Horizontal Scaling

RudderStack makes it easier for you to add more nodes to your Kubernetes cluster, to process the events in a timely and effective manner. This guide shows you how to effectively configure your Helm charts to trigger a horizontal scale up/down, based on the version of RudderStack you are using.

RudderStack Open-source Version

Each RudderStack node consists of a backend pod and a PostgreSQL pod. In case a RudderStack cluster is not able to process the events, it is advisable to add more more nodes to it. By changing the backendReplicaCount in the Helm configuration, you can trigger horizontal scale up/down. If you set this value to any positive number, helm charts will create as many RudderStack nodes.

The values.yaml file can be found here.

There are, however, a couple of challenges in scaling up or down this way.

Event Ordering

In a single-node cluster, all events are routed to the same RudderStack node. This guarantees the ordering of all events it has received for a given user. In case of a multi-node cluster, since the events of a same user can go to different nodes, the event ordering is not guaranteed by the cluster.

Scale Down

If you scale down, all pending events in the terminated node will not be delivered to the destination. You will need to make sure that all events in a node are drained before terminating it.

To understand how RudderStack handles hardware failures and guarantees High Availability, please go through our High Availability guide.

RudderStack Enterprise Version

The Enterprise version of RudderStack makes horizontal scaling seamless, while maintaining the event ordering and guaranteeing the delivery of all events.

Event Ordering

RudderStack Enterprise has a proxy service that implements a routing algorithm. It forwards a given set of user events to the same RudderStack node every time. Each node then guarantees the event ordering while delivering to the specified destinations. This proxy service is highly available and scalable.

Scale Down

RudderStack Enterprise includes a separate process named Jobs DB Migrator. This process is responsible for moving all pending events from the existing nodes to the new nodes after scaling up or down. It makes sure all events are migrated to the new nodes and the event ordering is maintained during and after the migration process.

Questions? Contact us by email or on Slack