Hi Björn,

On Fri, Dec 27, 2019 at 09:42:57AM +0100, Björn Jacke wrote:
> > > I'd like to propose to make a change for one of the next major
> > > versions of haproxy that "ip" gets aliases to what "ipv6" currently
> > > is and eventually introduce a new ipv4 type which equals to what
> > > type "ip" is now.
> > > 
> > > What do others here think about this?
> > 
> > Sounds good, would you like to create a patch for that?
> sure, I can start working out a patch if nobody disagrees that introducing
> "ipv4" for v4 only and making "ip" behave equally to "ipv6" is the way to
> go.

I thought I responded but can't find my response so it's possibly one of
those lost before sending. What I wanted to say is that I totally understand
your concern, but am not fond of pure config syntax changes like this which
cause trouble to 100% of the users when upgrading. For example some will
probably upgrade one node at a time in a cluster and will end up with two
nodes out of sync during all the transition period. Others might have
scripts used to automatically generate iptables or ipset rules based on
a "show table" output on the CLI, and they will break when IPv4 gets
reported as "ffff::".

This doesn't mean we shouldn't do anything. Maybe at least creating
"ipv4" and making "ip" remain "ipv4" but emit a warning could be another
option to warn users. Or maybe we should see if it's possible to make IPv4
and IPv6 more interoperable on the peers protocol and make the "show table"
output present v4-in-v6 output as pure IPv4 so that everything remains
exactly the same for all users. That's just an idea to fuel the discussion
of course, as I don't know the feasibility of all this.

Let's wait for some more opinions, especially in this period where everyone
is more or less on vacation :-/


Reply via email to