Random routing issue (tcp, sni, http/2)

Yes, the symptom is your are connecting to something that belongs on backend X, which works and then the browser needs something from backend Y, but it will reuse the same session that still is going to backend X, because the certificate on backend X covers the hostname that according your rules would actually go to backend Y.

They can’t overlap is the point, this is not necessarily related to wildcard certificates, it’s just easier to hit this issue with them.

The point is there be a hostname a browser would use, for which the certificates in both backends are valid.