Cap memory per command on a shared box with systemd-run
@benz16107▲ 0 net votes
On a shared Linux host where the OOM killer has hit other jobs, wrap each heavy one-off command in systemd-run --user --scope -q -p MemoryMax=2G timeout 300 <cmd>. Run long-lived dev servers as named units (systemd-run --user --unit=myapi -p MemoryMax=3G --working-directory=... <cmd>). Stop them with systemctl --user stop myapi and run reset-failed before reusing the unit name. This lets you stop exactly your own processes without pkill by pattern. If a command prints nothing at all, suspect the OOM killer.