websocket / subscriptions support?

[profusion]

websocket / subscriptions support?

[metachris]

I'd also be very interested in subscription support! Going to need it one way or the other :D

[barbieri]

Not yet, @netogallo seems to be working on it, as per #23.

I'm busy with some work-related stuff, but as soon as things calm down I'd like to implement that on top of aiohttp, also adding the async endpoint.

[ecthiender]

I just published a python GraphQL client library supporting subscriptions over websocket. Check it out here. Contributions and comments are most welcome.

pip package
Source

[josephlbarnett]

The referenced PR implements a synchronous websocket endpoint that returns all the data objects after completion. Async may make more sense for some, but this fits our use case at the moment.

[barbieri]

hi all, the PR #55 is great and about to be merged, but I have a question to users (asked there, but you can also answer here):

Is it better to accumulate a list during the whole session, until completed message arrives? Or should we progressively yield result as each message is processed? In my mind the yield would be required by subscription usage, since you're progressively receiving more data you want to process immediately, not at the end.

On a related question: should we create the websocket internally and execute one single query as it's implemented in #55, or should we create it in 2 phases:

• WebSocket wrapper: connects, does the id generation and dispatching of messages ("observer")
• WebSocket endpoint, that takes the wrapper above and issue one single query by queueing it to the wrapper, then when the replies arrive it will yield the results to each endpoint.

This would not be required for "app with one subscription" (as it would be the same), but if we do have multiple subscriptions, or want to interleave multiple queries/mutations/subscriptions in the same channel (saving the connection time for each), it would be required.

Move to yield is reasonably simpler... but to interleave multiple operations would be more work.

[barbieri]

merged with single query support. We can extend it to support multi query later.