rrset-roundrobin

[Gavin McCullagh]

Hi,

I think this is a move toward a better default behavior.  These are "sets"
(unordered) after all and most people in my experience prefer load to be
spread across them as evenly as possible.  So having the major resolvers do
this out of the box sounds like a good thing.

That said, I would caution that this will likely will be a breaking change
for some of your customers.  Whatever behavior you have today, some people
will be reliant on it and surprised when it changes.  At my place of work,
I have spoken to at least one customer who had an off-the-shelf GSLB setup
which returned multi-record answers sorted based on "nearest location
first".   In other words, these GSLBs seem to be set up on the assumption
the resolver doesn't alter the order of the rrsets. So, you may surprise
some people when the Unbound behavior changes.

I don't know NLNet's philosophy on breaking changes, but if you go ahead
I'd suggest you publicize this behavior change as visibly as you can.
Those who package Unbound for Linux distros etc. might want to think about
whether they would prompt upgrading users with a question (like "do you
want me to a. alter your config to preserve previous behavior or b. accept
the changing default?").

Gavin


On Thu, Apr 9, 2020 at 4:28 AM George Thessalonikefs via Unbound-users <
unbound-users at lists.nlnetlabs.nl> wrote:

> Hi all,
>
> We recently got a feature request
> (https://github.com/NLnetLabs/unbound/issues/215) to change the default
> value of 'rrset-roundrobin' from no to yes.
> Given that we don't have strong feelings either way I am reaching out to
> you in case you feel opposed to the change.
>
> Please let this not be derailed (too much) to a discussion about the
> effectiveness of the feature in different scenarios.
>
> Best regards,
> -- George
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.nlnetlabs.nl/pipermail/unbound-users/attachments/20200412/60f678c6/attachment.htm>