go-backpressure

Fun with TCP Flow Control

Purpose

You can measure TCP Flow Control ("back pressure") functionality by starting a websocket server and pushing more data to it than it can read at a time.

TCP Backpressure can be a tricky thing because it only applies to a connection between two hosts while in modern cloud applications, the client usually never talks to the application directly. In-between there are loadbalancers, routers, proxies and more loadbalancers. Depending on the type of loadbalancer in front, the "slow down" message might never reach the client.

This app is meant for various testing scenarios such as

• local deployments
• single VM public IP deployments
• VM behind loadbalancer(s)
• CloudFoundry App deployments

How to use

The app supports both standard TCP and websocket connections using the gorilla websocket implementation.

• To start the server:

  • Set protocol: tcp or protocol: websocket in server/config.yml
  • Run command

  go build
  ./go-backpressure -server server/config.yml
  

• To start the client:

  • Set protocol: tcp or protocol: websocket in client/config.yml
  • Run command

  go build
  ./go-backpressure -client client/config.yml
  

• To use the webclient:

  • The websocket server also provides a minimal webclient
  • Set protocol: websocket in server/config.yml
  • Point your browser to http://localhost:8080/client
  • To see console output, open developer tools

Capturing Traffic

To observe when the server advertises a zero window, using Wireshark is recommended. When running locally, use these settings:

• Capture on Loopback lo0
• Set filter to tcp.port == 8080 and not websocket to only see pure TCP traffic.
• Open one TCP PDU (=protocol data unit, i.e. packet / datagram / segment)
  • Righ-click on "Calculated Window Size", select "Apply as column"

When running remote, you will likely put TLS in front using HAproxy or NGINX. Notice that since the traffic will then be encrypted, Wireshark can no longer detect websocket traffic. All traffic will show as pure TCP and TLS.

Advertised window sizes will still be visible though.

Deploying on Docker

The project provides a simple Dockerfile to allow you to deploy it on any server that runs docker.

To build, simply run

docker build -t go-backpressure .

To deploy the server on a remote machine, run

docker run -d -p 8080:8080 go-backpressure

Deploying on CloudFoundry

The project also provides a minimal manifest for CF deployment.

After logging in to your org and space, simply run

cf push