[SignalR] Move to generic host

[BrennanConroy]

Part of #20964

Easier to review with whitespace turned off

[Tratcher]

Looks good. Still need to fix the ref assembly?

[BrennanConroy]

Still need to fix the ref assembly?

Yeah. And I'm still trying to figure out the issue with close. Wont remove from draft until I've figured that out.

[halter73]

@davidfowl @Tratcher

@BrennanConroy and I learned by investigating test failures related to this change that unlike WebHost.Dispose(), the generic Host.Dispose() just disposes all services.

Even when configured with ConfigureWebHost, Host.Dispose() does not fire ApplicationStopping, ApplicationStopped or call IServer.StopAsync(). Instead KestrelServer just gets disposed which completely bypasses graceful shutdown and aborts all connections. Naturally, this lead to DiagnosticMemoryPool errors in the tests.

I think we should make Host.Dispose() gracefully shutdown the server like WebHost.Dispose() did. Should I file an issue for this?

[Tratcher]

Feel free to open an issue for triage, it's out of scope for this PR.

I don't think Dispose should be graceful, that's why we added Stop and helpers like Run that call Stop and Dispose for you. I think this was a conscious choice back in 2.2 or 3.0, if not widely discussed.

[halter73]

I was in no way suggesting we change the generic host behavior in this PR. I'm just gauging whether it's worthwhile to even create an issue. It doesn't sound like we're too concerned. Having Dispose() be abortive does make it less sync-over-async which is nice.

[BrennanConroy]

I don't think I can merge this :/

For some reason it makes the issue at dotnet/runtime#30056 happen occasionally and we're trying to reduce overall flakiness, not introduce more!

[BrennanConroy]

Ran this multiple times on Mac after changing to new HostBuilder() and haven't seen the issue.

[Tratcher]