why the pid namespace is not compiled in the kernel ?

Daniel Lezcano daniel.lezcano at free.fr
Tue Jan 13 21:52:17 GMT 2009


Tim Gardner wrote:
> Daniel Lezcano wrote:
>   
>> Tim Gardner wrote:
>>     
>>> Daniel Lezcano wrote:
>>>  
>>>       
>>>> Daniel Lezcano wrote:
>>>>    
>>>>         
>>>>> Hi,
>>>>>
>>>>> I hope it is the right mailing list to ask  :)
>>>>>
>>>>> I tried the latest kernel version from "intrepid" and it looks like
>>>>> the namespaces are compiled in except the pid namespace (according
>>>>> the config file stored in /boot).
>>>>> Is there any particular reason ?
>>>>>
>>>>> Thanks.
>>>>>  -- Daniel
>>>>>
>>>>> ps: I recently subscribed to this mailing list, sorry if this
>>>>> question was already asked ...
>>>>>         
>>>>>           
>>>> did I ask to the right mailing list ?
>>>>
>>>>     
>>>>         
>>> Though there are a few features included in the config that depend on
>>> EXPERIMENTAL, CONFIG_PID_NS is not deemed sufficiently interesting to
>>> mess with.
>>>   
>>>       
>> Ah, I see, like the network namespace, it is experimental, that makes
>> sense.
>> We will have to wait a litlle before having a full featured container in
>> Ubuntu :)
>>
>> Thanks.
>>  -- Daniel
>>
>>     
>
> I'm not totally opposed, but you'll need to convince me with use cases
> and some stability analysis.
>   

The namespaces with the control group provides the ability to create a 
virtual private server.
You can launch an application like sshd or apache with its own private 
resources, that allows to make several instances of the same server on 
the same host without conflicts. You can launch several operating 
systems (eg. a debian) on the same host.
This is different from the virtual machine because the kernel is shared 
and it is up to it to handle the system resources per group of processes.
The advantage of this approach is the scalability and the very low 
overhead of the virtualization.

There are two projects implementing the container feature, the libvirt 
and the liblxc.

The pid namespace is enabled since fedora 9 and opensuse 11, and I 
didn't fall into any problem while using the liblxc, I guess we can 
consider it stable.
The network namespace is mutually exclusive with sysfs until 2.6.29,  I 
spotted 2 bugs in the netwok namespace and I am fixing them right now, 
one is leading to a kernel panic (fixed) and the last one just fails 
gracefully, sometimes, to create a network namespace when trying to 
instantiate a new network namespace in a infinite loop.

AFAICS, nobody complained about the namespaces being enabled in these 
different distros.

The namespaces tests are included in the ltp test suite, so IMHO, it is 
reasonable to say they are stable.
In any case, "experimental" is a scary word and I understand why the 
feature would not be enabled for a stable kernel version :)
If the features are missing I can live with a custom kernel until 
everything is enabled.

FYI, I added the lxc.7 man page to this email, I hope that can give some 
clues of what we can do with the namespaces and the cgroup :)

Thanks.
  -- Daniel






-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: lxc.txt
Url: https://lists.ubuntu.com/archives/kernel-team/attachments/20090113/1a13dfcc/attachment.txt 


More information about the kernel-team mailing list