node package manager
Share your code. npm Orgs help your team discover, share, and reuse code. Create a free org ยป


ConcussionJS Proxy Services Layer

What is it?

A distributed proxy built to support massive scale with rate limiting capabilities built to tier levels of service and provide extra layers of security.

ConcussionJS Proxy Services Layer is designed to be deployed along with ConcussionJS Core Platform, though it can be deployed independently as well per the instructions below.

Th ConcussionJS proxy is built on Hipache, originally developed at dotCloud, a popular platform-as-a-service, to replace its first-generation routing layer based on a heavily instrumented nginx deployment. It currently serves production traffic for tens of thousands of applications hosted on dotCloud.

Hipache was in turn built on the node-http-proxy library.

Run it!

1. Install it

From the shell:

$ npm install cjs-proxy -g

The '-g' option will make the 'cjs-proxy' bin-script available system-wide (usually linked from '/usr/local/bin')

2. Configuring the server (config.json)

cjs-proxy uses a Redis server to manage its configuration (and to share its state across the multiple workers). You can use the Redis server to change its configuration while it's running or simply check the health state of a backend.

    "server": {
        "accessLog": "/var/log/cjs_proxy_access.log",
        "port": 80,
        "workers": 5,
        "maxSockets": 100,
        "deadBackendTTL": 30,
        "https": {
            "port": 443,
            "key": "/etc/ssl/ssl.key",
            "cert": "/etc/ssl/ssl.crt"
    "redis": {
        "port": 6379,
        "host": ""
  • server.accessLog: location of the Access logs, the format is the same as nginx
  • server.port: Port to listen to (HTTP)
  • server.workers: Number of workers to be spawned (specify at least 1, the master process does not serve any request)
  • server.maxSockets: The maximum number of sockets which can be opened on each backend (per worker)
  • server.deadBackendTTL: The number of seconds a backend is flagged as `dead' before retrying to proxy another request to it
  • server.https: SSL configuration (omit this section to disable HTTPS)
  • redis: Redis configuration (host & port)

3. Spawn the server

By specifying your configuration file:

$ cjs-proxy --config config.json

Or from the shell:

$ cjs-proxy

Or if you use the port 80:

$ sudo cjs-proxy

Managing multiple configuration files:

The default configuration file is config.json. It's possible to have different configuration files named config_<suffix>.json. The suffix is got from an environment variable called SETTINGS_FLAVOR.

For instance, here is how to spawn the server with the config_test.json configuration file in order to run the tests.

$ SETTINGS_FLAVOR=test cjs-proxy

4. Configuring a vhost (redis)

All the configuration is managed through Redis. This makes it possible to update the configuration dynamically and gracefully while the server is running.

It also makes it simple to write configuration adapters. It would be trivial to load a plain text configuration file into Redis (and update it at runtime).

Different configuration adapters will follow, but for the moment you have to provision the Redis manually.

Let's take an example, I want to proxify requests to 2 backends for the hostname The 2 backends IP are and and they serve the HTTP traffic on the port 80.

redis-cli is the standard client tool to talk to Redis from the terminal.

Here are the steps I will follow:

  1. Create the frontend and associate an identifier

     $ redis-cli rpush mywebsite
     (integer) 1

The frontend identifer is mywebsite, it could be anything.

  1. Associate the 2 backends

     $ redis-cli rpush
     (integer) 2
     $ redis-cli rpush
     (integer) 3
  2. Review the configuration

     $ redis-cli lrange 0 -1
     1) "mywebsite"
     2) ""
     3) ""

While the server is running, any of these steps can be re-run without messing up with the traffic.


Load-balancing across multiple backends

As seen in the example above, multiple backends can be attached to a frontend.

All requests coming to the frontend are load-balanced across all healthy backends.

The backend to use for a specific request is determined at random. Subsequent requests coming from the same client won't necessarily be routed to the same backend (since backend selection is purely random).

Dead backend detection

If a backend stops responding, it will be flagged as dead for a configurable amount of time. The dead backend will be temporarily removed from the load-balancing rotation.

Multi-process architecture

To optimize response times and make use of all your available cores, Hipache uses the cluster module (included in NodeJS), and spreads the load across multiple NodeJS processes. A master process is in charge of spawning workers and monitoring them. When a worker dies, the master spawns a new one.

Memory monitoring

The memory footprint of Hipache tends to grow slowly over time, indicating a probable memory leak. A close examination did not turn up any memory leak in Hipache's code itself; but it doesn't prove that there is none. Also, we did not investigate (yet) thoroughly the code of Hipache's external dependencies, so the leaks could be creeping there.

While we profile Hipache's memory to further reduce its footprint, we implemented a memory monitoring system to make sure that memory use doesn't go out of bounds. Each worker monitors its memory usage. If it crosses a given threshold, the worker stops accepting new connections, it lets the current requests complete cleanly, and it stops itself; it is then replaced my a new copy by the master process.

Dynamic configuration

You can alter the configuration stored in Redis at any time. There is no need to restart Hipache, or to signal it that the configuration has changed: Hipache will re-query Redis at each request. Worried about performance? We were, too! And we found out that accessing a local Redis is helluva fast. So fast, that it didn't increase measurably the HTTP request latency!


Hipache supports the WebSocket protocol. It doesn't do any fancy handling for the WebSocket protocol; it relies entirely on the support in NodeJS and node-http-proxy.


If provided with a SSL private key and certificate, Hipache will support SSL connections, for "regular" requests as well as WebSocket upgrades.

Custom HTML error pages

When something wrong happens (e.g., a backend times out), or when a request for an undefined virtual host comes in, Hipache will display an error page. Those error pages can be customized.

Wildcard domains support

When adding virtual hosts in Hipache configuration, you can specify wildcards. E.g., instead (or in addition to) www.example.tld, you can insert *.example.tld. Hipache will look for an exact match first, and then for a wildcard one.

Note that the current implementation only tries to match wildcards against the last two labels of the requested virtual host. What does that mean? If you issue a request for some.thing.example.tld, Hipache will look for *.example.tld in the configuration, but not for *.thing.example.tld. If you want to serve requests for *.thing.example.tld, you will have to setup a wildcard for *.example.tld. It means that you cannot (yet) send requests for *.thing.example.tld and *.stuff.example.tld to different backends.

Active Health-Check

Even though Hipache support passive health checks, it's also possible to run active health checks. This mechanism requires to run an external program, you can find it on the hipache-hchecker project page.

Future improvements

Read the TODO page