---
metadata:
  - name: generator
    content: Diplodoc Platform v5.50.6
  - property: og:title
    content: Configuring external access to YTsaurus
  - property: og:description
    content: The instructions below describe how to configure external access to YTsaurus deployed in a Kubernetes cluster
  - property: og:type
    content: article
  - property: og:url
    content: https://ytsaurus.tech/docs/ru/admin-guide/cluster-access-proxy
  - property: article:tag
    content: k8s
  - property: article:modified_time
    content: '2026-03-17T20:00:00+03:00'
  - property: article:author
    content: Gorbacheva Olesya
alternate:
  - https://ytsaurus.tech/docs/en/admin-guide/cluster-access-proxy/spyt.md
  - https://ytsaurus.tech/docs/ru/admin-guide/cluster-access-proxy/spyt.md
---
> **Documentation Index:** Fetch the complete configuration index at https://ytsaurus.tech/docs/en/llms.txt


<!-- source: en/_includes/admin-guide/cluster-access-proxy/spyt.md -->
# Configure access to SPYT

For SPYT to work in [Standalone](https://ytsaurus.tech/docs/en/user-guide/data-processing/spyt/cluster/cluster-desc.md#spark-standalone) mode, the external Spark driver must connect directly to the Spark workers. These workers run not as separate Kubernetes pods, but as processes inside [Vanilla jobs](https://ytsaurus.tech/docs/en/user-guide/data-processing/operations/vanilla.md) in YTsaurus. Standard Kubernetes network abstractions (Services) can't route traffic to such processes.

To proxy these connections, `tcp_proxy` is used. This is a YTsaurus component that knows which node and port a specific worker process runs on, and forwards external traffic to it using routing tables stored in Cypress at `//sys/tcp_proxies/routes`.

To set up `tcp_proxy`:

1. Ensure the ports on which `tcp_proxy` runs (32000–32019 by default) are exposed via Kubernetes services (NodePort or LoadBalancer).
2. Specify the external addresses of these ports in the attribute `//sys/tcp_proxies/routes/<proxy_role>/@external_addresses` in Cypress.

This allows the Spark driver to use `tcp_proxy` as an intermediary to communicate with the internal workers.

#### Example

```bash
yt set //sys/tcp_proxies/routes/default/@external_addresses '["node1.example.com:32000"; "node2.example.com:32000"]'
```
<!-- endsource: en/_includes/admin-guide/cluster-access-proxy/spyt.md -->