Websocket test cases, for the messages lost in the reconnect gap

The reconnect gap is the defect worth building this set around. A socket drops, the client reconnects within a few seconds, and everything published during that window is gone. There is no error, no retry and no indication: the interface reports itself connected and displays state that is simply wrong. Fixing it requires a cursor or sequence number so the client can ask for what it missed, and testing it requires deliberately publishing while disconnected rather than testing on a stable connection where the gap never opens.

Connection state has to be honest for the same reason. A screen showing stale data while looking live is worse than one that admits it is offline, because the user has no reason to refresh. That in turn depends on detecting a half open connection, where the network has gone away without a close frame and the socket will wait indefinitely for messages that will never arrive. Only a heartbeat surfaces it.

Authorisation on sockets is commonly weaker than on the equivalent HTTP endpoints, because the socket feels like an internal detail of a page that already authenticated. It is not: the endpoint is directly reachable, so the handshake needs its own credential check and each channel subscription needs its own authorisation. Two related cases follow. A long lived connection outlives the access token that opened it, so expiry has to be handled on the live socket. And signing out, revoking a session or deactivating an account has to close the socket, otherwise the connection outlives the credential entirely.

Finally, scale changes the failure modes. Two clients on the same channel connected to different server instances must both receive every message, which requires a shared backplane and is the most common defect that appears only after horizontal scaling. A slow client needs backpressure, because an unbounded per client buffer is a memory exhaustion path. And a server restart produces a synchronised reconnect wave from every client at once, which will take the server down again unless the backoff has jitter.