Kubernetes

Edit page

Because every Xapiand node keeps its own data on disk and discovers its peers to form a cluster, it maps naturally onto a Kubernetes StatefulSet: each replica gets a stable network identity and its own persistent volume, and a headless Service gives the pods stable DNS names to find each other.

A node listens on three TCP ports, all of which the pods need to reach on each other:

Port	Protocol	Used for
`8880`	HTTP REST API	client requests
`9880`	Remote protocol	distributed searches across shards
`7880`	Replication protocol	keeping replicas in sync

A starting manifest

A three-node cluster with per-node storage. Each pod takes its node name from its own pod name and advertises its own pod IP for discovery:

apiVersion: v1
kind: Service
metadata:
  name: xapiand
spec:
  clusterIP: None          # headless: one stable DNS name per pod
  selector:
    app: xapiand
  ports:
    - { name: http,        port: 8880 }
    - { name: remote,      port: 9880 }
    - { name: replication, port: 7880 }
---
apiVersion: apps/v1
kind: StatefulSet
metadata:
  name: xapiand
spec:
  serviceName: xapiand
  replicas: 3
  selector:
    matchLabels:
      app: xapiand
  template:
    metadata:
      labels:
        app: xapiand
    spec:
      containers:
        - name: xapiand
          image: ghcr.io/kronuz/xapiand:latest
          args:
            - "--cluster=xapiand"
            - "--name=$(POD_NAME)"
            - "--discovery-interface=$(POD_IP)"
            - "--database=/data"
          env:
            - name: POD_NAME
              valueFrom: { fieldRef: { fieldPath: metadata.name } }
            - name: POD_IP
              valueFrom: { fieldRef: { fieldPath: status.podIP } }
          ports:
            - { name: http,        containerPort: 8880 }
            - { name: remote,      containerPort: 9880 }
            - { name: replication, containerPort: 7880 }
          volumeMounts:
            - { name: data, mountPath: /data }
  volumeClaimTemplates:
    - metadata:
        name: data
      spec:
        accessModes: ["ReadWriteOnce"]
        resources:
          requests:
            storage: 10Gi

The --cluster name must be the same on every node, --name must be unique (the pod name satisfies this), and --database points at the mounted volume so each node’s data survives a restart or reschedule.