Paul,

Thanks for the question, I'm sorry for the delayed response, I was
traveling home across the country yesterday.  It really depends on the
application itself, I've worked with customers that had issues with smart
caching implementations that pre-date serve stale where the stale data was
returned before performing the fetch and they had to disable the feature
because their applications couldn't tolerate it.  In that case it was in
the resolver versus the client, but the behavior was similar, using stale
data before fetching.  I don't know their underlying applications enough to
tell you why, but I've seen it more than once.

>From a normal GSLB perspective, in some cases it could run in to issues if
persistence was tracked very closely to the same timeframe as the TTL,
assuming that persistence is critically important, which it is in some
cases.  I've also come across issues in the past where clients not behaving
as normal stub resolvers (usually because they were running dnsmasq or a
local BIND instance) experienced unique application issues.  So making
clients behave in unexpected ways can certainly lead to unintended
consequences.  That is why I called it out as a potential issue, because
that is definitely not expected client behavior today, and thus, it could
lead to some negative unintended consequences.

Thanks,
Ross


On Fri, Jun 5, 2026 at 12:04 PM Paul Hoffman <[email protected]> wrote:

> On Jun 5, 2026, at 08:19, Ross Gibson <[email protected]> wrote:
> >
> > The concept of shifting that behavior to the client versus having the
> DNS infrastructure handle it could have a wide array of undesirable
> consequences.  Given the prevalence of the use of global server load
> balancing (GSLB) technology, which often relies on low TTLs to be able to
> control traffic flow, this kind of behavior could possibly break all kinds
> of application infrastructures both internally and externally.
>
> Please say more about what kind of breakage you are imagining. Maybe I'm
> being unimaginative, but I can't see why a record with a 30 second TTL that
> was last fetched in the last 60 seconds should have a negative effect if it
> is prefetched.
>
> --Paul Hoffman
>
> _______________________________________________
> DNSOP mailing list -- [email protected]
> To unsubscribe send an email to [email protected]
>
_______________________________________________
DNSOP mailing list -- [email protected]
To unsubscribe send an email to [email protected]

Reply via email to