NetworkManager hijacks LNST controlled interfaces

[Radek Pazdera]

Hi,

there's a problem I've been dealing with network manager running on the 
slaves.

Each time a new interface is hot-plugged to the machine, NM creates a 
default
connection profile for it and executes it. It means, that it tries to 
run dhclient
on that interface once in a while.

That's a problem, because we configure interfaces on slaves "manually" 
with ip
command. This configuration is flushed when by NM when it tries to 
enforce the
"Auto eth5" default profile on the interface. There's not even a DHCP
server
running in the environment I tested this in.

The thing is, that we need to have the network-manager running, because 
it starts
dhclient when the controller interface is plugged in. I don't know 
whether the
standard networking scripts can do this too. And if they can, we need a 
way to
turn NM off somehow anyway (which is hard, because we don't have shell 
at that
time yet).

NM has a D-Bus interface and a cli-wrapper around it called 'nmcli'. 
This utility
is shipped along with the NM package. It can be used to exclude some 
interfaces
from NM control by issuing

     nmcli dev disconnect iface eth2

command without necessity to restart the whole service (which is 
something we cannot
do, since we're connected through ssh).

Has this ever happened to you? It usually results a failure in the 
cleanup phase on

     ip addr del ...

command saying that it cannot clear the address (because NM already 
flushed it). Or
do you have NM off on slaves?

Radek :)

[Jan Tluka]

See some notes below ...

Mon, Aug 06, 2012 at 12:23:50PM CEST, rpazdera(a)redhat.com wrote:
> Hi,
>
> there's a problem I've been dealing with network manager running on
> the slaves.
>
> Each time a new interface is hot-plugged to the machine, NM creates a
> default
> connection profile for it and executes it. It means, that it tries to
> run dhclient
> on that interface once in a while.
>
> That's a problem, because we configure interfaces on slaves
> "manually" with ip
> command. This configuration is flushed when by NM when it tries to
> enforce the
> "Auto eth5" default profile on the interface. There's not even a DHCP
> server
> running in the environment I tested this in.
>  
I'd get rid off NM completely. I think that the safest method is to
disallow installation of NetworkManager package.

> The thing is, that we need to have the network-manager running,
> because it starts
> dhclient when the controller interface is plugged in. I don't know
> whether the
> standard networking scripts can do this too. And if they can, we need
> a way to
> turn NM off somehow anyway (which is hard, because we don't have
> shell at that
> time yet).
>  
Could we utilize the post-installation scripts? See anaconda
documentation [1] for details.

[1] http://fedoraproject.org/wiki/Anaconda/Kickstart#Chapter_5._Post-installa...

> NM has a D-Bus interface and a cli-wrapper around it called
> 'nmcli'.
> This utility
> is shipped along with the NM package. It can be used to exclude some
> interfaces
> from NM control by issuing
>
>     nmcli dev disconnect iface eth2
>
> command without necessity to restart the whole service (which is
> something we cannot
> do, since we're connected through ssh).
>
> Has this ever happened to you? It usually results a failure in the
> cleanup phase on
>
>     ip addr del ...
>
> command saying that it cannot clear the address (because NM already
> flushed it). Or
> do you have NM off on slaves?
>  
Yes, completely off :-)

> Radek :)
> _______________________________________________
> LNST-developers mailing list
> LNST-developers(a)lists.fedorahosted.org
> https://lists.fedorahosted.org/mailman/listinfo/lnst-developers

[Radek Pazdera]

On 08/06/2012 01:24 PM, Jan Tluka wrote:
>  See some notes below ...
>
>  Mon, Aug 06, 2012 at 12:23:50PM CEST, rpazdera(a)redhat.com wrote:
> > Hi,
> >
> > there's a problem I've been dealing with network manager running on
> > the slaves.
> >
> > Each time a new interface is hot-plugged to the machine, NM creates a
> > default
> > connection profile for it and executes it. It means, that it tries to
> > run dhclient
> > on that interface once in a while.
> >
> > That's a problem, because we configure interfaces on slaves
> > "manually" with ip
> > command. This configuration is flushed when by NM when it tries to
> > enforce the
> > "Auto eth5" default profile on the interface. There's not even a DHCP
> server
> > running in the environment I tested this in.
> >
>  I'd get rid off NM completely. I think that the safest method is to
>  disallow installation of NetworkManager package. 
Disabling NM would probably save us a lot of trouble, that's for sure ...

In that case we could think about dropping the dynamic controller
interface completely. As far as I know beaker set's up a working
interface so it shouldn't be a problem. Even when machines are
installed manually, there's always one NAT'ed interface present
by default.

I'm going to do the virtual netdevices first. And then we'll see if
the virtualization of controller interface is really necessary or
useful at all.

> > The thing is, that we need to have the network-manager running,
> > because it starts
> > dhclient when the controller interface is plugged in. I don't know
> > whether the
> > standard networking scripts can do this too. And if they can, we need
> > a way to
> > turn NM off somehow anyway (which is hard, because we don't have
> > shell at that
> > time yet).
> >
>  Could we utilize the post-installation scripts? See anaconda
>  documentation [1] for details.
>
>  [1] http://fedoraproject.org/wiki/Anaconda/Kickstart#Chapter_5._Post-installa...
>
> > NM has a D-Bus interface and a cli-wrapper around it called 'nmcli'.
> > This utility
> > is shipped along with the NM package. It can be used to exclude some
> > interfaces
>  >from NM control by issuing
> >     nmcli dev disconnect iface eth2
> >
> > command without necessity to restart the whole service (which is
> > something we cannot
> > do, since we're connected through ssh).
> >
> > Has this ever happened to you? It usually results a failure in the
> > cleanup phase on
> >
> >     ip addr del ...
> >
> > command saying that it cannot clear the address (because NM already
> > flushed it). Or
> > do you have NM off on slaves?
> >
>  Yes, completely off :-)
>
> > Radek :)
> > _______________________________________________
> > LNST-developers mailing list
> > LNST-developers(a)lists.fedorahosted.org
> > https://lists.fedorahosted.org/mailman/listinfo/lnst-developers