Preparing the Environment
There are no dependencies to install Railgun. The only external requirement is a Memcached instance for Railgun to use as a cache backend. However, you should ensure that the latest CA certificate bundle is installed for your Operating System as it will be used to connect securely back to CloudFlare during activation. CentOS, Red Hat, Debian, and Ubuntu users should make sure that the ca-certificates package is installed and up-to-date via yum or apt.
Users of mod_cloudflare, an Apache module which displays a visitor’s true source IP, should update their Apache configuration to include the IP of their Railgun instance as a trusted proxy. If Apache and Railgun run on the same server, the following line is needed within your Apache configuration (typically, /etc/apache2/httpd.conf or /etc/httpd/httpd.conf):
If Railgun and Apache are on two separate machines, configure mod_cloudflare to use the source IP of Railgun. This may be on a NATed address or the public IP, depending on your network configuration.
Railgun runs on port 2408 via TCP by default and this port will need to be open to connections from CloudFlare IPs. If you are unfamiliar with networking, please reach out to your hosting provider to determine the proper way to open the port for your environment. You can script out adding rules for software firewalls as follows (use with caution):
# for i in `curl https://www.cloudflare.com/ips-v4`; do ufw allow proto tcp from $i to any port 2408; done
# for i in `curl https://www.cloudflare.com/ips-v4`; do iptables -I INPUT -p tcp -s $i --dport 2408 -j ACCEPT; done
For users with very restrictive firewall egress (outbound) policies, outbound TCP port 443 will also need to be allowed for Railgun activation to function properly. In addition, you will need to allow Railgun to make outbound connections to your web server on any ports that it listens on.
Railgun is available only for 64-bit systems.
The best way of installing Railgun is via the Package Repository. Run the commands mentioned on the Package Repository page in order to set this up. Debian, CentOS, Red Hat Enterprise Linux, and Amazon Linux users may utilize the Railgun binary package repository. Ubuntu and Debian users will be automatically subscribed to the APT repo after installing one of the packages linked below. Other users will need to install the cloudflare-release RPM as described on the repository home page. Installation of that RPM will install the repository and GPG key.
Once this has been set up, run the following command as root:
# yum install railgun-stable
# apt-get install railgun-stable
Installing the package will automatically add /etc/init.d/railgun to be started on boot on GNU/Linux.
Configuration and activation
Railgun is configured in /etc/railgun/railgun.conf (GNU/Linux). The most important directives are wan.port, which specifies the port Railgun listens on, memcached.servers, a space separated list of host:port memcached instances for Railgun to utilize. It is also possible to specify the full path to a single socket file for memcached. Using a socket file is recommended for best performance if your memcached does not require network communication. Make sure at least one memcached instance is defined. Full details on the available options can be found here.
If you will be using Railgun in combination with a website hosted behind the same NAT scheme, then you will also need to configure a static IP mapping. In other words, if your website and Railgun node are running behind the same router or firewall, there are extra configuration steps necessary to properly route requests. More details in the Known Issues section below. Please contact support for assistance, or you can use the examples in /etc/railgun/railgun-nat.conf to define one yourself.
Railgun requires an activation step in order to use. To activate, you will need to update the configuration file with the external IP address or a hostname which resolves to the external IP of your Railgun instance for activation.public_ip and your activation token for activation.token. You can determine the external IP address of a server with the following command:
# curl icanhazip.com
By default, Railgun will listen on all interfaces, although it is good practice to specify the interface you would like Railgun to use. You can do that by setting the wan.ip option to the IP address of the interface you would like Railgun to listen on. This is especially important if you have a local link IPv6 address, as Railgun will prefer that to an IPv4 address and may not work as a result.
Business and Enterprise users can find the Railgun activation token listed on the Account Settings page. Details on this page can be found in the Administration documentation. Optimized Partners must use the init API method to acquire an activation token (rtkn). Business and Enterprise users should add a Railgun on the listing page, if you haven’t already. It can be named anything you like. After adding a Railgun, the token will be listed on the page and you can toggle the Railgun to ‘On.’ The token is a 32 character hash of numbers and letters. Copy this hash into the activation.token option in your Railgun configuration file. If your public IP and activation token are properly configured, the first time you start Railgun it will register and download a newly generated SSL certificate to encrypt traffic. To make sure that activation works properly, check your logs for a successful activation response when starting Railgun:
# tail -f /var/log/messages
Oct 27 22:29:41 www rg-listener: [Activation] Activation POST completed.
Oct 27 22:29:41 www rg-listener: [Activation] Assigned Railgun ID: 1
Oct 27 22:29:41 www rg-listener: [Activation] Acquired cert from server
If there is a certificate error in the system log or during startup, ensure that the necessary certificate bundle is installed as described in Preparing the Environment above so that Railgun may connect securely back to CloudFlare’s activation servers.
The newly generated certificate is signed against the port2408.net domain and our system will automatically set your new port2408.net subdomain to resolve to the IP specified by activation.public_ip. You can use this hostname to test Railgun without enabling the service for all users. More details are in the Testing Railgun section below.
Starting the service
You should start the service in the manner that your operating system recommends. This is likely either by using the init script directly, or by invoking the service command:
# /etc/init.d/railgun start
Starting railgun: [ OK ]
# service railgun start
Starting railgun: [ OK ]
After running the init script, there should be an rg-listener process listening on the configured port (2408 by default). You can use netstat to make sure:
# netstat -plnt | grep 2408
tcp 0 0 :::2408 :::* LISTEN 2981/rg-listener
Provided that your port is open to CloudFlare traffic, it’s time for testing. If you don’t see the process running, then there may be an issue with activation or your configuration. The Railgun panic log, which is written to /var/log/railgun/panic.log by default, may contain more information. You can also attempt to start Railgun without the init script to see if any errors are present at start:
# sudo -u railgun /usr/bin/rg-listener -config=/etc/railgun/railgun.conf
You should be familiar either with a utility like cURL (available on all major operating systems) or with a browser extension that allows you to set custom HTTP headers like TamperData on Firefox or HTTP Headers for Chrome. If you are not familiar with these utilities, we can help you test Railgun to ensure things are functioning.
The following testing method will only affect your requests. Your other users will not be impacted in any way. To cause a web request to CloudFlare to use Railgun, you set the cf-setopt-wan-id HTTP header to contain the port2408.net hostname and port 2408. Business and Enterprise can determine this hostname in the Railgun section of the Speed Settings page for your domain. Select the desired Railgun from the dropdown and then click ‘Test’ to reveal the hostname. It is not necessary to turn the Railgun toggle to On. Optimized Partners can determine the port2408.net hostname of a Railgun using the conn_set method with mode set to 0 or the host_get_all method.
Once you have the assigned port2408.net hostname, you can test with curl as follows:
# curl -o /dev/null -iv -H "cf-setopt-wan-id: rg-d65dffddfddia75fbbde99a7lleede90.port2408.net:2408" www.example.com
> GET / HTTP/1.1
> User-Agent: curl/7.21.4 (universal-apple-darwin11.0) libcurl/7.21.4 OpenSSL/0.9.8r zlib/1.2.5
> Host: www.example.com
> Accept: */*
< HTTP/1.1 200 OK
< Server: cloudflare-nginx
< Date: Mon, 25 Jun 2012 17:30:59 GMT
< Transfer-Encoding: chunked
< Content-Type: text/html
< Connection: keep-alive
< Cf-Railgun: baa8902ccd 0000 normal eece
< Last-Modified: Thu, 31 May 2012 18:50:00 GMT
< CF-RAY: 46bdbbbd12b0b2d
< Set-Cookie: __cfduid=d19d3919cb9af8538a2053ac233a4e8fa1340645459; expires=Mon, 23-Dec-2019 23:50:00 GMT; path=/; domain=.example.com
< Set-Cookie: __cfduid=d19d3919cb9af8538a2053ac233a4e8fa1340645459; expires=Mon, 23-Dec-2019 23:50:00 GMT; path=/; domain=.www.example.com
HTTP header browser extension users need to add a header with a name of cf-setopt-wan-id and a value of the port2408.net hostname with the Railgun port appended (e.g., rg-d65dffddfddia75fbbde99a7lleede90.port2408.net:2408) when visiting your domain. Consult your extension’s documentation for details on how to do so.
The response should contain a CF-Railgun HTTP header. If this header contains direct, then we were unable to connect to your Railgun instance and fell-back to a direct HTTP request. If the header is set to normal or if it contains numerical values, then Railgun successfully handled the request. The log file can also be used to monitor results. By default, we log only errors. For testing, you need to raise the log.level option from 0 to 5 in your Railgun configuration and then restart the service. You can then tail the logs to watch the requests being processed:
# tail -f /var/log/messages
Oct 27 23:35:57 www railgun[18.104.22.168:22114]: Handling new WAN connection
Oct 27 23:35:57 www railgun[22.214.171.124:22114]: Rx [0000000000... FnPing 2012-10-27 23:37:55.872617 +0000 UTC]
Oct 27 23:35:57 www railgun[126.96.36.199:22114]: Tx [0000000000... FnPong 2012-10-27 23:37:55.872617 +0000 UTC]
Oct 27 23:35:57 www railgun[188.8.131.52:22114]: Transmit time: 550us
Oct 27 23:36:06 www railgun[184.108.40.206:22114]: Rx [ab18927f79... FnBegin ab18927f79... FnRequest HEAD http://example.com/ HTTP/1.1 220.127.116.11 ab18927f79... FnData DENone 374 ab18927f79... FnPush]
Oct 27 23:36:06 www railgun[18.104.22.168:22114]: ab18927f79... waiting for request
Oct 27 23:36:06 www railgun[22.214.171.124:22114]: ab18927f79... HEAD http://example.com/ HTTP/1.1 126.96.36.199 
Oct 27 23:36:06 www railgun[188.8.131.52:22114]: ab18927f79... Starting connection to LAN
Oct 27 23:36:06 www railgun[184.108.40.206:22114]: ab18927f79... connected to 220.127.116.11:80
Oct 27 23:36:06 www railgun[18.104.22.168:22114]: ab18927f79... Request sent to server
Oct 27 23:36:06 www railgun[22.214.171.124:22114]: ab18927f79... waiting for response
Oct 27 23:36:06 www railgun[126.96.36.199:22114]: ab18927f79... Response is ready
Oct 27 23:36:06 www railgun[188.8.131.52:22114]: ab18927f79... Received response
Oct 27 23:36:06 www railgun[184.108.40.206:22114]: ab18927f79... Forcing wantedDelta
Oct 27 23:36:06 www railgun[220.127.116.11:22114]: ab18927f79... HTTP/1.1 301 Moved Permanently
Oct 27 23:36:06 www railgun[18.104.22.168:22114]: Tx [ab18927f79... FnData DENone 328 ]
Oct 27 23:36:06 www railgun[22.214.171.124:22114]: ab18927f79... completed (0 body bytes in 0.001572 seconds/0)#012
Oct 27 23:36:06 www railgun[126.96.36.199:22114]: ab18927f79... waiting for request
Oct 27 23:36:06 www railgun[188.8.131.52:22114]: Transmit time: 88us
Oct 27 23:36:06 www railgun[184.108.40.206:22114]: Tx [ab18927f79... FnPush]
Oct 27 23:36:06 www railgun[220.127.116.11:22114]: Transmit time: 48us
Railgun will produce 5xx messages when the rg-listener service is unable to reach the origin web server. Checking that the route between the Railgun server and the web server is clear is essential before contacting support. The easiest way of checking that us just by performing a curl command from the Railgun server to the origin web server. If Railgun and the web server are both run on the same physical server, check that it allows loopback HTTP connections on ports 80 and 443 in its firewall settings. Please ensure the correct port is open and contact support if errors persist.
It is recommended that you test your site using the aforementioned browser extensions and the cf-setopt-wan-id HTTP header set thoroughly before enabling Railgun for all visitors to your site. When you wish to go live, Enterprise and Business users should select the desired Railgun for your domain on the CloudFlare Settings page from the drop-down and then toggle the switch to ‘On.’ Optimized Partners should use the conn_set with mode set to 0 or conn_setmode_enabled method to enable Railgun. Railgun may take up to 5 minutes to fully activate, after which you should see the CF-Railgun HTTP header present in responses from all your active CloudFlare DNS records.
If you notice consistent 523, 524 or other error responses, please check the System Status Map and contact support as needed. Railgun will fall-back to direct HTTP requests if our endpoints can’t contact your Railgun daemon, but consistent error responses may indicate a system or origin server problem. When contacting support, please provide a screenshot of http://www.yourdomain.com/cdn-cgi/trace if you are able, or a traceroute to your domain so we know which datacenter your requests are hitting. You can then pause CloudFlare via the website to disable the service and resume normal website traffic.
Railgun does not perform DNS queries when it receives a request for maximum efficiency and to prevent tampering. This means that the daemon is unaware of NAT routing or firewalls. NAT does not allow for addressing a public interface from within the associated LAN and Railgun requests will timeout and produce 502 errors. This can be corrected by setting up a static IP mapping. You can set that either through the hosts file for your system (usually at the path /etc/hosts) or through the railgun-nat.conf file in the same directory as the railgun.conf file. Please contact support if you require assistance with the NAT configuration file.