---
layout: default
title: s3
parent: Sinks
grand_parent: Pipelines
nav_order: 55
---

# s3

The `s3` sink saves batches of events to [Amazon Simple Storage Service (Amazon S3)](https://aws.amazon.com/s3/) objects.

## Usage

The following example creates a pipeline configured with an s3 sink. It contains additional options for customizing the event and size thresholds for which the pipeline sends record events and sets the codec type `ndjson`:

```
pipeline:
  ...
  sink:
    - s3:
        aws:
          region: us-east-1
          sts_role_arn: arn:aws:iam::123456789012:role/Data-Prepper
          sts_header_overrides:
        max_retries: 5
        bucket:
          name: bucket_name
          object_key:
            path_prefix: my-elb/%{yyyy}/%{MM}/%{dd}/
        threshold:
          event_count: 2000
          maximum_size: 50mb
          event_collect_timeout: 15s
        codec:
          ndjson:
        buffer_type: in_memory
```

## Configuration 

Use the following options when customizing the `s3` sink.

Option | Required | Type | Description
:--- | :--- | :--- | :---
`bucket` | Yes | String | The object from which the data is retrieved and then stored. The `name` must match the name of your object store.
`codec` | Yes | [Buffer type](#buffer-type) | Determines the buffer type. 
`aws` | Yes | AWS | The AWS configuration. See [aws](#aws) for more information.
`threshold` | Yes | [Threshold](#threshold-configuration) | Configures when to write an object to S3. 
`object_key` | No | Sets the `path_prefix` and the `file_pattern` of the object store. Defaults to the S3 object `events-%{yyyy-MM-dd'T'hh-mm-ss}` found inside the root directory of the bucket.
`compression` | No | String | The compression algorithm to apply: `none`, `gzip`, or `snappy`. Default is `none`. 
`buffer_type` | No | [Buffer type](#buffer-type) | Determines the buffer type. 
`max_retries` | No | Integer | The maximum number of times a single request should retry when ingesting data to S3. Defaults to `5`.

## aws

Option | Required | Type | Description
:--- | :--- | :--- | :---
`region` | No | String | The AWS Region to use for credentials. Defaults to [standard SDK behavior to determine the Region](https://docs.aws.amazon.com/sdk-for-java/latest/developer-guide/region-selection.html).
`sts_role_arn` | No | String | The AWS Security Token Service (AWS STS) role to assume for requests to Amazon SQS and Amazon S3. Defaults to `null`, which will use the [standard SDK behavior for credentials](https://docs.aws.amazon.com/sdk-for-java/latest/developer-guide/credentials.html).
`sts_header_overrides` | No | Map | A map of header overrides that the IAM role assumes for the sink plugin.
`sts_external_id` | No | String | The external ID to attach to AssumeRole requests from AWS STS.


## Threshold configuration

Use the following options to set ingestion thresholds for the `s3` sink.

Option | Required | Type | Description
:--- | :--- | :--- | :---
`event_count` | Yes | Integer | The maximum number of events the S3 bucket can ingest. 
`maximum_size` | Yes | String | The maximum number of bytes that the S3 bucket can ingest after compression. Defaults to `50mb`.
`event_collect_timeout` | Yes | String | Sets the time period during which events are collected before ingestion. All values are strings that represent duration, either an ISO_8601 notation string, such as `PT20.345S`, or a simple notation, such as `60s` or `1500ms`.


## Buffer type

`buffer_type` is an optional configuration that records stored events temporarily before flushing them into an S3 bucket. The default value is `in_memory`. Use one of the following options: 

- `in_memory`: Stores the record in memory.
- `local_file`: Flushes the record into a file on your machine. 
- `multipart`: Writes using the [S3 multipart upload](https://docs.aws.amazon.com/AmazonS3/latest/userguide/mpuoverview.html). Every 10 MB is written as a part.

## Object key configuration

Option | Required | Type | Description
:--- | :--- | :--- | :---
`path_prefix` | Yes | String | The S3 key prefix path to use. Accepts date-time formatting. For example, you can use `%{yyyy}/%{MM}/%{dd}/%{HH}/` to create hourly folders in S3. By default, events write to the root of the bucket. 


## codec

The `codec` determines how the `s3` source formats data written to each S3 object.

### avro codec

The `avro` codec writes an event as an [Apache Avro](https://avro.apache.org/) document.

Because Avro requires a schema, you may either define the schema yourself, or Data Prepper will automatically generate a schema.
In general, you should define your own schema because it will most accurately reflect your needs.

We recommend that you make your Avro fields use a null [union](https://avro.apache.org/docs/current/specification/#unions).
Without the null union, each field must be present or the data will fail to write to the sink.
If you can be certain that each each event has a given field, you can make it non-nullable.

When you provide your own Avro schema, that schema defines the final structure of your data.
Therefore, any extra values inside any incoming events that are not mapped in the Arvo schema will not be included in the final destination.
To avoid confusion between a custom Arvo schema and the `include_keys` or `exclude_keys` sink configurations, Data Prepper does not allow the use of the `include_keys` or `exclude_keys` with a custom schema.

In cases where your data is uniform, you may be able to automatically generate a schema.
Automatically generated schemas are based on the first event received by the codec.
The schema will only contain keys from this event.
Therefore, you must have all keys present in all events in order for the automatically generated schema to produce a working schema.
Automatically generated schemas make all fields nullable.
Use the sink's `include_keys` and `exclude_keys` configurations to control what data is included in the auto-generated schema.


Option | Required | Type | Description
:--- | :--- | :--- | :---
`schema` | Yes | String | The Avro [schema declaration](https://avro.apache.org/docs/current/specification/#schema-declaration). Not required if `auto_schema` is set to true.
`auto_schema` | No | Boolean | When set to `true`, automatically generates the Avro [schema declaration](https://avro.apache.org/docs/current/specification/#schema-declaration) from the first event.

 
### ndjson codec

The `ndjson` codec writes each line as a JSON object.

The `ndjson` codec does not take any configurations.


### json codec

The `json` codec writes events in a single large JSON file.
Each event is written into an object within a JSON array.


Option | Required | Type | Description
:--- | :--- | :--- | :---
`key_name` | No | String | The name of the key for the JSON array. By default this is `events`.


### parquet codec

The `parquet` codec writes events into a Parquet file.
When using the Parquet codec, set the `buffer_type` to `in_memory`.

The Parquet codec writes data using the Avro schema.
Because Parquet requires an Avro schema, you may either define the schema yourself, or Data Prepper will automatically generate a schema.
However, we generally recommend that you define your own schema so that it can best meet your needs.

For details on the Avro schema and recommendations, see the  [Avro codec](#avro-codec) documentation.


Option | Required | Type | Description
:--- | :--- | :--- | :---
`schema` | Yes | String | The Avro [schema declaration](https://avro.apache.org/docs/current/specification/#schema-declaration). Not required if `auto_schema` is set to true.
`auto_schema` | No | Boolean | When set to `true`, automatically generates the Avro [schema declaration](https://avro.apache.org/docs/current/specification/#schema-declaration) from the first event.