petersulyok/smfc v6.2.0

on GitHub

Added

• New exit_level= parameter in the [Ipmi] section (int, [-1..100]%, default=100). It is the fan level applied to all configured IPMI zones when the service terminates. The special value -1 means "do not change the fan levels", so the zones stay at the last applied level. On X14 motherboards the level is applied and then manual fan control is released, so automatic BMC fan control is restored - see README chapter 1.5 and chapter 6 for the details of this platform difference. This parameter was added while fixing issue #118.

Changed

• docker/Docker.md was renamed to docker/DOCKER.md, so all documentation files of the repository use the same upper-case naming. All references were updated, including the ones in the earlier entries of this changelog, but external links pointing to the old path (e.g. bookmarks) have to be updated manually.
• docker/DOCKER.md documents what happens to the fans when the container is stopped: docker stop sends SIGTERM, so the configured exit_level= is applied.
• The section headings of this changelog use the Keep a Changelog categories consistently now (Added, Changed, Deprecated, Removed, Fixed). The earlier New, New/Added, Fix and Change headings were renamed accordingly in all releases, the entries themselves are unchanged. The same categories are used in the DEB (debian/changelog) and RPM (smfc.spec) changelogs, where the New: bullet prefix became Added:.
• The fan mode is not changed at exit any more. smfc is already running in FULL mode at that point, so the redundant set_fan_mode() call only cost a fan_mode_delay long sleep on every service stop.

Deprecated

• The -ne command-line option. It is still accepted and still means "no fan level change at exit" (it is equivalent to exit_level=-1), but it logs a deprecation warning and it will be removed in a future release.

Fixed

• The documented safe shutdown did not happen on a normal service stop. Fan levels were restored through an atexit handler only, and CPython does not run atexit callbacks when the process is terminated by SIGTERM, which is the default kill signal of systemd. As a result nothing happened on systemctl stop or systemctl restart: the BMC was left in FULL mode at the last applied duty cycle with no component regulating it, which is exactly the state FULL mode is not safe in. smfc installs a SIGTERM handler now, so a normal service stop performs an ordinary interpreter shutdown and the configured exit_level= is applied. If you prefer the previous behavior, set exit_level=-1. See issue #118.
• smfc-client reported the wrong fan level for controllers that lose a shared IPMI zone arbitration. Both the Fan controllers table and the --verbose block showed the level applied to the zone instead of the level the controller itself requested, so a losing controller contradicted its own Window:/Curve: lines (e.g. an NVME controller at 39.9C displaying Level: 74 % while its curve maps that temperature to 35%). Each controller now reports its own request, and when the zone ended up at a different level that value is appended explicitly as (zone N applied: Z %). Controllers on non-shared zones and the winner of a shared zone are unchanged.
• Documentation of smfc-client --verbose: shared=yes was described as meaning that another controller is currently driving the row's IPMI zone. It marks participation in zone arbitration and is reported for every controller on a shared zone, the current winner included.