[ovs-dev] [PATCH ovn] expr: Combine multiple ipv4 with wildcard mask.(Internet mail)

[Dumitru Ceara]

On 4/20/21 3:57 AM, gmingchen(陈供明) wrote:
> 
> 
> On 2021/4/12, 4:48 PM, "Dumitru Ceara" <dceara at redhat.com> wrote:
> 
>     On 4/8/21 1:24 PM, gmingchen(陈供明) wrote:
>     > 
>     > 
>     > On 2021/4/7, 12:01 AM, "Dumitru Ceara" <dceara at redhat.com> wrote:
>     > 
>     >     On 4/6/21 12:11 PM, Mark Gray wrote:
>     >     > I had a conversation with Dumitru and this patch came up in the
>     >     > conversation. He made an interesting suggestion (Dumitru, please correct
>     >     > me if I get this wrong) that this could be refactored as an external
>     >     > tool. This cmdline tool could, for example, take a set of IP addresses
>     >     > and return the wildcard representation. For example,
>     >     > 
>     >     > # ./ovs-new-tool <set of addresses>
>     >     > <ip address/prefix> ... <ip address/prefix>
>     >     > 
>     >     > Then this tool could optionally be used by a CMS. The other advantage of
>     >     > this is that it could also be used in other places.
>     > 
>     >     Given that OVN is essentially a stack of compilers (NB -> SB ->
>     >     openflow) I was wondering if it would be reasonable and worth it to go a
>     >     step forward and implement this feature to "combine IP addresses with
>     >     wildcard" as a library.  This would allow using it at different levels
>     >     in the stack, e.g.:
>     >     - at CMS level/as a standalone cmdline tool.
>     > 
>     > My concern is whether the process after cms can correctly handle this
>     > non-standard cidr format (1.1.1.0/255.255.255.253).
> 
>     I would expect this to not be a problem but it depends on CMS indeed.
> 
> Yes.
> 
>     > 
>     >     - in ovn-northd (for example when generating SB address sets from NB
>     >     address sets)
>     > 
>     > If the non-standard cidr format ip address can be handled reasonably
>     > here, I think this is a good idea.
> 
>     These are OVN internals so the approach you were proposing in
>     ovn-controller could be implemented at this level too.
> 
> It is indeed the case.
> 
> I wonder if the scale of the IP address is not so important, so that fewer
> people respond now. So, I wonder if the patch can be put here first, if
> this problem is more common, then further targeted discussion.

Sounds good to me, although one thing we need to take into account in
ovn-northd is that in the C implementation there is no incremental
processing so the translation from address set individual IPs to
combined with wildcard will be executed at every run.

Regards,
Dumitru