Feb 27, 2011

Dashboard for enabled services in systemd

Systemd does a good job at monitoring and restarting services. It also keeps track of "failed" services, which you can easily see in systemctl output.

Problem for me is that services that should be running at the machine don't always "fail".
I can stop them and forget to start again, .service file can be broken (like, reload may actually terminate the service), they can be accidentally or deliberately killed or just exit with 0 code due to some internal event, just because they think that's okay to stop now.
Most often such "accidents" seem to happen on boot - some services just perform sanity checks, see that some required path or socket is missing and exit, sometimes with code 0.
As a good sysadmin, you take a peek at systemctl, see no failures there and think "ok, successful reboot, everything is started", and well, it's not, and systemd doesn't reveal that fact.
What's needed here is kinda "dashboard" of what is enabled and thus should be running with clear indication if something is not. Best implementation of such thing I've seen in openrc init system, which comes with baselayout-2 on Gentoo Linux ("unstable" or "~" branch atm, but guess it'll be stabilized one day):
root@damnation:~# rc-status -a
Runlevel: shutdown
  killprocs  [ stopped ]
  savecache  [ stopped ]
  mount-ro   [ stopped ]
Runlevel: single
Runlevel: nonetwork
  local      [ started ]
Runlevel: cryptinit
  rsyslog    [ started ]
  ip6tables  [ started ]
...
  twistd     [ started ]
  local      [ started ]
Runlevel: sysinit
  dmesg      [ started ]
  udev       [ started ]
  devfs      [ started ]
Runlevel: boot
  hwclock    [ started ]
  lvm        [ started ]
...
  wdd        [ started ]
  keymaps    [ started ]
Runlevel: default
  rsyslog    [ started ]
  ip6tables  [ started ]
...
  twistd     [ started ]
  local      [ started ]
Dynamic Runlevel: hotplugged
Dynamic Runlevel: needed
  sysfs      [ started ]
  rpc.pipefs [ started ]
...
  rpcbind    [ started ]
  rpc.idmapd [ started ]
Dynamic Runlevel: manual
Just "grep -v started" and you see everything that's "stopped", "crashed", etc.
I tried to raise issue on systemd-devel, but looks like I'm the only one who cares about it, so I went ahead to write my own tool for the job.
Implementation uses extensive dbus interface provided by systemd to get a set of all the .service units loaded by systemd, then gets "enabled" stuff from symlinks on a filesystem. Latter are easily located in places /{etc,lib}/systemd/system/*/*.service and systemd doesn't seem to keep track of these, using them only at boot-time.
Having some experience using rc-status tool from openrc I also fixed the main annoyance it has - there's no point to show "started" services, ever! I always cared about "not enabled" or "not started" only, and shitload of "started" crap it dumps is just annoying, and has to always be grepped-out.

So, meet the systemd-dashboard tool:

root@damnation:~# systemd-dashboard -h
usage: systemd-dashboard [-h] [-s] [-u] [-n]

Tool to compare the set of enabled systemd services against currently running
ones. If started without parameters, it'll just show all the enabled services
that should be running (Type != oneshot) yet for some reason they aren't.

optional arguments:
 -h, --help  show this help message and exit
 -s, --status Show status report on found services.
 -u, --unknown Show enabled but unknown (not loaded) services.
 -n, --not-enabled Show list of services that are running but are not
   enabled directly.

Simple invocation will show what's not running while it should be:

root@damnation:~# systemd-dashboard
smartd.service
systemd-readahead-replay.service
apache.service

Adding "-s" flag will show what happened there in more detail (by the grace of "systemctl status" command):

root@damnation:~# systemd-dashboard -s

smartd.service - smartd
  Loaded: loaded (/lib64/systemd/system/smartd.service)
  Active: failed since Sun, 27 Feb 2011 11:44:05 +0500; 2s ago
  Process: 16322 ExecStart=/usr/sbin/smartd --no-fork --capabilities (code=killed, signal=KILL)
  CGroup: name=systemd:/system/smartd.service

systemd-readahead-replay.service - Replay Read-Ahead Data
  Loaded: loaded (/lib64/systemd/system/systemd-readahead-replay.service)
  Active: inactive (dead)
  CGroup: name=systemd:/system/systemd-readahead-replay.service

apache.service - apache2
  Loaded: loaded (/lib64/systemd/system/apache.service)
  Active: inactive (dead) since Sun, 27 Feb 2011 11:42:34 +0500; 51s ago
  Process: 16281 ExecStop=/usr/bin/apachectl -k stop (code=exited, status=0/SUCCESS)
  Main PID: 5664 (code=exited, status=0/SUCCESS)
  CGroup: name=systemd:/system/apache.service

Would you've noticed that readahead fails on a remote machine because the kernel is missing fanotify and the service apparently thinks "it's okay not to start" in this case? What about smartd you've killed a while ago and forgot to restart?

And you can check if you forgot to enable something with "-n" flag, which will show all the running stuff that was not explicitly enabled.

Code is under a hundred lines of python with the only dep of dbus-python package. You can grab the initial (probably not updated much, although it's probably finished as it is) version from here or a maintained version from fgtk repo (yes, there's an anonymous login form to pass).

If someone will also find the thing useful, I'd appreciate if you'll raise awareness to the issue within systemd project - I'd rather like to see such functionality in the main package, not hacked-up on ad-hoc basis around it.

Update (+20d): issue was noticed and will probably be addressed in systemd. Yay!

Member of The Internet Defense League