We got a support case where the system was bogged down with massive amount of connections in TIME_WAIT state. The cause seems to be the session PulseAudio server that goes bonkers and spams connections (similar to some misbehaving PulseAudio clients). This happens when PulseAudio on the client has died, but the ssh tunnel is still open.
The problem is that our poll logic for reconnect sees the port as open and keeps trying to connect to it. We need to tweak things a bit so we get a delay between attempts even if the port is open.
This is the triggering cause of bug 6125.
If I kill the pulseaudio process on the client side I always get 100% CPU usage from pulseaudio on the server. Verified using these servers:
* Mate and GNOME on both CentOS 7.6 with ThinLinc 4.9.0
* Mate, XFCE and GNOME on Fedora 29 with ThinLinc nightly server
This seems very brittle.
Moving to --- for discussion.