Re: [9front] mailing list issues

[Toby <toby@clftn.uk>]

On Tuesday, 08 June 2021 at 16:31:53 -0400, Stanley Lieber wrote:
>On June 8, 2021 3:07:10 PM EDT, Julien Blanchard <julien@typed-hole.org> wrote:
>>> Le 8 juin 2021 à 20:37, Stanley Lieber <sl@stanleylieber.com> a 
>>> écrit :
>>>
>>> messages arrive out of order. messages arrive late. messages don't 
>>> arrive at all.
>>>
>>> I see a lot of stuff like:
>>>
>>> 1ess Jun  8 14:10:41 delivery '/net.alt/dns' does not exist ([known good hostname])
>>>
>>Not sure it can help (I don’t know much about email) but I got DMARC 
>>issues every time I sent an email to the list when a year earlier it 
>>seemed to work fine. From my server that is so maybe I’m at fault 
>>here.
>
>upas can't do any kind of dmarc at all.
>
>trouble amplified when we moved off of ramnode ssd-backed disk to 
>9cloud. we had similar problems years ago when we were on 9cloud before 
>we moved to ramnode.
>
>the way mail is sent, runq hammered the disk, which severely tanked 
>performance on spinning platter disks. we tinkered with it a bit and 
>arrived at the current situation, where sending a message to a list no 
>longer brings the entire system to a crawl, but still exhibits the same 
>other defects.
>
>we ignored this problem for years by throwing an ssd at it.

Do you think that the problem is still related to disk access, or is 
there more going on here, given that some but not all of the trouble has 
gone away with the move to ssd?

Given the messages you see in the log, and that certain aspects of the 
problem (I'm not quite clear which from what you write) seem to exist 
with one cloud provider more than with another, it isn't the case that 
you're hitting some kind of rate limit on the cloud provider's dns 
server, or something like that, is it?

In any case, my guess is that this will have more to do with messages 
arriving late and out of order than not-at-all. I generally haven't 
noticed messages being referred to that never arrive with very few 
exceptions, and when I have noticed it, I think I've always found a 
corresponding entry in my spam filter's log saying that a message from 
1ess has been rejected.

The main cause of spamminess for messages from this list that pass 
through my filter seems to be that the Reply-To header is set equal to 
the To address (9front@9front.org), which seems to be the default 
behaviour for mlmgr.  Other mailing lists I'm on seem to usually set a 
List-Post header to do this job rather than a Reply-To.  This can also 
be useful in the case that one wishes to reply only to the sender of a 
message rather than the whole list.

DMARC policy issues for me give lower contributions to the spam score 
than the above problem, and in any case they are hard to avoid for 
mailing lists. (Something which seems to be reflected in how much notice 
spam filters take of DMARC, especially for messages which look like they 
come from mailing lists.)

A more minor point is that 1ess does not seem to be doing starttls when 
it sends messages on. Whilst my spam filter does not attach very much 
significance to such things, I have heard of people caring rather more, 
sometimes to the point of rejecting mail if it does not reach them in a 
hop that uses starttls. If this can be fixed by adding the -s option in 
whatever call to smtp(8) takes place, perhaps it is worth doing.

Toby