TCP proxy that also duplicates traffic to a secondary host
I built this because I needed a way to "tap" production traffic and shoot it at a new system to see how it handles load.
npm install -g duplicator
duplicator -f localhost:80 -d localhost:3000 -p 8080 [-i 127.0.0.1]
- forward all traffic to
localhost:80
- duplicate all traffic to
localhost:3000
, ignoring responses - listen on port
8080
- if -i is specified only listen to incoming traffic for ip address 127.0.0.1
Note: the cli automatically uses the cluster API to run several workers to handle connections, and restart workers if they die.
The equivalent to the call above, in node.js:
var duplicator = require('duplicator')
var server = duplicator(function(connection, forward, duplicate) {
forward('localhost:80')
duplicate('localhost:3000')
}).listen(8080, ["127.0.0.1"])
Creates a new net.Server
. When cb
is called on a successful connection, it will receive the connection to the client and two special callback functions, forward
and duplicate
.
duplicator(function(connection, forward, duplicate) {
forward('origin:80')
duplicate(Math.random() < 0.5 ? 'host1:80' : 'host2:80')
})
In this example, we always use origin:80
as the primary server to forward requests to, but we send half the duplicated rquests to host1
, and the other half to host2
.
Forward stream
to host
. stream
defaults to the current connection unless otherwise specified.
duplicator(function(connection, forward, duplicate) {
// forward every connection to localhost:80
forward('localhost:80')
})
Note that it is not safe to forward the same connection multiple times, as the responses from the different servers may interfere with one another. Therefore if you do call forward
multiple times, only the first call will succeed, and subsequent calls will merely duplicate the connection.
Duplicate on average rate
copies of stream
to host
. stream
defaults to the current connection unless otherwise specified, rate
defaults to 1.
Rate is the expected number of copies sent per connection, and is interpreted as follows:
- 0.1 -> send once with 10% probability
- 1 -> send exactly one copy
- 5 -> send out 5 copies
- 2.5 -> send out 2 copies, with a 50% chance of a third
Note that both stream
and rate
are optional, but have sensible defaults. If only two parameters are specified, the second will be interpreted as rate
if it's a number, stream
otherwise.
duplicator(function(connection, forward, duplicate) {
// duplicate 50% of requests to localhost:3000
duplicate('localhost:3000', 0.5)
// duplicate all connections to stdout
duplicate('localhost:3000', process.stdout)
})
A host can be any one of:
- an object as
net.connect
would expect it ({ host: 'google.com', port: 80 }
) - a number representing a port (
80
,3000
), host is implied to belocalhost
- a string representing a host (
localhost:3000
,google.com:80
) - a function that behaves like
net.connect
(function(cb) { net.connect({ port:80 }, cb) }
)
Thanks to @netroy for contributing the cluster support!
If you'd like to contribute, please open a pull request or file an issue.