Shopify/toxiproxy

Toxiproxy is a framework for simulating network conditions. It's made specifically to work in testing, CI and development environments, supporting deterministic tampering with connections, but with support for randomized chaos and customization. Toxiproxy is the tool you need to prove with tests that your application doesn't have single points of failure. We've been successfully using it in all development and test environments at Shopify since October, 2014. See our blog post on resiliency for more information.

Toxiproxy usage consists of two parts. A TCP proxy written in Go (what this repository contains) and a client communicating with the proxy over HTTP. You configure your application to make all test connections go through Toxiproxy and can then manipulate their health via HTTP. See Usage below on how to set up your project.

For example, to add 1000ms of latency to the response of MySQL from the Ruby client:

Toxiproxy[:mysql_master].downstream(:latency, latency: 1000).apply do
  Shop.first # this takes at least 1s
end

To take down all Redis instances:

Toxiproxy[/redis/].down do
  Shop.first # this will throw an exception
end

While the examples in this README are currently in Ruby, there's nothing stopping you from creating a client in any other language (see Clients).

• Toxiproxy
  • Table of Contents
  • Why yet another chaotic TCP proxy?
  • Clients
  • Example
  • Usage
    • 1. Installing Toxiproxy
      • Upgrading from Toxiproxy 1.x
    • 2. Populating Toxiproxy
    • 3. Using Toxiproxy
    • 4. Logging
    • Toxics
      • latency
      • down
      • bandwidth
      • slow_close
      • timeout
      • reset_peer
      • slicer
      • limit_data
      • packet_loss
    • HTTP API
      • Proxy fields:
      • Toxic fields:
      • Endpoints
      • Populating Proxies
    • CLI Example
    • Metrics
    • Frequently Asked Questions
    • Development
    • Release

The existing ones we found didn't provide the kind of dynamic API we needed for integration and unit testing. Linux tools like nc and so on are not cross-platform and require root, which makes them problematic in test, development and CI environments.

• toxiproxy-ruby
• toxiproxy-go
• toxiproxy-python
• toxiproxy.net
• toxiproxy-php-client
• toxiproxy-node-client
• toxiproxy-java
• toxiproxy-haskell
• toxiproxy-rust
• toxiproxy-elixir

Let's walk through an example with a Rails application. Note that Toxiproxy is in no way tied to Ruby, it's just been our first use case. You can see the full example at sirupsen/toxiproxy-rails-example. To get started right away, jump down to Usage.

For our popular blog, for some reason we're storing the tags for our posts in Redis and the posts themselves in MySQL. We might have a Post class that includes some methods to manipulate tags in a Redis set:

class Post < ActiveRecord::Base
  # Return an Array of all the tags.
  def tags
    TagRedis.smembers(tag_key)
  end

  # Add a tag to the post.
  def add_tag(tag)
    TagRedis.sadd(tag_key, tag)
  end

  # Remove a tag from the post.
  def remove_tag(tag)
    TagRedis.srem(tag_key, tag)
  end

  # Return the key in Redis for the set of tags for the post.
  def tag_key
    "post:tags:#{self.id}"
  end
end

We've decided that erroring while writing to the tag data store (adding/removing) is OK. However, if the tag data store is down, we should be able to see the post with no tags. We could simply rescue the Redis::CannotConnectError around the SMEMBERS Redis call in the tags method. Let's use Toxiproxy to test that.

Since we've already installed Toxiproxy and it's running on our machine, we can skip to step 2. This is where we need to make sure Toxiproxy has a mapping for Redis tags. To config/boot.rb (before any connection is made) we add:

require 'toxiproxy'

Toxiproxy.populate([
  {
    name: "toxiproxy_test_redis_tags",
    listen: "127.0.0.1:22222",
    upstream: "127.0.0.1:6379"
  }
])

Then in config/environments/test.rb we set the TagRedis to be a Redis client that connects to Redis through Toxiproxy by adding this line:

TagRedis = Redis.new(port: 22222)

All calls in the test environment now go through Toxiproxy. That means we can add a unit test where we simulate a failure:

test "should return empty array when tag redis is down when listing tags" do
  @post.add_tag "mammals"

  # Take down all Redises in Toxiproxy
  Toxiproxy[/redis/].down do
    assert_equal [], @post.tags
  end
end

The test fails with Redis::CannotConnectError. Perfect! Toxiproxy took down the Redis successfully for the duration of the closure. Let's fix the tags method to be resilient:

def tags
  TagRedis.smembers(tag_key)
rescue Redis::CannotConnectError
  []
end

The tests pass! We now have a unit test that proves fetching the tags when Redis is down returns an empty array, instead of throwing an exception. For full coverage you should also write an integration test that wraps fetching the entire blog post page when Redis is down.

Full example application is at sirupsen/toxiproxy-rails-example.

Configuring a project to use Toxiproxy consists of three steps:

1. Installing Toxiproxy
2. Populating Toxiproxy
3. Using Toxiproxy

Linux

See Releases for the latest binaries and system packages for your architecture.

Ubuntu

$ wget -O toxiproxy-2.12.0.deb https://github.com/Shopify/toxiproxy/releases/download/v2.12.0/toxiproxy_2.12.0_linux_amd64.deb
$ sudo dpkg -i toxiproxy-2.12.0.deb
$ sudo service toxiproxy start

OS X

With Homebrew:

$ brew tap shopify/shopify
$ brew install toxiproxy

Or with MacPorts:

$ port install toxiproxy

Windows

Toxiproxy for Windows is available for download at https://github.com/Shopify/toxiproxy/releases/latest/download/toxiproxy-server-windows-amd64.exe, with the matching CLI at https://github.com/Shopify/toxiproxy/releases/latest/download/toxiproxy-cli-windows-amd64.exe

Docker

Toxiproxy is available on Github container registry. Old versions <= 2.1.4 are available on on Docker Hub.

$ docker pull ghcr.io/shopify/toxiproxy
$ docker run --rm -it ghcr.io/shopify/toxiproxy

If using Toxiproxy from the host rather than other containers, enable host networking with --net=host. The image binds the API to 0.0.0.0, so with host networking it is reachable on every host interface; pass -e TOXIPROXY_AUTH_TOKEN=<token> if other machines can reach the host (see HTTP API).

$ docker run --rm --entrypoint="/toxiproxy-cli" -it ghcr.io/shopify/toxiproxy list

Source

If you have Go installed, you can build Toxiproxy from source using the make file:

$ make build
$ ./dist/toxiproxy-server

In Toxiproxy 2.0 several changes were made to the API that make it incompatible with version 1.x. In order to use version 2.x of the Toxiproxy server, you will need to make sure your client library supports the same version. You can check which version of Toxiproxy you are running by looking at the /version endpoint.

See the documentation for your client library for specific library changes. Detailed changes for the Toxiproxy server can been found in CHANGELOG.md.

When your application boots, it needs to make sure that Toxiproxy knows which endpoints to proxy where. The main parameters are: name, address for Toxiproxy to listen on and the address of the upstream.

Some client libraries have helpers for this task, which is essentially just making sure each proxy in a list is created. Example from the Ruby client:

# Make sure `shopify_test_redis_master` and `shopify_test_mysql_master` are
# present in Toxiproxy
Toxiproxy.populate([
  {
    name: "shopify_test_redis_master",
    listen: "127.0.0.1:22220",
    upstream: "127.0.0.1:6379"
  },
  {
    name: "shopify_test_mysql_master",
    listen: "127.0.0.1:24220",
    upstream: "127.0.0.1:3306"
  }
])

This code needs to run as early in boot as possible, before any code establishes a connection through Toxiproxy. Please check your client library for documentation on the population helpers.

Alternatively use the CLI to create proxies, e.g.:

toxiproxy-cli create -l localhost:26379 -u localhost:6379 shopify_test_redis_master

We recommend a naming such as the above: <app>_<env>_<data store>_<shard>. This makes sure there are no clashes between applications using the same Toxiproxy.

For large application we recommend storing the Toxiproxy configurations in a separate configuration file. We use config/toxiproxy.json. This file can be passed to the server using the -config option, or loaded by the application to use with the populate function.

An example config/toxiproxy.json:

[
  {
    "name": "web_dev_frontend_1",
    "listen": "[::]:18080",
    "upstream": "webapp.domain:8080",
    "enabled": true
  },
  {
    "name": "web_dev_mysql_1",
    "listen": "[::]:13306",
    "upstream": "database.domain:3306",
    "enabled": true
  }
]

Use ports outside the ephemeral port range to avoid random port conflicts. It's 32,768 to 61,000 on Linux by default, see /proc/sys/net/ipv4/ip_local_port_range.

To use Toxiproxy, you now need to configure your application to connect through Toxiproxy. Continuing with our example from step two, we can configure our Redis client to connect through Toxiproxy:

# old straight to redis
redis = Redis.new(port: 6380)

# new through toxiproxy
redis = Redis.new(port: 22220)

Now you can tamper with it through the Toxiproxy API. In Ruby:

redis = Redis.new(port: 22220)

Toxiproxy[:shopify_test_redis_master].downstream(:latency, latency: 1000).apply do
  redis.get("test") # will take 1s
end

Or via the CLI:

toxiproxy-cli toxic add -t latency -a latency=1000 shopify_test_redis_master

Please consult your respective client library on usage.

There are the following log levels: panic, fatal, error, warn or warning, info, debug and trace. The level could be updated via environment variable LOG_LEVEL.

Toxics manipulate the pipe between the client and upstream. They can be added and removed from proxies using the HTTP api. Each toxic has its own parameters to change how it affects the proxy links.

For documentation on implementing custom toxics, see CREATING_TOXICS.md

Add a delay to data on a single proxy stream. The delay is equal to latency +/- jitter.

Toxics are directional: by default a latency toxic is attached to the downstream stream (server -> client). It does not delay the opposite direction unless you also add a latency toxic with stream=upstream. For a symmetric round-trip delay, add latency on both streams.

Attributes:

• latency: time in milliseconds
• jitter: time in milliseconds

Bringing a service down is not technically a toxic in the implementation of Toxiproxy. This is done by POSTing to /proxies/{proxy} and setting the enabled field to false.

Limit a connection to a maximum number of kilobytes per second.

Attributes:

• rate: rate in KB/s

Delay the TCP socket from closing until delay has elapsed.

Attributes:

• delay: time in milliseconds

Stops all data from getting through, and closes the connection after timeout. If timeout is 0, the connection won't close, and data will be dropped until the toxic is removed.

Attributes:

• timeout: time in milliseconds

Simulate TCP RESET (Connection reset by peer) on the connections by closing the stub Input immediately or after a timeout.

Attributes:

• timeout: time in milliseconds

Slices TCP data up into small bits, optionally adding a delay between each sliced "packet".

Attributes:

• average_size: size in bytes of an average packet (must be greater than 0)
• size_variation: variation in bytes of an average packet (must be at least 0 and at most average_size)
• delay: time in microseconds to delay each packet by

Closes connection when transmitted data exceeded limit.

• bytes: number of bytes it should transmit before connection is closed

Randomly drops chunks flowing through the proxy simulating flaky Wi-Fi, mobile, or satellite network conditions.

Attributes:

• loss_rate: probability [0.0-1.0] that a chunk is dropped (default 0.0)
• correlation: extra drop probability when the previous chunk was dropped, modeling burst loss (default 0.0)

All communication with the Toxiproxy daemon from the client happens through the HTTP interface, which is described here.

Toxiproxy listens for HTTP on port 8474.

The API is unauthenticated by default and can create proxies to arbitrary upstreams, so never expose it to an untrusted network. To require a token, start the server with TOXIPROXY_AUTH_TOKEN set; every request must then send Authorization: Bearer <token>. toxiproxy-cli reads the same variable (or --auth-token), and the Go client takes it via Client.AuthToken. The token is required on every endpoint, including /version and /metrics, so health checks and Prometheus scrapers must send the header too (see METRICS.md).

• name: proxy name (string)
• listen: listen address (string)
• upstream: proxy upstream address (string)
• enabled: true/false (defaults to true on creation)

To change a proxy's name, it must be deleted and recreated.

Changing the listen or upstream fields will restart the proxy and drop any active connections.

If listen is specified with a port of 0, toxiproxy will pick an ephemeral port. The listen field in the response will be updated with the actual port.

If you change enabled to false, it will take down the proxy. You can switch it back to true to reenable it.

• name: toxic name (string, defaults to <type>_<stream>)
• type: toxic type (string)
• stream: link direction to affect (defaults to downstream)
• toxicity: probability of the toxic being applied to a link (defaults to 1.0, 100%)
• attributes: a map of toxic-specific attributes

See Toxics for toxic-specific attributes.

The stream direction must be either upstream or downstream. upstream applies the toxic on the client -> server connection, while downstream applies the toxic on the server -> client connection. This can be used to modify requests and responses separately.

All endpoints are JSON.

• GET /proxies - List existing proxies and their toxics
• POST /proxies - Create a new proxy
• POST /populate - Create or replace a list of proxies
• GET /proxies/{proxy} - Show the proxy with all its active toxics
• POST /proxies/{proxy} - Update a proxy's fields
• DELETE /proxies/{proxy} - Delete an existing proxy
• GET /proxies/{proxy}/toxics - List active toxics
• POST /proxies/{proxy}/toxics - Create a new toxic
• GET /proxies/{proxy}/toxics/{toxic} - Get an active toxic's fields
• POST /proxies/{proxy}/toxics/{toxic} - Update an active toxic
• DELETE /proxies/{proxy}/toxics/{toxic} - Remove an active toxic
• POST /reset - Enable all proxies and remove all active toxics
• GET /version - Returns the server version number
• GET /metrics - Returns Prometheus-compatible metrics

Proxies can be added and configured in bulk using the /populate endpoint. This is done by passing a json array of proxies to toxiproxy. If a proxy with the same name already exists, it will be compared to the new proxy and replaced if the upstream and listen address don't match.

A /populate call can be included for example at application start to ensure all required proxies exist. It is safe to make this call several times, since proxies will be untouched as long as their fields are consistent with the new data.

$ toxiproxy-cli create -l localhost:26379 -u localhost:6379 redis
Created new proxy redis
$ toxiproxy-cli list
Name   Listen           Upstream        Enabled  Toxics
======================================================================
redis  127.0.0.1:26379  localhost:6379  enabled  None

Hint: inspect toxics with `toxiproxy-cli inspect <proxyName>`

$ redis-cli -p 26379
127.0.0.1:26379> SET omg pandas
OK
127.0.0.1:26379> GET omg
"pandas"

$ toxiproxy-cli toxic add -t latency -a latency=1000 redis
Added downstream latency toxic 'latency_downstream' on proxy 'redis'

To set more than one attribute, repeat the -a flag, for example -a latency=1000 -a jitter=500.

$ redis-cli -p 26379
127.0.0.1:26379> GET omg
"pandas"
(1.00s)
127.0.0.1:26379> DEL omg
(integer) 1
(1.00s)

$ toxiproxy-cli toxic remove -n latency_downstream redis
Removed toxic 'latency_downstream' on proxy 'redis'

$ redis-cli -p 26379
127.0.0.1:26379> GET omg
(nil)

$ toxiproxy-cli delete redis
Deleted proxy redis

$ redis-cli -p 26379
Could not connect to Redis at 127.0.0.1:26379: Connection refused

Toxiproxy exposes Prometheus-compatible metrics via its HTTP API at /metrics. See METRICS.md for full descriptions

How fast is Toxiproxy? The speed of Toxiproxy depends largely on your hardware, but you can expect a latency of < 100µs when no toxics are enabled. When running with GOMAXPROCS=4 on a Macbook Pro we achieved ~1000MB/s throughput, and as high as 2400MB/s on a higher end desktop. Basically, you can expect Toxiproxy to move data around at least as fast the app you're testing.

Can Toxiproxy do randomized testing? Many of the available toxics can be configured to have randomness, such as jitter in the latency toxic. There is also a global toxicity parameter that specifies the percentage of connections a toxic will affect. This is most useful for things like the timeout toxic, which would allow X% of connections to timeout.

I am not seeing my Toxiproxy actions reflected for MySQL. MySQL will prefer the local Unix domain socket for some clients, no matter which port you pass it if the host is set to localhost. Configure your MySQL server to not create a socket, and use 127.0.0.1 as the host. Remember to remove the old socket after you restart the server.

Toxiproxy causes intermittent connection failures. Use ports outside the ephemeral port range to avoid random port conflicts. It's 32,768 to 61,000 on Linux by default, see /proc/sys/net/ipv4/ip_local_port_range.

Should I run a Toxiproxy for each application? No, we recommend using the same Toxiproxy for all applications. To distinguish between services we recommend naming your proxies with the scheme: <app>_<env>_<data store>_<shard>. For example, shopify_test_redis_master or shopify_development_mysql_1.

Use Go 1.27 and golangci-lint v2.14.0 (built with Go 1.27 or newer), matching CI. Download the linter from its v2.14.0 release and ensure both tools are in your $PATH before running the commands below.

• make. Build a toxiproxy development binary for the current platform.
• make all. Build Toxiproxy binaries and packages for all platforms. Requires to have Go compiled with cross compilation enabled on Linux and Darwin (amd64) as well as goreleaser in your $PATH to build binaries the Linux package.
• make test. Run the Toxiproxy tests.

See RELEASE.md