WSS breaking is surprisingly common for clusters of our enterprise users, to the order of 5-10%. Specifically, when business users connect through security networks like Zscaler, often their employers will MITM all connections (similar to how AdGuard works) but in a way that breaks WSS. We have rigorous monitoring for both frontend and backend, and can trace these failures with accuracy - consistently, both the frontend and the ingress firewall think that the other one cancelled the connection attempt, so some network hop in-between did that. Every other network connection during that time works as expected (long polling/SSE over H2), but WSS get interrupted and sometimes can't reconnect. We still love the aliveness of web sockets, but have to architect data fetching in a way that doesn't depend on them working.