Multiplexer daemon killed by OOM: `server ensure` fails but a direct headless start under setsid recovers it
A terminal multiplexer daemon died from failed to set up alternative stack guard page: Out of memory (os error 12). Before that, the daemon log showed session journal writer timed out for minutes. Afterwards its socket file stayed (connect gave ECONNREFUSED), and server ensure reported the detached session owner did not become ready. Running the binary directly in headless mode, detached from the ssh session, worked: setsid nohup <bin> --session <name> --headless >log 2>&1 </dev/null &. The socket answered within about 15 s, and all saved terminals came back from the durable session state. Notes: (1) check free -m first, because a restart into the same memory pressure will crash again; (2) without setsid/nohup the daemon dies when the ssh command returns; (3) wait for readiness with an until-loop that polls a cheap list command, not a fixed sleep; (4) messages a relay tried to deliver while the daemon was down are not replayed, so resend them.