Reconnection will almost never work

[SteveSandersonMS]

This comes out of #11793 where I'm trying to pin down what will make people actually successful at handling connection instability and migration of state across servers.

We've invested a lot in making server-side Blazor circuits theoretically capable of reconnecting and continuing after a transient connection loss, but with the current implementation, it gives up permanently if the first auto-reconnect attempt fails. And it continues to show the "Retry?" button which will do nothing if you click it.

This is because in Boot.Server.ts, the flow looks like:

• Disconnection is detected
• After 3 seconds, a timer calls reconnect, which in turn calls connection.start
• connection.start will throw (assuming the network connection hasn't been restored already, and remember, we only waited 3 seconds so it almost certainly hasn't been).
• The exception is caught and passed to unhandledError, which sets a flag renderingFailed = true
• Now all subsequent calls to reconnect just bail out immediately without doing anything because of the renderingFailed flag.

So the timer which calls reconnect every 3 seconds won't do anything, and the "Retry?" button which also calls reconnect won't do anything.

Minimal fixes

As an absolute minimum, assuming we want to ship reconnection as a feature for 3.0, I think we must do the following. This is not a suggestion of the best UX; it's the smallest set of improvements I think we could ship with.

• Don't switch into a "permanently failed" mode just because one reconnection attempt failed if the reason was unknown.
  • We'll need to distinguish exceptions due to network errors from exceptions due to components themselves throwing. Not calling unhandledError from initializeConnection, and instead catching upsteam, might be enough for that.
• Change the "reconnect" UI so that alongside the "Retry" button, it also has a secondary button (or link text) saying "Reload page", which does a full-page reload.

Better fixes

If we wanted better reconnection UX, we should

• Make the "retry" button show some acknowledgement when you click it
  • For example, disable it for 1 second following the click
  • If the manually-initiated retry fails, then change the dialog text to say "Reconnection failed. Try reloading the page if you're unable to reconnect.". It will still have the two options, "[Retry]" and "[Reload page]".
• Explicitly detect connection failures due to "server says unknown circuit ID", and in those cases, auto-reload the whole page since we know reconnection isn't realistically going to work.
  • We need to avoid infinite refresh loops, e.g., put a flag in sessionStorage to say we did an auto-reload, clear that flag the next time a connection is successful, and don't do an auto-reload any time the flag is still there.

I recognize that some of this duplicates #10496 and #9256. They are down as 3.1.0 and backlog respectively, but I'm not convinced this is a shippable feature without addressing at least the "minimal fixes" above.

[mkArtakMSFT]

Let's tackle the Minimal fixes in Preview 9 and handle the rest post 3.0.

[SteveSandersonMS]

Closing this one as done in #12420, with remaining suggestions for future enhancement now moved to #12442.