If I understand Mark correctly, his data indicates that DNS works great
unless you are:

   - dependent on Microsoft as a DNS client to submit the query or
   - dependent on Microsoft as a DNS server to respond to the query.


Sadly, that is a lot of dependent traffic.


On Mon, Nov 4, 2024 at 7:18 PM Mark Alley <mark.alley=
[email protected]> wrote:

> Just sharing on the topic of authentication, DNS, and temperrors - I have
> recent statistics timeline (attached) involving DKIM failure rates for a
> high volume of mail for $dayjob that directly correlates the effect of TTLs
> and caching on temperrors and authentication (with Microsoft being an
> anomaly in this dataset). This volume in total roughly accounts for 738k
> DKIM failures out of 695.5m emails over 30 days, which is about a 0.1% DKIM
> failure rate overall.
>
> Splitting the failures up by reporter, Microsoft currently is the
> overwhelming offender when it comes to temperrors, by magnitudes of several
> thousands of percentage points, compared to all other receivers combined.
> Even increasing TTLs inordinately high had barely any effect on Microsoft's
> DNS errors. All other reporters fell off a cliff once the DKIM TTLs were at
> 24 hours.
>
> This also seems to apply to SPF as well.
>
> - Mark Alley
> On 11/4/2024 5:31 PM, Douglas Foster wrote:
>
> Surprisingly, for email, DNS errors are not rare.   I have been collecting
> data from three sources:
>
> - The incoming messages to my server
> - The Aggregate Report data about messages that came from me or appear to
> come from me, and
> - The Authentication Results header fields on incoming messages, which is
> evaluated by unrelated servers about other unrelated domains.
>
> In each case, my TempError rate is 1 in 350 messages or higher.   I
> expected something on the order of 1 in a million or 1 in a billion.
>
> I don't understand why the error rates are so high.   I may undertake a
> study to determine what percentage of TempErrors are transient events and
> what percentage are highly repeatable events, but I have not done so yet..
>  Perhaps Google or someone else can contribute some insight.
>
> Doug Foster
>
>
>
>
> On Sun, Nov 3, 2024 at 1:29 PM Alessandro Vesely <[email protected]> wrote:
>
>> On Sun 03/Nov/2024 13:56:42 +0100 Douglas Foster wrote:
>> > The other problem is that the Tree Walk makes the entire document
>> > experimental.      Tree Walk is a great tool for batch-mode detection
>> of
>> > PSL errors, but it has multiple problems as a real-time tool.
>> >
>> >     - It disables best-guess DMARC based on default policy.   Instead of
>> >     disabling best-guess DMARC, this document should be embracing it.
>>
>>
>> Best-guess DMARC can be a local filtering method, but it's not being
>> standardized.  Using the Tree Walk doesn't preclude to use the PSL for
>> any
>> local intent.
>>
>>
>> >     - It is prone to inconsistent results and indeterminate results
>> caused
>> >     by DNS errors.
>>
>>
>> On a well supplied server, DNS errors are quite rare.  An error
>> retrieving
>> DMARC records, which can happen also when the org domain is discovered by
>> PSL,
>> deserves a 4xx temporary error.
>>
>>
>> >     - Multiple DNS lookups are reasonably expected to have lower
>> performance
>> >     than one database lookup, especially since high-throughput
>> environments can
>> >     use memory-resident databases for the PSL lookup.
>>
>>
>> All the involved domains can be queried in one shot.  That way the time
>> required is about the same as querying a single record.
>>
>>
>> >     -  Inconsistent results can be mitigated by using long-term caching
>> >     outside of DNS, but that adds complexity and overhead.
>>
>>
>> Agreed.  It is much better to deploy a dedicated DNS server.
>>
>>
>> > [...]
>> >
>> > In aggregate, I see no justification for DMARCbis to be given Standards
>> > Track status, unless the Tree Walk is separated out as an experimental
>> > option, either in an appendix or in a separate document.
>>
>>
>> Using the PSL was the experiment.  Standardization provides a
>> better-defined
>> method by which a domain owner can decide which domains are
>> "organizational"
>> without having to seek help from a vaguely defined group that maintains a
>> list
>> designed for a different purpose.
>>
>> I don't think experimental status is required for that.  Anyway, that is
>> to be
>> decided by the IESG or the RFC-Editor, methinks.  Our charter says:
>>
>>      The task of defining a standard mechanism for identifying
>> organizational
>>      domain is out of scope for this working group. However the working
>> group
>>      can consider extending the base DMARC specification to accommodate
>> such a
>>      standard, should it be developed during the life of this working
>> group.
>>
>>
>> Best
>> Ale
>> --
>>
>>
>>
>> _______________________________________________
>> dmarc mailing list -- [email protected]
>> To unsubscribe send an email to [email protected]
>>
>
> _______________________________________________
> dmarc mailing list -- [email protected]
> To unsubscribe send an email to [email protected]
>
> _______________________________________________
> dmarc mailing list -- [email protected]
> To unsubscribe send an email to [email protected]
>
_______________________________________________
dmarc mailing list -- [email protected]
To unsubscribe send an email to [email protected]

Reply via email to