/lib/postgresql/14/main
            - name: nominatim-other
              mountPath: /nominatim
          readinessProbe:
            httpGet:
              path: /status
              port: 8080
            initialDelaySeconds: 30
            periodSeconds: 10
          livenessProbe:
            httpGet:
              path: /status
              port: 8080
            initialDelaySeconds: 10
            periodSeconds: 15
          resources:
            requests:
              cpu: 400m
              memory: 750M
            limits:
              memory: 2000M
  volumeClaimTemplates:
  - metadata:
      name: nominatim-db
    spec:
      accessModes: [ "ReadWriteOnce" ]
      resources:
        requests:
          storage: 10Gi
  - metadata:
      name: nominatim-other
    spec:
      accessModes: [ "ReadWriteOnce" ]
      resources:
        requests:
          storage: 10Gi
---
apiVersion: v1
kind: Service
metadata:
  name: nominatim
spec:
  type: ClusterIP
  selector:
    app: nominatim
  ports:
    - protocol: TCP
      port: 8080
      targetPort: 8080

It would also be nice if all files including database would line in one location such as /nominatim, this way we only need to use a single persistentVolumeClaim.

[mausch]

I recommend checking out https://github.com/robjuz/helm-charts/tree/99a5d1e47150ef9f57066116ebbeea1872c0cdca/charts/nominatim