Phoenix LiveView doesn't handle server restart gracefully

[mtrudel]

Phoenix LiveView doesn't handle server restart gracefully

[derekkraan]

Thanks for the quick turnaround on this, will push that to prod now. Shortest-lived fork I have probably ever had to run in prod.

[mtrudel]

I agonized over this corner of behaviour for awhile when I was working up the first pass at WebSocket support - I've always understood RFC 6455§7.4.1 to prescribe 1001 as the appropriate close code for a server going down. The spec is a bit vague on specifics here though, so it's definitely open to interpretation. Let's look at some prior art, noting that 1000 is specified to 'indicate[] a normal closure, meaning that the purpose for which the connection was established has been fulfilled':

• Cowboy uses 1000 for stop and timeout cases. Setting aside the semantics of the timeout case (which is pretty clearly incorrect), it's arguable that a 1000 is correct for the server shutting down.
• The ws library also uses 1000 for the shutdown case
• Autobahn isn't prescriptive one way or the other

The above combined with the inherent vagueness of the spec makes this seem like a bit of a silly hill to die on; I'll be updating Bandit's behaviour to return 1000 in this case shortly.

[mtrudel]

Fixed in 0.6.8!