Skip to main content

Streams Mode

A Benthos stream consists of four components; an input, an optional buffer, processor pipelines and an output. Under normal use a Benthos instance is a single stream, and these components are configured within the service config file.

Alternatively, Benthos can be run in streams mode, where a single running Benthos instance is able to run multiple entirely isolated streams. Adding streams in this mode can be done in two ways:

  1. Static configuration files allows you to maintain a directory of static stream configuration files that will be traversed by Benthos.

  2. An HTTP REST API allows you to dynamically create, read the status of, update, and delete streams at runtime.

These two methods can be used in combination, i.e. it's possible to update and delete streams that were created with static files.

When running Benthos in streams mode it is still necessary to provide a general service wide configuration with the -c/--config flag that specifies observability configuration such as the metrics, logger and tracing sections, as well the http section for configuring how the HTTP server should behave.

You can import resources either in the general configuration, or using the -r/--resources flag, the same as when running Benthos in regular mode.

benthos -r "./prod/*.yaml" -c ./config.yaml streams


When running Benthos in streams mode resource components are shared across all streams. The streams mode HTTP API also provides an endpoint for modifying and adding resource configurations dynamically.


Metrics from all streams are aggregated and exposed via the method specified in the config of the Benthos instance running in streams mode, with their metrics prefixed by their respective stream name.

For example, a Benthos instance running in streams mode with the configured prefix benthos running a stream named foo would have metrics from foo registered with the prefix

This can cause problems if your streams are short lived and uniquely named as the number of metrics registered will continue to climb indefinitely. In order to avoid this you can use the path_mapping field to filter metric names.

# Only register metrics for the stream `foo`. Others will be ignored.
prefix: benthos
path_mapping: if !this.has_prefix("foo") { deleted() }

Or use it to rename prefixes in order to reduce the cardinality of names:

# Rename all stream metric prefixes of the form `foo_<uuid_v4>` to just `foo`.
prefix: benthos
address: localhost:8125
flush_period: 100m
path_mapping: this.re_replace("foo_[0-9\\-a-zA-Z]+\\.(.*)","foo.$1")