An output is a sink where we wish to send our consumed data after applying an optional array of processors. Only one output is configured at the root of a Benthos config. However, the output can be a broker which combines multiple outputs under a chosen brokering pattern.

An output config section looks like this:

bucket: TODO
path: "${!metadata:kafka_topic}/${!json_field:message.id}.json"
# Optional list of processing steps
- jmespath:
query: '{ message: @, meta: { link_count: length(links) } }'

Back Pressure

Benthos outputs apply back pressure to components upstream. This means if your output target starts blocking traffic Benthos will gracefully stop consuming until the issue is resolved.


When a Benthos output fails to send a message the error is propagated back up to the input, where depending on the protocol it will either be pushed back to the source as a Noack (e.g. AMQP) or will be reattempted indefinitely with the commit withheld until success (e.g. Kafka).

It's possible to instead have Benthos indefinitely retry an output until success with a retry output. Some other outputs, such as the broker, might also retry indefinitely depending on their configuration.

Multiplexing Outputs

It is possible to perform content based multiplexing of messages to specific outputs either by using the switch output, or a broker with the fan_out pattern and a filter processor on each output, which is a processor that drops messages if the condition does not pass. Conditions are content aware logical operators that can be combined using boolean logic.

For more information regarding conditions, including a full list of available conditions please read the docs here.

Dead Letter Queues

It's possible to create fallback outputs for when an output target fails using a try output.