Fix journalctl leak

[mtaufen]

This fixes the journalctl leak that occurs when a process that uses
cadvisor exits. See issues #1725 and kubernetes/kubernetes#34965.

I tested this by changing the Kubelet configuration 100 times (which causes 100 restarts of cadvisor). After this, there were no leaked inotify nodes:

root@test-cos-beta-61-9765-31-0:~# for foo in /proc/*/fd/*; do readlink -f $foo; done | grep inotify | sort | uniq -c | sort -nr | awk '{print $2}' | cut -d/ -f 3 | xargs ps -p
  PID TTY      STAT   TIME COMMAND
    1 ?        Ss     0:01 /usr/lib/systemd/systemd noresume noswap cros_efi
  383 ?        Ss     0:00 /usr/lib/systemd/systemd-udevd
  516 ?        Ssl    0:00 /usr/lib/systemd/systemd-timesyncd
  579 ?        Ss     0:00 /usr/bin/dbus-daemon --system --address=systemd: --nofork --nopidfile --systemd-activation
  617 ?        Ss     0:00 /usr/lib/systemd/systemd-resolved
  631 ?        Ss+    0:00 /sbin/agetty --keep-baud 115200,38400,9600 ttyS0 vt220

whereas before there were many journalctl processes associated with leaked nodes.

[dashpole]

thanks a bunch