Kubeblocks Addon Milvus

Deploy Milvus on KubeBlocks

Overview

Deploy Milvus vector database clusters on Kubernetes using KubeBlocks. Milvus powers embedding similarity search and AI applications. Supports standalone mode (single node, dev/test) and cluster mode (distributed, production) with etcd, MinIO, and optional Pulsar/Kafka for log storage.

Official docs: https://kubeblocks.io/docs/preview/kubeblocks-for-milvus/01-overview Quickstart: https://kubeblocks.io/docs/preview/kubeblocks-for-milvus/02-quickstart

Prerequisites

• A running Kubernetes cluster with KubeBlocks installed (see install-kubeblocks)
• The Milvus addon must be enabled:

# Check if milvus addon is installed
helm list -n kb-system | grep milvus

# Install if missing
helm install kb-addon-milvus kubeblocks/milvus --namespace kb-system --version 1.0.0

Available Topologies

Default ports: 19530 (gRPC), 9091 (metrics).

Supported Versions

List available: kubectl get cmpv milvus

Workflow

- [ ] Step 1: Ensure addon is installed
- [ ] Step 2: Create namespace
- [ ] Step 3: Create cluster (choose topology)
- [ ] Step 4: Dry-run, then apply
- [ ] Step 5: Watch until Running, verify success
- [ ] Step 6: Connect and verify

Step 1: Ensure Addon Is Installed

helm list -n kb-system | grep milvus

If not found:

helm install kb-addon-milvus kubeblocks/milvus --namespace kb-system --version 1.0.0

Step 2: Create Namespace

kubectl create namespace demo --dry-run=client -o yaml | kubectl apply -f -

Step 3: Create Cluster

Standalone (Dev/Test)

Single-node deployment with etcd and MinIO co-located. Suitable for development and testing:

apiVersion: apps.kubeblocks.io/v1
kind: Cluster
metadata:
  name: milvus-standalone
  namespace: demo
spec:
  clusterDef: milvus
  topology: standalone
  terminationPolicy: Delete
  componentSpecs:
    - name: etcd
      replicas: 1
      resources:
        limits: {cpu: "0.5", memory: "0.5Gi"}
        requests: {cpu: "0.5", memory: "0.5Gi"}
      volumeClaimTemplates:
        - name: data
          spec:
            accessModes: [ReadWriteOnce]
            resources: {requests: {storage: 20Gi}}
    - name: minio
      replicas: 1
      resources:
        limits: {cpu: "0.5", memory: "0.5Gi"}
        requests: {cpu: "0.5", memory: "0.5Gi"}
      volumeClaimTemplates:
        - name: data
          spec:
            accessModes: [ReadWriteOnce]
            resources: {requests: {storage: 20Gi}}
    - name: milvus
      replicas: 1
      resources:
        limits: {cpu: "0.5", memory: "0.5Gi"}
        requests: {cpu: "0.5", memory: "0.5Gi"}
      volumeClaimTemplates:
        - name: data
          spec:
            accessModes: [ReadWriteOnce]
            resources: {requests: {storage: 20Gi}}

Cluster (Production)

Distributed deployment with proxy, mixcoord, datanode, indexnode, querynode. Requires pre-created etcd, minio, and optionally Pulsar/Kafka clusters. See cluster topology docs for full setup.

Key components: milvus-proxy (access), milvus-mixcoord (coordination), milvus-datanode, milvus-indexnode, milvus-querynode (compute), plus etcd, minio. Cluster mode uses serviceRef to reference external etcd, minio, and pulsar/kafka clusters.

Step 4: Dry-Run and Apply

Before applying, run server-side dry-run to catch RBAC, webhook, or schema errors:

kubectl apply -f cluster.yaml --dry-run=server

If dry-run succeeds, apply:

kubectl apply -f cluster.yaml

Step 5: Wait for Cluster Ready

kubectl -n demo get cluster <cluster-name> -w

Success condition: STATUS shows Running. Typical duration: 2–5 min. Investigate after 10 min if still Creating/Updating (see safety-patterns).

Check pods:

kubectl -n demo get pods -l app.kubernetes.io/instance=<cluster-name>

Step 6: Connect

Port-Forward (Standalone)

kubectl port-forward pod/<cluster-name>-milvus-0 -n demo 19530:19530

Connect via localhost:19530 (gRPC). For cluster mode, port-forward to the proxy pod.

Verify with Python

from pymilvus import connections
connections.connect(host="localhost", port=19530)

Troubleshooting

Cluster stuck in Creating:

kubectl -n demo describe cluster <cluster-name>
kubectl -n demo get events --sort-by='.lastTimestamp'

Milvus pod not starting:

kubectl -n demo logs <milvus-pod>

etcd or minio dependency issues:

• Ensure etcd and minio pods are Running before milvus starts
• Check component order in componentSpecs

Out of memory:

• Increase resources.limits.memory for milvus component
• For cluster mode, scale datanode/indexnode/querynode resources

Day-2 Operations

Safety Patterns

Follow safety-patterns: dry-run before apply, confirm success condition after watch, pre-deletion checklist before delete.