[lxc-users] System daemons don't start inside containers after some number of containers was created

Ivan Kurnosov zerkms at zerkms.ru
Fri Jul 21 03:44:38 UTC 2017


Hi,

I have a very strange situation and I'm not even sure if it's the right
mail list to post, but let's see.

I have successfully reproduced it on 2 machines: on a real hardware and in
virtualbox vm.

Given, the host OS Ubuntu 16.04.2 (will all updates), I'm creating a bunch
of containers. To reproduce the problem I'm using the following script:

lxc launch ubuntu:16.04 container-1
sleep 5
lxc exec container-1 ps ax

lxc launch ubuntu:16.04 container-2
sleep 5
lxc exec container-2 ps ax

lxc launch ubuntu:16.04 container-3
sleep 5
lxc exec container-3 ps ax
....

etc up to container-30.

At some point (different on both machines, but consistent) the container is
created, started but has /sbin/init as it ONLY running process. So no
systemd or any other system daemon running there apart of /sbin/init. And
after that point every other container I create looks broken in the very
same way.

Here is how the output from the script above looks on the edge between
"proper containers" and "broken containers":

Creating container-12
Starting container-12
  PID TTY      STAT   TIME COMMAND
    1 ?        Ss     0:00 /sbin/init
   53 ?        Ss     0:00 /lib/systemd/systemd-udevd
   57 ?        Ss     0:00 /lib/systemd/systemd-journald
  236 ?        Ss     0:00 /sbin/dhclient -1 -v -pf /run/dhclient.eth0.pid
-lf /var/lib/dhcp/dhclient.eth0.leases -I -df
/var/lib/dhcp/dhclient6.eth0.leases eth0
  292 ?        Rs     0:00 /usr/bin/python3 /usr/bin/cloud-init init
  295 ?        S      0:00 /bin/sh -c tee -a /var/log/cloud-init-output.log
  296 ?        S      0:00 tee -a /var/log/cloud-init-output.log
  300 ?        Rs+    0:00 ps ax
Creating container-13
Starting container-13
  PID TTY      STAT   TIME COMMAND
    1 ?        Ss     0:00 /sbin/init
  221 ?        Rs+    0:00 ps ax

All the containers after container-13 are created "broken". If I create
another container now - it also would not start properly.

But if I only leave 11 containers running and create another one or restart
one of "broken" ones - then it would start fine.

It only is reproducible with ubuntu 16.04 containers, 17.04 run fine (at
least up to 30 simultaneously running containers).

The number of containers the OS "allows" to run properly is different on
both of machines I tried (12 on the real hardware, and 20 in the virtualbox
vm).
There is plenty of memory available, so memory is not a problem.

There is nothing particularly interesting in the host machine's syslog or
lxd.log.

And in the container there are no logs to read actually, since journalctl
and rsyslog weren't even started.


Any suggestions on where I could dig further?

-- 
With best regards, Ivan Kurnosov
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.linuxcontainers.org/pipermail/lxc-users/attachments/20170721/36e314ca/attachment.html>


More information about the lxc-users mailing list