systemd Service (Advanced)

A systemd service unit fits in five lines, but the ones deciding production behaviour are almost never among those five. The restart policy, the interval between attempts and the memory and CPU limits are what separate a service that recovers on its own from one that restarts in a loop until it takes the machine down.

Fill in the description, the user, the command and the working directory, choose the restart policy and the limits, and the page assembles the complete unit with its three sections. The policy usually wanted is restarting on any exit rather than only on failure — that way the service comes back even after exiting with code zero by mistake.

The memory limit deserves an explanation: it is enforced by the kernel resource controller, and exceeding it gets the process killed by the out-of-memory mechanism. It is protection against leaks, not a performance setting. The CPU limit, by contrast, is proportional and merely slows the process under contention, without killing it. After writing the unit, systemd's configuration must be reloaded for it to see the changes.