I'll change this to staging after deployment and landing this stack, for now, I preferer this - it's easier for me to work on this.

We should retry here, but I want to avoid unnecessary computation on the client right now, when the socket most likely will be closed because the clients are not logged in to Identity and auth will fail. Improving this is tracked here.

yes!

As far as I know close and error comes along, but close is later (as a result of error) so it's usually better to handle only close events. link to standard