Feature request: Make messaging reliable across reconnects (no lost messages) about engine.io

[rauchg]

We ideally would provide the right interfaces to make this pluggable. I like this though, thanks for bringing it up @EugenDueck

[EugenDueck]

Fyi, regarding my comment about "MBWS doesn't support volatile packages" - MBWS's solution to that problem is MBLWS, the lightweight variant of the protocol without the sequence number and resync feature. According to Mark Hapner:

If you have a mix of messages, some requiring once-and-only-once delivery QoS (MBWS) and some requiring best-effort QoS (MBLWS) then you would use a separate socket for each.
The HyBi working group is currently in the process of extending WS to support multiplexing of logical WS sockets over a TCP socket. Once this is available, you can have a single socket that carries both.

[rauchg]

This would be cool to have a as a default with a configurable timeout. For you still want disconnections to be fired and buffers be discarded (or given to the developer?) past a certain threshold.

For example, if a reconnection is established within 15 seconds, to the user/developer we're like "a disconnection never happened".

Although this might be a better fit at the socket.io level, since it would bloat engine.io considerably.

[EugenDueck]

Agree, the buffers should be prevented from growing indefinitely.

A timeout and/or one of the solutions proposed here would do: #45 (comment)

[rauchg]

@EugenDueck I want something like this for socket 1.0/1.0+. Engine.IO will remain WS-like as far as the message reception guarantees.

[EugenDueck]

This is not a high priority feature for me right now, will try to figure out in the future how/if this could be done in the socket.io layer.

[rauchg]

@EugenDueck per your other message, basically by controlling what gets written at all times with the callback, instead of relying on engine.io's internal buffer