# High Availability

You can run a cluster of Typesense nodes for high availability. Typesense uses the Raft consensus algorithm to manage the cluster and recover from node failures.

In cluster mode, Typesense will automatically replicate your entire dataset to all nodes in the cluster, automatically and continuously. Read and write API calls can be sent to any nodes in the cluster - read API calls will be served by the node that receives it, write API calls are automatically forwarded to the leader of the cluster internally.

Since Raft requires a quorum for consensus, you need to run a minimum of 3 nodes to tolerate a 1-node failure. Running a 5-node cluster will tolerate failures of up to 2 nodes, but at the expense of slightly higher write latencies.

# High Availability in Typesense Cloud

In Typesense Cloud (opens new window), we manage High Availability for you.

When you flip the setting ON when launching a cluster, you'll see a special Load Balanced endpoint in addition to the individual hostnames*, in your cluster dashboard:

Typesense Cloud HA Hostnames

Requests sent to the Load-Balanced endpoint are distributed between all the 3 nodes in the cluster. If a particular node has an infrastructure issue, or is inaccessible for any reason, it is automatically quarantined and traffic is re-routed to the other healthy nodes.


You will only see the Load Balanced endpoint for HA clusters provisioned after June 16, 2022.

For HA clusters provisioned before June 16, 2022, you will only see the individual hostnames. Health-checking and traffic re-routing are done client-side in our official client libraries. See Client Configuration below. If you'd like to enable server-side load-balancing for your existing clusters, please reach out to us at support at typesense d0t org with your Cluster ID.

# High Availability when Self-Hosting Typesense

# Configuring a Typesense cluster

To start a Typesense node as part of a cluster, create a new file on each node that's part of the cluster with the following format, and use the --nodes server configuration to point to the file.

Each node definition in the file should be in the following format, separated by commas:


peering_address, peering_port and api_port should match the corresponding Server Configuration Parameters used when starting the Typesense process on each node.

All nodes in the cluster should have the same bootstrap --api-key for security purposes.

# Nodes File Example

Here's an example of a --nodes file for a 3-node cluster:

In the example above

  • The peering_address (the IP address used for cluster operations) is 192.168.12.x
  • The peering_port (the port used for cluster operations) is 8107
  • The api_port (the actual port to which clients connect to) is 8108

Here's the corresponding command to start the Typesense process on each node:


  • --peering-address should be a Private IP address, since it is only meant for internal cluster operations and contains unencrypted Raft data that is exchanged between nodes.

  • --api-address can be a public or private IP address. This is the IP address that your end users/clients will connect to interact with the Typesense API.

  • We strongly recommend setting --api-port to 443 (HTTPS) in a production setting, and configuring SSL certs using the --ssl-certificate and --ssl-certificate-key server parameters.


If you are using Docker, make sure that you've configured the Docker network in such a way that the Typesense process within the Docker container can communicate with the other Typesense processes using their IP addresses. Read more about Docker Networking (opens new window).

# Verifying Cluster Formation

Once you've setup all the nodes in a cluster, you can verify that they've successfully formed a cluster by sending a GET request to the /debug endpoint of each node:

curl "http://${TYPESENSE_HOST}/debug/" \

On one of the nodes you should see the following response:

  "state": 1,
  "version": "x.x.x"

where state: 1 indicates that this is the node that was elected to be the leader.

All the other nodes should return a response of:

  "state": 4,
  "version": "x.x.x"

where state: 4 indicates that the node was selected to be a follower.

If you see state: 1 on more than one node, that indicates that the cluster was not formed properly. Check the Typesense logs (usually in /var/log/typesense/typesense.log) for more diagnostic information. Ensure that the nodes can talk to each other on the ports you've configured as the HTTP port and the Peering port.

If you see a value other than state: 4 or state: 1 that indicates an error. Check the Typesense logs (usually in /var/log/typesense/typesense.log) for more diagnostic information.

# Recovering a cluster that has lost quorum

A Typesense cluster with N nodes can tolerate a failure of at most (N-1)/2 nodes without affecting reads or writes.

So for example:

  • A 3 node cluster can handle a loss of 1 node.
  • A 5 node cluster can handle a loss of 2 nodes.

If a Typesense cluster loses more than (N-1)/2 nodes at the same time, the cluster becomes unstable because it loses quorum and the remaining node(s) cannot safely build consensus on which node is the leader. To avoid a potential split brain issue, Typesense then stops accepting writes and reads until some manual verification and intervention is done.

To recover a cluster in this state:

  1. Force one of the nodes to become a single-node cluster by editing its nodes file to contain just its own IP address. You don't have to restart the Typesense process, since changes to the node file are automatically picked up within 30s.
  2. Once this node returns ok when you call /health, edit the nodes file and then add back another node in the cluster, and wait for it to catch up with the new leader.
  3. Repeat Step 2 for each node until the cluster is healthy again.

# Client Configuration

# When deployed without a load balancer

Typesense clients allow you to specify one or more nodes during client initialization. So you can specify the individual hostnames in the cluster when instantiating the client library, and it will load balance reads & writes across all nodes and will automatically strive to recover from transient failures through built-in retries.

Here's a sample 3-node client configuration:

# When using Typesense Cloud or a Load Balancer

If you use Typesense Cloud (with load-balancing enabled on your cluster, which is enabled by default for all clusters provisioned after June 16, 2022), or if you choose to set up a server-side load-balancer for convenience, you can specify the single load-balanced endpoint, instead of specifying each of the individual ones in the client libraries.

Here xxx.a1.typesense.net is a Load Balanced endpoint.

Last Updated: 4/6/2024, 6:43:57 AM