systemd 262-rc1 adds LUOSession= to service units. It helps coordinate supported state across a kexec-based kernel replacement, but does not turn an ordinary service into a transparently live-updatable application.
A handover of supported state
The version-pinned NEWS file says systemd creates LUO sessions and passes them through its file-descriptor store interface. The same release adds reporting for kexec counts and shutdown timestamps.
The kernel LUO documentation describes named sessions and handlers that preserve selected resources, then retrieve them under the new kernel. PID 1 does not simply continue executing unchanged across kexec. The new kernel and userspace must reconstruct access to preserved state.
The diagram marks that boundary explicitly. Preserving data is different from preserving all process execution, every device and every open file.
Memfd support has concrete limits
The memfd preservation documentation warns that preservation is not transparent. Selected properties, including contents and size, survive; other properties reset. FD_CLOEXEC must be set again after restoration, and Hugetlb-backed memfds are not supported by the documented mechanism. The interface remains subject to compatibility changes.
For an application experiment, inventory the resources it needs after the transition. Identify the handler and restoration path for each, then test both a successful handover and an aborted update. Do not infer support for a resource merely because the LUO framework names that class as an intended use case.
Two separate controls for ordinary service starts
The release candidate also introduces RestartRandomizedDelaySec= to spread automatic restarts and ActivatingConcurrencyMax= on slices to limit concurrent activation. They address different problems: synchronized retries versus too many units starting together. Neither implements LUO by itself.
Before adopting a new directive, check the installed systemd version and validate the unit with that version’s tooling in a test environment. For services using Type=notify-reload, the release also tightens signal handling requirements at readiness; review the full compatibility notes before upgrading.
This is a release-candidate review, not an instruction to replace a production kernel. We have not performed a live kernel handover. A useful acceptance criterion is the application’s externally observed pause and correct restored state, including failure handling, rather than the presence of the new unit setting alone.
Sources
- systemd v262-rc1 release
- Version-pinned NEWS and compatibility changes
- Kernel Live Update Orchestrator
- Memfd preservation limitations
Clarified that PID 1 and arbitrary processes do not transparently survive kexec; added actual memfd limitations and separated restart controls from LUO.