Re: Latest kvm autotest patch queue

[jason wang <jasowang@redhat.com>]

sudhir kumar 写道:
> On Mon, Jun 22, 2009 at 6:54 PM, Lucas Meneghel Rodrigues<lmr@redhat.com> wrote:
>   
>> KVM Autotest patch queue
>>
>> --------------------------------------------------------------------------------
>> Michael Goldish (mgoldish@redhat.com)
>>
>> 1) KVM test: optionally convert PPM files to PNG format after test - Reviewed
>>
>> 2) Step file tests: introducing a new feature and some small changes - Reviewed
>>
>> 3) Introducing kvm_subprocess - Reviewed
>>
>> --------------------------------------------------------------------------------
>> Jason Wang (jasowang@redhat.com)
>>
>> 1) TAP network support in kvm-autotest
>>
>> Michael Goldish has a similar patch to this one, so we need to decide which one
>>     
> I remember Michael's patch using nmap to get the ip of the guest. I
> tested the patch and found nmap taking too long to do a lookup and get
> the guest ip. I will suggets two things here.
> 1. Modify nmap options to make nmap lookup really faster.
> 2. Provide an option to use a dictionary based(mac address to ip
> mapping) lookup in config file and return mapped ip for a mac address.
>
> I can send a patch for option 2 on top of Michael's patch. In the
> simplest form it will generate a dictionary from a macaddress vs IP
> address list(provided by the dhcp server in your lab/environment. you
> can put this list in a file and use it). You need to specify macaddr
> parameter in config file from this list. Looking the mac address the
> parser will return the corresponding ip address from the dictionary
> generated.
>
> Please let me know if 2 sounds good. Also I think I missed to look at
> Jason's patch?
>   
My patch could be seen at http://patchwork.kernel.org/patch/23524/. I 
think the mac:ip dictionary is not flexible enough especially in a 
network with public dhcp server. The scanning method could be used in 
either public or private bridge. But my patch still have some problems 
to solve and need advices:
1 Mac address generation: I could rework the patch and make it could 
accept user specified address range, but the user should guarantee its 
validity.
2 IP scanner: Is there any option of nmap could make the scanning faster 
than -sP? I also agree that nmap is not a elegant method, but it seems 
the sole choice ( or write a simple scanner through python?) . Another 
method is to snoop the dhcp session through tcpdump during the 
vm.create() but this method does not feed for all dhcp server.
3 I want the make network method to be an exclusive options and want to 
know if it is useful to have both user mode nics and bridge nics 
simultaneously.
4 Based on advices of Michael Goldish I would add an index to 
get_address() and get_port() to adapt the environments with multiple nics.

I would rework the patches based on the comments, and like to see the 
patch of michael.
Thanks.
**