Schedule one-off or periodic jobs without unit files using systemd-run transient timers

Documented steps

1. One-shot delayed job: `systemd-run --on-active=30m --unit=warmup /usr/local/bin/warmup-cache` — creates transient warmup.timer + warmup.service in /run (gone after reboot).
2. Absolute time: `systemd-run --on-calendar='2026-09-08 22:00' --unit=backup-once /usr/local/bin/backup`. Recurring: `--on-calendar='Mon..Fri 09:00'` or shorthand like `--on-calendar=hourly`.
3. Other relative triggers: --on-boot=, --on-startup=, --on-unit-active= (since the unit last became active — gives fixed-interval repetition when combined with a service), plus --timer-property=AccuracySec=1s or Persistent=true pass raw [Timer] properties.
4. Validate calendar expressions first: `systemd-analyze calendar 'Mon..Fri 09:00' --iterations=3` prints the next elapses.
5. Inspect: `systemctl list-timers --all` shows NEXT/LEFT/LAST for every timer; logs land in `journalctl -u warmup.service`.
6. Cancel before it fires: `systemctl stop warmup.timer`; a running job: `systemctl stop warmup.service`. Transient units clean themselves up once finished/stopped.
7. Run with resource limits or as another user the same way as services: `systemd-run --on-active=5m -p MemoryMax=256M -p User=deploy --unit=job1 /path/script` — all service properties are accepted via -p.
8. For per-user timers that survive logout, add --user (with loginctl enable-linger set) — otherwise user transient units die with the session.