Investigate one-to-one dbus calling about go-systemd HOT 5 CLOSED

Assigning to @squeed for the initial investigation. This may help removing a dbus requirement from some tools expected to run in minimal environments, like runc.

from go-systemd.

This is already supported, but needs to be explicitly requested. If we like, we could change the logic of New() to reflect systemd's logic: connect to the private bus if uid == 0.

from go-systemd.

What guarantee do we have that it's 1:1 with dbus? I'm wary of this because of the discussion in systemd/systemd#3610 (comment)

from go-systemd.

We don't have that guarantee per se. However, all the current systemd tools operate this way - they try kdbus, then private socket, then public bus.
We have some important scenarios where we need to talk to systemd without dbus running, so adding this as a fallback would be really useful.

from go-systemd.

No explicit guarantee, but systemd internally is doing this kind of automatic fallback so it seems to handle all cases transparently without special-casing endpoints on different buses. I think this gave us some confidence margin that there won't be big differences.

Just to be more conservative, second iteration of #208 switched to using the p2p mode only as a fallback if the original dbus connection fails. Sounds more reasonable?

from go-systemd.