Christopher J. 'CJ' Wright

I think one issue with storing strong references in the pipeline, which is that `del` might not do the right thing. Since the graph keeps binding references to itself deleting...

Ah, yes so working with `__del__` won't work. But an extension of `disconnect` to work with sinks might work?

also xref: https://github.com/python-streamz/streamz/issues/71

Hmm. This is also an issue with `sink_to_list` which returns the list and not the node.

My understanding of the track we're on: Currently we store weak references for everything except `sink` nodes which are tracked in the global sink set. The sink set causes problems...

What problems are you facing? The approach I took with ZMQ was to use a dispatcher which called `source.emit` when new data was sent over.

Here is a short sketch. 1. Have a system which receives data and calls a function when that data is obtained, see [here](https://github.com/bluesky/bluesky/blob/master/bluesky/callbacks/zmq.py#L186) 2. Build a pipeline see streamz docs...

Cool. I guess I need to start playing around more with the dask stuff.

I guess we'd need to implement a `reset` system first :smile:. One could also consider this existing as a utility, which builds the graph of the pipeline and then check...

Can you elaborate on what is gained from the knowledge of the framework? I do have a working implementation of this which I'll try to PR soon.