Longpolling creates unexpected Error

[PJUllrich]

Longpolling creates unexpected Error

[mtrudel]

The fallback to long polling is most likely related to being on the latest Phoenix; there's an improvement that shipped in 1.7.11 that changes how sockets fail back to long polling in certain error cases; see the Phoenix changelog for details.

In terms of those errors, they're something we know about on long poll connections. They're benign and 'expected'; we're just not quite sure how to handle them yet.

The data point that you're seeing them a lot is helpful though; do you have a rough idea of how the volume of those messages correlates to site traffic? Like percentage wise or whatever other measures you may have? It'll help deciding how acute of an issue this may be. Thanks!

[PJUllrich]

Thanks Mat, glad you’re ahead of this already :) Unfortunately, I don’t have these numbers but my app doesn’t have many visitors (mostly just me). I experience it myself that LiveView falls back to long polling on every second or third page view. So, I navigate to LV1, it uses a websocket. I navigate to LV2, it uses long polling. I refresh the page, it uses a Websocket now. I can’t spot why it does that though. Let me know if I can help you in any way.

[mtrudel]

Should be fixed in 1.2.1, just released. Logging these messages just isn't worth the support burden of having to answer this same question for the next thousand years. There's a log_unknown_messages config option in case people want the previous (verbose) behaviour.

Thanks for the issue!