benthos

README

Benthos is a low latency stream buffer service for piping large volumes of data from one source to another, able to protect against back pressure caused by data surges or service outages further down the pipeline.

Benthos supports a range of input and output protocols out of the box, allowing you to drop benthos in between many existing pipeline services with minimal changes to your architecture.

A range of internal buffer strategies are available, allowing you to choose a balance between latency, protection against back pressure and file based persistence.

## Design

Benthos has inputs, an optional buffer, and outputs, which are all set in a single config file.

+-------------------------+                  +-------------------------+
|      Input Stream       |                  |     Output Stream       |
| ( ZMQ, HTTP Post, etc ) |--+            +->| ( ZMQ, HTTP Post, etc ) |
+-------------------------+  |            |  +-------------------------+
                             v            |
             +--------------------------------------------+
             |                   Buffer                   |
             | ( Memory-Mapped Files, Memory, None, etc ) |
             +--------------------------------------------+

For a full list of available inputs, outputs or buffers you can run:

# Print inputs
benthos --print-inputs | less

# Print buffers
benthos --print-buffers | less

# Print outputs
benthos --print-outputs | less

## Support

Currently supported input/output targets:

• ZMQ4 (PUSH, PULL, SUB, PUB)
• Nanomsg/Scalability Protocols (PUSH, PULL, SUB, PUB)
• HTTP 1.1 POST
• STDIN/STDOUT
• File
• Kafka 0.9

You can also have multiple outputs or inputs by configuring a routing strategy (fan in, fan out, round robin, etc).

Speed and Benchmarks

The main goal of Benthos is to be stable, low-latency and high throughput. Benthos comes with two benchmarking tools using ZMQ4 benthos_zmq_producer and benthos_zmq_consumer. The producer pushes an endless stream of fixed size byte messages with a configured time interval between each message. The consumer will read an endless stream of messages and print a running average of latency, throughput (per second) and total messages received.

These tools are useful for getting ballpark figures, but it would be good to get some more meaningful third party benchmarks to publish.

Heres a table of results from an 8-core (2.4ghz) machine using the ZMQ4 input, mmap_file persisted buffer, ZMQ4 output configuration, and with messages of size 5000 bytes:

Stream Interval	Avg. Latency (us)	Msg. Rate (msgs/s)	Byte Rate (MB/s)
100ms	247	9.66	0.05
10ms	518	75.71	0.36
1ms	561	606.35	2.90
100us	734	4454.05	21.32
10us	2665	7178.02	34.36
1us	174099	20098.26	96.20

Install

go get github.com/jeffail/benthos/cmd/...

ZMQ4 Support

Benthos supports ZMQ4 for both data input and output. To add this you need to install libzmq4 and use the compile time flag when building benthos:

go install -tags "ZMQ4" ./cmd/...

Run

benthos -c ./config.yaml

## Config

Create a fully populated default configuration file:

benthos --print-yaml > config.yaml
benthos --print-json > config.json

The configuration file should contain a section for an input, output, and a buffer. For example, if we wanted to output to a ZMQ4 push socket our output section in a YAML config might look like this:

output:
  type: zmq4
  zmq4:
    addresses:
      - tcp://*:1234
    socket_type: PUSH

More docs

For further information check out the ./docs directory.

## Vendoring

Versions of go above 1.6 should automatically go get all vendored libraries. Otherwise, while cloning use --recursive:

git clone https://github.com/jeffail/benthos --recursive

Or, if the repo is already cloned, get the latest libraries with:

git submodule update --init

To add new libraries simply run:

PACKAGE=github.com/jeffail/util
git submodule add https://$PACKAGE vendor/$PACKAGE"

It might be handy to set yourself a function for this in your .bashrc:

function go-add-vendor {
	git submodule add https://$1 vendor/$1
}