Some SCSI devices are listed as "tentative" by systemd

SUSE Linux Enterprise Server 15 SP3
SUSE Linux Enterprise Server for SAP Applications 15 SP3

On systems with late versions of systemd, it is possible that some SCSI devices, which are path devices of a dm-multipath device, are shown with state "tentative" although the system is fully booted:

# /bin/systemctl --no-pager --all list-units
dev-sdf.device               loaded    activating tentative /dev/sdf
dev-sdl.device               loaded    activating tentative /dev/sdl
dev-disk-by\x2did-dm\x2duuid\x2dmpath\x2d360002ac000000000000000070002225b.device  loaded    active     plugged   /dev/disk/by-id/dm-uuid-mpath-360002ac000000000000000070002225b

As long as the respective dm-multipath device (last line in the example above under Situation) is shown as "active/plugged", this is not an error; it "works as designed".

This phenomenon can occur for multipath path devices if the multipath daemon ('multipathd.service') is only enabled in the fully booted system, and not in the initial RAM file system (initramfs).
During boot, the SCSI device (e.g. /dev/sdf' in the example above under Situation) will be seen as "active/plugged" by systemd while processing the initramfs, because multipathd is inactive.
After switching the root file system, these devices are not "active" any more, and unusable for systemd, because they will now become members of a multipath device.

When systemd switches to the root file system, it re-executes itself, passing a list of previously detected device units to the re-executed instance by "serializing" and "deserializing" through a mechanism similar to a socket.
Recent systemd releases have changed the logic with which it treats the state of such "deserialized" devices before the devices are re-discovered by udev.

Older systemd releases would keep treating previously "plugged" device as "active/plugged".
This could cause a wrong setup of the storage stack and eventually boot failure for multipath setups.
Assuming that these devices are "inactive/dead" is also wrong; it could lead to other fatal errors, such as unmounting file systems that are still needed.

Therefore systemd was updated,and treats such devices as "tentative" now, until udev re-discovers them and reports them back to systemd. Using "tentative" state has the advantage that the devices will neither be prematurely activated nor deactivated by systemd, avoiding the above-mentioned errors.

For SCSI devices that are multipath members, udev sets the "SYSTEMD_READY=0" device property, and does not report these devices to systemd at all. Therefore they remain in systemd's internal device list in "tentative" state, and will be shown in this state by the `systemctl --all list-units` command.
This has no negative consequences. 
If you are observing this device state just verify that the multipath device, that the respective SCSI devices are members of, is correctly activated.

The behavior changed with a change set that is mentioned in the changelog of the systemd package as follows:

7b47b3c306 core/device: ignore DEVICE_FOUND_UDEV bit on switching root
(bsc#1137373 bsc#1181658 bsc#1194708 bsc#1195157 bsc#1197570)