When a client logs in again with the same run ID, frps replaces the old
control and closes its control stream, but the yamux session that carried it
stays open until keepalive reclaims it. The replaced control's work
connections and streams share that session, so they are left behind for up to
tcpMuxKeepaliveInterval plus the ping budget.
A mux session carries exactly one control connection, so the control now
records its session and closes it from the single teardown point shared by
replacement, fatal exit and shutdown. QUIC and non-multiplexed transports keep
the previous path.
Closing the session ends every stream that still belongs to the replaced
control. Those proxies are already being torn down, so the cost is that
in-flight connections of the old control are interrupted immediately instead
of at the keepalive deadline. A replacement control is not affected, and the
ownership rules for a still-online control do not change.
Tests: replacement and fatal exit close the session while the replacement
control keeps working, and concurrent logins with the same run ID leave
exactly one live session.
Use net/http Server.Protocols and the merged golib listener path instead of the deprecated h2c handler. Keep HTTP/1.1 and cleartext HTTP/2 prior-knowledge support, while intentionally dropping HTTP/1.1 Upgrade: h2c.
fatedier