Re: [Zinq-general] Caching SERVFAIL?

On Sat, May 9, 2009 at 6:47 PM, Jason Haar <[email protected]> wrote:
> On 05/09/2009 06:43 AM, Mark Johnson wrote:
>> I think 5 minutes is waaaay to long.  I'd tend towards less than 30 seconds.
>>
> We used to run it here - but I removed it as it was causing us to
> effectively block mail to a local ISP 3-5 times a day for 5 minutes at a
> time. End result - dozens of bounced emails as Qmail (calling dnscache)
> started receiving "no such host" responses to MX/A lookups due to that
> patch.
>
> As far as I'm aware, it only ever affected our connectivity to that one
> ISP - I'm sure their DNS servers was timing out under load - but the
> reaction of the patch - to effectively say hosts don't exist - was
> definitely not the correct response.

There aren't any existing patches for djbdns that I know of that deal
with this properly.  When and if I get around to dealing with this, it
will probably take the form of some kind of list of query / type / IP
tuples (separate from the cache, which is just query / type ) that
result in SERVFAIL.  I think that's the direction djb was headed in.