The centralized journal daemon and Desktop Bus broker with a single Desktop Bus are two of the major weaknesses in the systemd architecture, sources of problems as varied as log floods affecting unrelated services, journald (originally) taking everything with it when it died (a failure to understand the lessons of daemontools, also explaining why originally every service was proofed against SIGPIPE), and the Desktop Bus Death Rattle.
One does not need SystemTap to see the mount unit problems, never fear. As the article notes, near the bottom, the behaviour is documented in the systemd.mount manual page; and one can look at the Desktop Bus traffic with busctl monitor.
But again, this is a failure to learn the lessons of previous softwares; in particular the lessons of the many softwares that do/did data updates over limited bandwidth channels. From everything, from redrawing the screen in a full-screen TUI program like vi, through sending the Fidonet nodelist around the world, to doing quick tape backups, one learns the lesson of sending just the diffs. As the author noted, it seems daft for systemd-logind to be told, again, all about the 1999 of the 2000 mount points that have not changed.
The sad thing is that systemd-logind isn't even interested in mount units. It registers an interest in all unit properties change announcements, but is only really interested in the properties of session and "user" slice units.
* http://jdebp.uk./FGA/dbus-death-rattle.html
One does not need SystemTap to see the mount unit problems, never fear. As the article notes, near the bottom, the behaviour is documented in the systemd.mount manual page; and one can look at the Desktop Bus traffic with busctl monitor.
But again, this is a failure to learn the lessons of previous softwares; in particular the lessons of the many softwares that do/did data updates over limited bandwidth channels. From everything, from redrawing the screen in a full-screen TUI program like vi, through sending the Fidonet nodelist around the world, to doing quick tape backups, one learns the lesson of sending just the diffs. As the author noted, it seems daft for systemd-logind to be told, again, all about the 1999 of the 2000 mount points that have not changed.
The sad thing is that systemd-logind isn't even interested in mount units. It registers an interest in all unit properties change announcements, but is only really interested in the properties of session and "user" slice units.