On Sun, 07 Feb 2016 at 08:35:55 +0100, Jonas Smedegaard wrote: > Afinstallerer dbus (1.8.20-0+deb8u1) ... > Job for dbus.service canceled. > invoke-rc.d: initscript dbus, action "stop" failed.
I have not been able to reproduce exactly this failure mode on a jessie VM. However, "Job for %s canceled" means that the transaction being added conflicts with one that was already in progress; in this case that's probably an attempt to *start* dbus, which can be done on-demand by a connection to its socket. By removing dbus while a script repeatedly connected to the system bus in a loop, I was able to get a jessie system into a similarly bad state. Because systemd and systemctl will use the system dbus-daemon if present, systemctl can also communicate with polkit which also uses the system dbus-daemon, and various other system facilities optionally use D-Bus, it's probably relatively easy to get a situation where something is connecting to the system bus during removal, causing the system bus to be started up again. Please try these packages: https://people.debian.org/~smcv/bug813970/ or just overwrite /var/lib/dpkg/info/dbus.prerm with the one from that directory before attempting to purge or remove dbus (that and the changelog are the only source differences). On Sun, 07 Feb 2016 at 13:43:22 +0100, Jonas Smedegaard wrote: > ...but then reinstalling (to try get you some more enlightening logdata) > caused the install to hang: ... > Udpakker dbus (1.8.20-0+deb8u1) ... > Behandler udløsere for man-db (2.7.0.2-5) ... > Behandler udløsere for systemd (215-17+deb8u3) ... > Sætter dbus (1.8.20-0+deb8u1) op ... If you look in your syslog around this time, did you get messages "Looping too fast. Throttling execution a little." from systemd? If so, then I was able to reproduce this in the same way I described above. I think there's some bug in the way systemd's event loop interacts with failure to connect to D-Bus; this might have been fixed post-jessie. In this situation, you may be able to get systemd unstuck by telling it to re-exec itself: "kill 1". The modified prerm that I've proposed seems to avoid this issue. > a moment ago (before trying to reinstall dbus): > > root@xayide:~# LC_ALL=C systemctl -l status dbus > * dbus.service > Loaded: not-found (Reason: No such file or directory) > Active: active (running) since Sun 2016-02-07 09:19:41 CET; 3h 50min ago > Main PID: 32549 (dbus-daemon) > CGroup: /system.slice/dbus.service > `-32549 /usr/bin/dbus-daemon --system --address=systemd: --nofork > --nopidfile --systemd-activation OK, so dbus has been removed, but not stopped, and is still running. > How do I fetch data from Journal? The most important bits end up in the syslog (if present), but for your information, see journalctl(1) for more general Journal-reading. > Another detail: The version I reported initially was wrong (issue found > on a server running stable, whereas I report using my laptop running > unstable). Bug should instead be tied to version 1.8.20-0+deb8u1 Thanks, I've corrected that. The bug might also be present in later versions, but for now I'm only preparing test packages for jessie, because that's where you were able to reproduce this. S
