Skip to main content

Multi-node setup

Recall that we have deployed Weaviate on a single node in our Kubernetes cluster with Minikube. Now, let's scale it up to a multi-node setup.

To do this, we will need a Kubernetes cluster with multiple nodes. Then, we can configure Weaviate to make use of these nodes.

Add nodes to your Weaviate cluster

We'll stop the current Weaviate deployment and then deploy a new one with multiple nodes with Minikube.

Minikube limitations

Keep in mind that this runs multiple containers on the same host device for learning. In a production environment, you would typically have a managed Kubernetes cluster with multiple, isolated, physical or virtual nodes.

Stop the current Weaviate deployment

First, stop the tunnel by pressing Ctrl+C at the terminal where you ran minikube tunnel.

Then, stop Minikube:

minikube stop

Since the minikube cluster exists, we will have to delete it before we can start a new one with multiple nodes:


You can also add nodes to an existing Minikube cluster. But we'll delete the current one and start a new one for simplicity.

minikube delete

Start a multi-node Kubernetes cluster

To start a multi-node Kubernetes cluster with Minikube, run:

minikube start --nodes <n>

Replace <n> with the number of nodes you want in your cluster. In our case, let's spin up a 6-node cluster:

minikube start --nodes 6

Once that's finished, you can check the status of your nodes by running:

kubectl get nodes

And you should see output like:

NAME           STATUS   ROLES           AGE   VERSION
minikube Ready control-plane 78s v1.28.3
minikube-m02 Ready <none> 60s v1.28.3
minikube-m03 Ready <none> 50s v1.28.3
minikube-m04 Ready <none> 39s v1.28.3
minikube-m05 Ready <none> 29s v1.28.3
minikube-m06 Ready <none> 18s v1.28.3

Now let's update the Weaviate deployment to use these nodes. To do this, we'll update the replicas field in the values.yaml file to match the number of nodes in the cluster. Along with our reduced resource requests and limits, this section should look as follows:

replicas: 6
type: RollingUpdate
resources: {}
cpu: '300m'
memory: '150Mi'
cpu: '500m'
memory: '300Mi'

If we restart weaviate by running:

helm upgrade --install \
"weaviate" \
weaviate/weaviate \
--namespace "weaviate" \
--values ./values.yaml

Then, you should see Weaviate pods come up on each of the nodes in your cluster. You can check this by running:

kubectl get pods -n weaviate

And you should see output like:

NAME         READY   STATUS            RESTARTS   AGE
weaviate-0 1/1 Running 0 31s
weaviate-1 0/1 PodInitializing 0 9s

As pods are created one by one.

Eventually, you will see something like:

weaviate-0 1/1 Running 0 113s
weaviate-1 1/1 Running 0 91s
weaviate-2 1/1 Running 0 79s
weaviate-3 1/1 Running 0 57s
weaviate-4 1/1 Running 0 35s
weaviate-5 1/1 Running 0 13s

Open a new terminal and run minikube tunnel to expose the Weaviate service to your local machine as before.

Utilizing a multi-node setup

There are two main ways to utilize a multi-node setup, namely replication and horizontal scaling. Let's take a look at each of these.


Replication is the process of creating multiple copies of the same data across multiple nodes.

This is useful for ensuring data availability and fault tolerance. In a multi-node setup, you can replicate your Weaviate data across multiple nodes to ensure that if one node fails, the data is still available on other nodes. A replicated setup can also help distribute the load across multiple nodes to improve performance, and allow zero-downtime upgrades and maintenance.

Weaviate can handle replication by setting a replication factor in the database collection definition. This tells Weaviate how many copies of the data to keep across the nodes in the cluster.

The code examples show how to configure replication. Keep in mind that the specified port may be different for you (e.g. 80) than what is shown in the code snippet.

import weaviate
from weaviate.classes.config import Property, DataType, Configure

client = weaviate.connect_to_local(port=8180, grpc_port=50151)

articles = client.collections.create(
Property(name="title", data_type=DataType.TEXT)


Database sharding

On the other hand, sharding can be used to horizontally scale Weaviate. Sharding is the process of splitting a database into smaller parts, called shards, and distributing these shards across multiple nodes.

So a database that holds 2 million records do not need to store all 2 million records on a single node. Instead, it can split the records into smaller parts and store them on different nodes. This allows the database to scale horizontally by adding more nodes to the cluster.

Sharding with replication

You can use both sharding and replication together to horizontally scale and ensure fault tolerance in your Weaviate setup.

In our example with 6 nodes, we could have a replication factor of 3 and shard the data across 2 nodes each. This way, we have 3 copies of the data spread across 6 nodes, ensuring fault tolerance and high availability.

For a production setup, this is a common approach to ensure that your Weaviate setup can handle high loads and remain available even if a node fails. Additionally, you can add more nodes to the cluster as needed to scale your Weaviate setup horizontally, ensuring that it can handle more data and more requests.


That's it for this guide. You've successfully deployed Weaviate on a multi-node Kubernetes cluster and learned about replication and sharding.

If you would like, you can access Weaviate as described before, and work with it. We include further resources on the next page for you to explore.

When you are finished, you can stop the tunnel by pressing Ctrl+C in the terminal where you ran minikube tunnel, Then, stop Minikube and delete the cluster:

minikube stop
minikube delete

Questions and feedback

If you have any questions or feedback, please let us know on our forum. For example, you can: