Wed, Dec 10, 2014 at 03:25:31PM CET, colin(a)trancecode.co.uk wrote:
Hi Jiri
>
> I do not understand why you try to run it by hand and with NM at the
> same time.
>
I know of no other option but to explicitly define all networking
settings which are to be integrated with a docker container.
The pattern is to create a link, then move it into another network namespace.
I certainly cannot rely on (nor would I want to ) NM interfering
inside a container bound process.
Experimentation has revealed that even
[root@localhost scripts]# teamd -d
causes a similar trace in /var/log/messages <attached>
Weird thing is that teamd is started and killed at the very second. I
suspect NM from doing this. Maybe you have some teamd connection in NM?
What "nmcli con" command says?
Weird indeed. Feel free to cc NM people/maling-list.
I have found a workaround to this NetworkManager problem:
which is to use NetworkManager to create a team device.
[root@localhost scripts]# nmcli connection edit type team
This hardly seems like a great solution when the whole point of
hosting lxc containers is to have them be self contained and not rely
on NetworkManager.
The problem that I have been experiencing has gone from
explicitly creating a team device does not work and disables all
networking on a VM :-( :-( :-(
to
explicitly creating a team device does not work with NM :-( :-(
to
I will have to rewrite teamd scripts that use docker to use 'nmcli'
instead because of NM :-(
Perhaps I have failed to appreciate some aspect of the implementation.
Should I infer from you answer that it is mandatory to completely
disable NM on F21 if you wish to explicitly create team devices?
[root@localhost ~]# systemctl stop NetworkManager.service
<futzes about fixing networking for a bit>
[root@localhost ~]# teamd -d
[root@localhost ~]# ip link show team0
16: team0: <BROADCAST,MULTICAST> mtu 1500 qdisc noop state DOWN mode
DEFAULT group default
link/ether 76:91:c0:24:ff:f2 brd ff:ff:ff:ff:ff:ff
[root@localhost ~]# teamd --team-dev teamRR2 -d
Ah I see that worked.
I used to have to avoid NM because it didn't play well with OVS, but
that is fixed now, and NM is what you get in a default install.
Nowhere have I seen any mention that there is a known problem between
NM and teamd.
Perhaps this information could be added to the developer
guide/docs/webite etc. so that others can avoid spending their time
troubleshooting it?
Thanks
Colin