> ## Documentation Index
> Fetch the complete documentation index at: https://docs.goldsky.com/llms.txt
> Use this file to discover all available pages before exploring further.

# S2

> Publish Turbo pipeline output to S2.dev streams for serverless event streaming, with configurable batch_size and batch_flush_interval delivery.

## Overview

Publish to an [S2.dev](https://s2.dev) stream.

S2 is a serverless HTTP API for durable, append-only streams with support for high read fanout and bottomless storage on object storage.

Use it to share streams of data with front-ends and back-ends as a simpler alternative to Kafka, AWS SQS, or Google Pub/Sub.

<Tip>
  S2 is ideal for serverless architectures and teams using TypeScript or Python stacks who want a simpler alternative to Kafka for event streaming. Try the [S2 Playground](https://s2.dev/playground) to get started.
</Tip>

## Configuration

```yaml theme={null}
sinks:
  my_s2_sink:
    type: s2_sink
    from: my_transform
    access_token: your_access_token_here
    basin: your-basin-name
    stream: your-stream-name
```

## Parameters

<ParamField path="type" type="string" required>
  Must be `s2_sink`
</ParamField>

<ParamField path="from" type="string" required>
  The transform or source to read data from
</ParamField>

<ParamField path="access_token" type="string" required>
  S2 access token for authentication. Create one at [s2.dev](https://s2.dev).
</ParamField>

<ParamField path="basin" type="string" required>
  Basin name. Must be a valid S2 [BasinName](https://s2.dev/docs/basins).
</ParamField>

<ParamField path="stream" type="string" required>
  Stream name to append records to.
</ParamField>

<ParamField path="batch_size" type="integer" default="1000">
  Runtime-level parameter — maximum number of rows the Turbo runtime buffers before appending them to the S2 stream. See [Batching](/turbo-pipelines/sinks/overview#batching) on the overview page.
</ParamField>

<ParamField path="batch_flush_interval" type="string" default="1s">
  Runtime-level parameter — maximum time the Turbo runtime waits before flushing a batch, even if `batch_size` has not been reached. Accepts humantime durations such as `500ms`, `1s`, or `30s`. See [Batching](/turbo-pipelines/sinks/overview#batching).
</ParamField>

## Behavior

* **Append-only**: Each pipeline record is appended to the S2 stream as a single record. S2 streams are append-only by design, so updates and deletes are not supported.
* **JSON encoding**: Records are serialized as line-delimited JSON (one JSON object per S2 record) before being appended.
* **Auto batching**: Records are grouped into `AppendRecordBatch` instances up to S2's per-batch size limits. When a batch fills up, the sink starts a new one and keeps appending.
* **Serverless**: No infrastructure to manage. The sink talks to S2 over its HTTP API.

## Example

Stream ERC-20 transfers from a Goldsky dataset to an S2 stream:

```yaml theme={null}
name: erc20-transfers-to-s2
resource_size: s

sources:
  transfers:
    type: dataset
    dataset_name: ethereum.erc20_transfers
    version: 1.2.0
    start_at: latest

transforms:
  filtered_transfers:
    type: sql
    primary_key: id
    sql: |
      SELECT *
      FROM transfers
      WHERE CAST(value AS DECIMAL) > 1000000000000000000

sinks:
  s2_output:
    type: s2_sink
    from: filtered_transfers
    access_token: your_access_token_here
    basin: your-basin-name
    stream: erc20-transfers
```


## Related topics

- [Turbo - Supported sinks](/turbo-pipelines/sinks/overview.md)
- [Benefits](/benefits.md)
- [Mirror - Supported sinks](/mirror/sinks/supported-sinks.md)
- [Agent Skills](/ai-skills.md)
- [Mirror vs. Turbo pipelines](/mirror-vs-turbo.md)