#rpc-stream

RpcStream is a dead simple (75 loc) rpc system that works over any full-duplex text/byte/json stream. There's also a Python port, e.g. for cross-language RPC-over-SSH.

ports

• rpyc-stream Python

rant

There are a bunch of people have already written node rpc systems, but most of them has done it right yet. My beef is that all these systems is that they are tightly coupled to, or wrapped around servers. what if I want to encrypt the messages? what if I want to send the messages over stdin/out, or over ssh? of write them to a temp file? there is one abstraction that matches all of these cases; the Stream

I should just be able to do this:

  REMOTE_SSH_STREAM           //<-- pipe data from a remote source
    .pipe(DECRYPT_STREAM)    //through some ('middleware') streams (ssh already encrypts, but I'm paranoid)
    .pipe(GUNZIP_STREAM)
    .pipe(RPC)               //<--- pipe the data through the RPC system.
    .pipe(GZIP_STREAM)
    .pipe(ENCYPT_STREAM)
    .pipe(REMOTE_SSH_STREAM)  //<-- and back to the remote

  //with something very similar on the other side.

RPC framework (AHEM! RPC MODULE!), you just worry about calling the right function, I'll decide where you go...

update: dnode@1.0.0 now has first class streams, and you can pipe it where you like!

usage

var rpc = require('rpc-stream')

//create a server, that answers questions.
//pass in functions that may be called remotely.
var server = rpc({hello: function (name, cb) {
  cb(null, 'hello, '+name)
}})

//create a client, that asks questions.
var client = rpc()

//pipe rpc instances together!
client.pipe(server).pipe(client)

var remote = client.wrap(['hello'])
remote.hello('JIM', function (err, mess) {
  if(err) throw err
  console.log(mess)
})

over tcp

server

net.createServer(function(con) {
  // create one server per connection
  var server = rpc(/* ... */)
  server.pipe(con).pipe(server)
}).listen(3000))

client

var client = rpc()
var con = net.connect(3000)
client.pipe(con).pipe(client)

var remote = client.wrap(['hola'])
remote.hola('steve', function(err, res) {
  console.log(res)
})

rpc(methods, opts)

returns a RpcStream that will call methods when written to.

If opts.raw is set to true, JSON.stringify() is turned off and you just get a stream of objects, in case you want to do your own parsing/stringifying.

With opts.flattenError you can override the default method of converting errors to plain js objects. For example, to include non-enumerable properties too, pass:

{flattenError: function (err) {
  if(!(err instanceof Error)) return err
  var err2 = { message: err.message }
  var props = Object.getOwnPropertyNames(err)
  for(var k in err)
    err2[k] = err[k]
  return err2
}}

RpcStream#wrap(methodNames)

returns a wrapped object with the remote's methods. the client needs to already know the names of the methods. accepts a string, and array of strings, or a object. if it's an object, wrap will use the keys as the method names.

//create rpc around the fs module.
var fsrpc = rpc(require('fs'))
//pipe, etc

then, in another process...

var fsrpc = rpc()
//pipe, etc

//wrap, with the right method names.
var remoteFs = fsrpc.wrap(require('fs'))

remoteFs.mkdir('/tmp/whatever', function (err, dir) {
  //yay!  
})

now, the second process can call the fs module in the first process! wrap does not use the methods for anything. it just wants the names.

RpcStream#rpc(name, args, cb)

this gets invoked by wrap. but you could call it directly.

rpc().wrap('hello').hello(name, callback)
//is the same as
rpc().rpc('hello', [name], callback)

RpcStream#pipe

this is why we are here. read this and this

license

MIT/APACHE2