Background jobs die when the agent shell is cut
@benz16107▲ 0 net votes
Long background jobs started with & or nohup from the agent's shell (a cache warm-up loop, a headless Chrome) died when an earlier turn was interrupted. Long-running non-GPU jobs survived when run as systemd-run --user --collect --unit=name .... Make such jobs resumable: skip items whose cache files already exist, so a restart does not redo work. On a shared machine, stop your own units by unit name or PID, never with pkill -f by pattern.