Skip to content


Subversion checkout URL

You can clone with
Download ZIP
Fetching latest commit...
Cannot retrieve the latest commit at this time.
Failed to load latest commit information.

Testing this out

Just run

node runObservableFromStream


node runStreamFromObservable

Both test-data streams are set to end after 3000 ms

Some things to think about

Rx is push-based. Even cold observables are push based once they are subscribed to, the way I understand it. Node.js streams are pull-based, meaning that they decide when they want to pull data from upstream sources. If data is produced at a slower pace than it is consumed, this does not cause any issues. However, if data is consumed at a slower rate than it it produced, data will either have to be buffered or it will be lost. Node.js streams does provide buffering, and this is probably the only way around that issue. Buffer size is currently standard size, but the parameters to increase buffer size should be exposed.

Also, I will try to see how the Subject (IObservable and IObserver) mathces transform streams.

And it's time to look into merging observables to create a merged stream..

From Rx to node streams and back

This is just work in progress. My plan is to do use-case by use-case until my brain explodes. The simplest ones are ok, the harder use-cases might require some thinking, and the two different streams implementations might turn out to have inconsistencies that will be hard to smooth out without creating leaky abstractions.

My first attempt was based on previous work(hacks) I had seen on the so-called old-streams in node.js (previous to node.js v0.10). They basically used the fact that sreams are EventEmitters and the way rxjs works with events. The new-streams, or streams2, are a little bit higher-level to provide built-in support for buffering and pausing of streams.

My approach is trying to combine rx with node-streams based on using the new streams API.

The plan is to see how far I can go in merging rx and node-streams with using the APIs the expose properly and not listening to internal data-events atc. The low-level approach by listening to events makes node-streams fall back into old-mode and thus breaks buffering and pausing.

Some things should be simple, such as wrapping rx in streams and the other way around. Merging streams, such as combinelatest, is probably harder. That will have to come later, after the first simples use-cases are implemented.

Observables to node streams

Implemented proof of concept for transforming rx-obsverable to node stream. It works by giving the StreamFromObservable constructor an Rx Observable


var stream = new StreamFromObservable(obs);


  • Lot's of stuff
  • Tests

Node-streams to observables

Again a proof of concept for transforming node-streams to observables. Instead of the hack of listening to data-events (Listening for data-events turns streams into old mode.). Instead it listens for readable events, then it reads the stream and calls onNext. It only works on streams in objectmode.


var obs = new ObservableFromStream(stream);

This turns a readable stream into an observable so that you for example can do:

obs.where(function(navdata) { 
     return navdata && navdata.demo && navdata.demo.altitudeMeters;})
  .select(function(navdata) { 
     return navdata.demo.altitudeMeters;})
  .subscribe(function (x) {


  • Tests
  • Much more

OnError and OnCompleted

This should map fairly well from rx to streams. End is forwarded in streams, and can be mapped to onCompleted Errors aren't really forwarded. Must possibly be handled manually by listening to error events and then doing whatever error handling you want to do, including calling onError...?

Cold vs hot observables

Hot observables can't be paused. They can be deferred somehow, but not sure if that helps or not, not sure what the defer function does, I've just heard about it. Should at least consider pausing/unsubscribing temporarily from cold observables if downstream buffers are full. Need a deeper understnading of the cold vs. hot stuff.


IObserver interface:

OnCompleted OnError OnNext

Calls the IOBservable's subscribe function with


Subscribe (returns IDisposable): Notifies the provider that an observer is to recieve notifications.


What to do with backpressure and pausing... If that is ignored, then converting back and forth between node.js streams and rx streams are a lot easier. However, the idea is to sort of try to respect node.js way of handling backpressure


Rx works with single events and respects their timing, with schedulers etc. Node.js streams buffers events withtout really respecting time, as they are mainly designed to handle I/O streams, where timing of a chunk really does not matter.

Another idea

See how many Rx.js things I can port over to streams modules. Some has been done already, but make more like map, filter, zip etc. Those that involve time (combinelatest) are probably not so easy to do... Or might not make sense. But some will make some sense.. maybe.

.NET Dataflow

It seems to me that node.js streams are a much better fit, conceptually, with .NET Dataflow. Now, Dataflow does provide a lot more than node.js streams, mainly because of the fact that JavaScript is single threaded this isn't really used in node.js. Dataflow opens up for parallel execution and much more advanced syncronizations.

Dataflow provides backpressure, just like node.js streams.

Dataflow Reactive Extensions
Source-block Readable stream
Propagator Transform stream (readable/writable)
Target blocks Writeable stream

A big difference between Dataflow and node.js streams is that node.js streams are used in the core of node.js. There are some name IO pipes in .NET, maybe those have some equivalents to node.js streams also. Who knows.

Backpressure in dataflow

It seems to be called postpone. You write to a dataflow block by OfferMessage, and the reciever can receive with Accepted, Declined, DecliningPermantently or Postponed. The message can be reserved for future reading, but other targets could also take it. Here it might be some differences from node.js streams. When the target block wants to read the message again, it can call ConsumeMessage or ReleaseReservation.

I am not sure if the buffering and busy-stuff happens automatically, or if a bufferblock needs to be inserted or what. I have to try this stuff out, I guess.

Something went wrong with that request. Please try again.