RProxy

RProxy is a reverse proxy server written with performance and scale in mind.

Building and installing

Dependencies

Building with all dependencies compiled and statically linked

cd build
cmake -DRPROXY_BUILD_DEPS:STRING=ON ..
make

Build using system-wide dependencies

cd build
cmake ..
make

Configuration

Base Configuration

Before any other sub-section of the configuration is processed, the following options can be set.

daemonize  = false
rootdir    = /tmp
user       = nobody
group      = nobody
max-nofile = 1024

daemonize

The value of this option is a boolean (either true or false). If the value is true, RProxy will daemonize after start, otherwise the server will run in the foreground.
rootdir

If RProxy is configured to daemonize, the service will daemonize into this i directory.
user

Drop permissions to this user once root operations have been executed. The default of this is to run as the current user.
group

Drop permissions to this group once root operations have been executed. The default of this is to run as the current group.
max-nofile

If your system supports set/get rlimits, this sets the maximum number of file-descriptors the server can use at one time. Since RProxy will attempt to keep all of the downstream connections alive, it is suggested that this number be pretty high.

It should be noted that by default, most systems won't allow a user to go over a static number (most of the time 1024) even with setrlimit. In this case, a user must perform system-wide configurations.

On linux you can add the following to the file /etc/security/limits.conf:
```
  "* hard nofile 590000"
```
Note a reboot is required for this setting to take effect.

On OSX, the following command can be run
```
  "sudo launchctl limit maxfiles 590000 590000"
```

Base Server Configuration

RProxy configuration must contain one or more "server" configuration sections. These sections contains all the required information to deal with an incoming request. The base configuration for a server is as follows:

addr            = 127.0.0.1
port            = 8080
threads         = 4
read-timeout    = { 0, 0 }
write-timeout   = { 0, 0 }
pending-timeout = { 0, 0 }
high-watermark  = 0
max-pending     = 0
backlog         = 0

addr

The IP address in which to listen on.
port

The port to listen on
threads

The number of threads to use for this server (note that downstream connections are multiplied by this number).
read-timeout

The timeout in { 'seconds', 'microseconds' } for a client connection who has not sent any data. For example, if the value of this was "{ 1, 0 }", if the client has not been active for 1 second, the connection is closed.

Setting this will evade potential idle connection DoS attacks.
write-timeout

The timeout in { 'seconds', 'microseconds' } to wait for data to the client to be written. If a client is blocking on the read for this long, the connection is closed.
pending-timeout

When a connection is first made to RProxy, it is not immediately processed, instead it is placed in a pending queue. Only when a downstream connection has become available for use does the client get serviced. If a downstream does not become available for this amount of time, the client connection is shut down and removed from the pending queue.

This makes sure both the RProxy service and the downstream services are never overloaded.
max-pending

If there are this many clients waiting for a backend server to be available, the incoming connection will not be accepted.

This setting assures that the server cannot be overloaded with too many connections.
high-watermark

In many cases, a downstream may write faster to the RProxy than it can to the client. If the client is unable to keep up with the speed the backend server is sending, you may experience very high memory consumption. This is called a fast-writer/slow-reader effect.

If the number of bytes in the output buffer to the client goes over this number, RProxy will disable reading data from the backend until all data in the output queue to the client has been sent.

Server::Downstream Configuration

Each server section contains one or more "downstreams" (in other words, a backend service). Each downstream configuration directive is named, which is referenced by a rule (discussed later).

The information contained within the downstream configuration is global to a server; all virtual host and rule will use the downstream information contained within each downstream section.

downstream <NAME> {
    enabled        = true
    addr           = x.x.x.x
    port           = nnnn
    connections    = 4
    high-watermark = 0
    read-timeout   = { 0, 0 }
    write-timeout  = { 0, 0 }
    retry          = { 0, 500000 }
}

NAME

Each downstream must have a unique name.
enabled

If this value is set to false, RProxy will not attempt to make connections or utilize this downstream in any of the rules.
addr

The FQDN or IP address of this downstream server.
port

The listening port of the downstream server.
connections

The number of connections RProxy will attempt to keep available. Note that this number is multiplied by the number of threads configured for the server. For example if you have 4 threads, and 2 connections, RProxy actually maintains 8 connections.
high-watermark

If number of bytes in the sendQ is over the value of this number (in bytes), further reading from the client is disabled until the sendQ has been emptied.

This setting assures that backend servers will not be overloaded by clients.
read-timeout

If no data has been read from this downstream for this many { seconds, microseconds }, the connection is terminated.
write-timeout

If a write request takes over { seconds, microseconds } to happen, the connection is terminated.
retry

If one of the downstream connections has been terminated for some reason, this is the time in { seconds, microseconds } RProxy will wait until it tries to re-establish the connection.

Server::SSL Configuration

When this configuration section is present and enabled, RProxy will treat incoming connections as SSL.

ssl {
    enabled           = false
    protocols-on      = { ALL }
    protocols-off     = { }
    cert              = bleh.crt
    key               = bleh.key
    ca                = blah.ca
    capath            = /capath/
    ciphers           = "ECDHE-RSA-AES128-GCM-SHA256:ECDHE-RSA-AES256-GCM-SHA384:ECDHE-RSA-RC4-SHA:ECDHE-RSA-AES128-SHA:RC4-SHA:RC4-MD5:ECDHE-RSA-AES256-SHA:AES256-SHA:ECDHE-RSA-DES-CBC3-SHA:DES-CBC3-SHA:AES128-SHA"
    verify-peer       = false
    enforce-peer-cert = false
    verify-depth      = 0
    context-timeout   = 172800
    cache-enabled     = true
    cache-timeout     = 1024
    cache-size        = 65535
    crl               = {}
}

enabled

If set to true, SSL is enabled, otherwise SSL is disabled.
cert

The servers SSL cert
key

The servers private SSL key
ca A specific CA file
capath Relative path to search for valid CA's
ciphers

Accepted ciphers
protocols-(on|off)

The SSL options for enabling or disabling SSL specific protocols. Options: SSLv2, SSLv3, TLSv1, or ALL
verify-peer

Enables SSL client peer verification
enforce-peer-cert

If true, a client is rejected if it does not supply a client certificate.
cache-enabled

Enable SSL certificate cache
cache-size

Maximum size of the SSL cache
cache-timeout

The lifetime a cert will be kept in the cache.
context-timeout

Timeout for (OpenSSL >= 1.0) session timeouts.

Server::SSL::CRL Configuration

A server / vhost can be configured to use a CRL list or file which can be reloaded without restarting or signals. This configuration is turned off by default.

  ssl {
      crl {
          file    = "/path/to/crl.pem"
          dir     = "/path/to/crldir/"
          reload  = { 10, 0 }
      }
  }

file

Read from a specific file for CRLs.
dir

Read from a specific directory for CRLs
reload

A timer (in secuds, useconds) to check for changes and reload the CRL configuration if changes have been mae.

Bagarre/RProxy