Longpoll fallback is preserved after server restart and browser reload

[clarkware]

Longpoll fallback is preserved after server restart and browser reload

[josevalim]

Perhaps we should change the built in generators as well.

[chrismccord]

We still fallback when websocket drops and fails to connect within the failback threshold, but we now only memorize the fallback for the session if the websocket transport never successfully established a connection. This strikes a nice balance between saving the user from moving to a network that perm fails websockets (but their tab previously had a good websocket) and unnecessarily memorizing for the duration of the session.

[mainlymortal]

I have the exact same issue but assumed the default transport was mistakenly changed to longpoll instead of websocket.

A member on the elixir forum also has this issue by the looks of it as they can only connect via websocket by opening a new tab.

https://elixirforum.com/t/liveview-falls-back-to-longpoll-after-dev-server-restart/61736

[SteffenDE]

You could do something like this

longPollFallbackMs: location.host.startsWith("localhost") ? null : 2500,

to only configure longpoll when not in development.

[mainlymortal]

Would this issue still not happen in production though if a server a client is connected to happens to go down for any reason. The client will be stuck on long polling until they open a new browser tab / window.

[clarkware]

That works nicely. It also seems to have resolved a spurious Presence leave event I was getting which I'm guessing was related to memorizing the fallback.

Thanks for the quick fix!