[lxc-devel] [Lxc-users] Request for inclusion into mainline LXC utils

[lxc at zitta.fr]

Le 01/02/2010 00:15, Daniel Lezcano a écrit :
> Dominik Schulz wrote:
>   
>> Am Samstag 30 Januar 2010 21:54:29 schrieb Guillaume ZITTA:
>>     
>>> Sorry for the late response, I was on holidays.
>>> I do think joining efforts is always a good thing.
>>> I think some things needs to be defined :
>>> - best practices for a good container is (no udev, syslog conf...)
>>> - what minimal features we expect from container creation scripts.
>>> - who works on it.
>>>       
>> Hi,
>> I'm rather new to LXC but I'm already working on improving the existing tools.
>>
>> My work is based on that of Nigel Mcnie [1]. Since he doesn't seem to  be 
>> fully involved into LXC I'm looking for a place to contribute my patches to.
>>
>> I propose a clear separation of concerns. The core package "lxc" should only 
>> include the essential userland tools, mostly those written in C.
>>     
> I agree.
>
me too.
>> The fancy ones should go into a package of their own. Either separated by distribution 
>> (lxc-debian, lxc-redhat, ...) or all in one (lxc-utils).
>>     
>
> There is too much combination of containers configuration, IMO it should 
> be preferable to keep them separated:
>   lxc-debian (lenny, sid, ...)
>   lxc-fedora (f10, f11, ...)
>   lxc-opensuse (10.1, 11.0, 11.1, ...)
>   lxc-busybox (statically linked or not)
>
> That would be nice to identify clearly who handle a script(s).
>
> That do not prevent to build on top of these scripts a single one.
>
> There is also the sysvrc vs upstart configuration.
>
> We have to deal with the host vs container distro too.
>
> There is the container configuration itself (eg. macvlan, vlan, veth, 
> etc ... ) to be interactive or not, and the distros configuration (eg. 
> static ip or dhcp).
>
> Note people would be interested by templates which are not only distros 
> but also simple applications like sshd or apache+mysql. Why running a 
> full container to host a web browser ?
>
>   
I think we have 4 or 5 levels of configuration :
- Common to all Linux
    example : /etc/resolv.conf
- Distro family
    example for debian-like : /etc/network/interfaces
- Distro ( useful? )
- Distro version
    example for Ubuntu karmic : upstart, mountall...
- Application or user specific (a gentoo webserver, a debian mailserver,
...)

We should make a modular program so that everybody can simply add a new
distro or appliance.

>> Further I propose not to separate tools which should be united in one. I'd 
>> like to see the a separation of the container-creation tools based on the 
>> lower level programs they use. Something like lxc-debootstrap for Debian-based 
>> distributions and something alike for the ones based on RPM. Because 
>> separating Debian and Ubuntu doesn't seem to support achieving our objectives. 
>> They are just to similar in terms of creating containers.
>>     
> There is the febootstrap command.
>
>   
>> (Partly) in contrast to the proposal of Daniel Lezcano [2] I'd propose to keep 
>> the core utils small and simple (following the well known KISS principle) and 
>> don't go for templates which are called by lxc-create. Instead I'd keep lxc-
>> create as small as possible and incorporate it into other tools, which I've 
>> mentioned above.
>>     
> That makes senss.
>
> Should we have a separate project ? or shall we keep these scripts in 
> the lxc source tree in a different location in order to have the core 
> and the templates synced ? For example, Michael H. Warfield and Tony 
> Risinger are writing some useful scripts to shutdown / reboot the 
> containers, I hope that won't be a third package, so the user will be 
> totally lost.
>
>   

I agree.

Dominik, you said that you started some work. anything visible?

Regards,
Guillaume ZITTA