Open Stack

Here's a final (hopefully) review of uptake for the
openstack-discuss mailing list, now that the old lists it replaced
have been closed for three months. See my previous reports[0][1]
from the one- and two-month anniversaries.

The volume of posts from non-subscribers landing in the moderation
queue has continued to dwindle to only a few each week now. We've
reached 934 subscribers to openstack-discuss (up from 830 in the
previous report). 76% of the addresses used to send 10 or more
messages to the old lists in 2018 are now subscribed to the new one
(it was 75% a month ago so we're probably in the long tail now). As
anticipated, posting volume rose in February to 1083 messages as
compared to January's 958; for reference that's an 11% drop from the
1196 we had in Februafy 2018 (and again recall that right at 10% of
the messages on the old lists were duplicates from cross-posting).

We're continuing to investigate causes of messages through the list
failing DKIM signature validation:

* The most frequent cause was Mailman's recipient deduplication,
  since it works by filtering list subscribers out of the Cc header
  on receipt prior to processing for redelivery. As an experiment
  we've removed[2] it from Mailman's module pipeline, which has
  significantly reduced the number of invalid DKIM signatures (but
  at the expense of an increase in people receiving duplicate copies
  of list replies on which they were also Cc'd directly).

* For the next most common cause, we've observed some mail servers
  using the much more strict "simple" header canonicalization
  algorithm[3] which makes whitespace in message headers significant
  (as opposed to the more popular "relaxed" canonicalization), and
  Mailman has a penchant for normalizing whitespace in message
  headers prior to forwarding which invalidates these particular
  signatures. We're not sure what the best option for addressing
  this is yet, but are considering reaching out to the affected
  individuals to see if they'll lobby their respective organizations
  to reconsider their choice of canonicalization.

* The third and least common cause we've identified is a participant
  whose References header incorrectly adds a comma separator (RFC
  822 parser spec[4] says space/tab/CRLF are the only allowed
  separators here), so when Mailman reformats the references list
  prior to forwarding it concatenates some elements. We don't have a
  great solution for this one either aside from asking the affected
  individual to work out whether the problem is being introduced by
  their MUA or their organization's MTA.

[0] http://lists.openstack.org/pipermail/openstack-discuss/2019-January/001386.html
[1] http://lists.openstack.org/pipermail/openstack-discuss/2019-February/002335.html
[2] https://review.openstack.org/635078
[3] http://dkim.org/specs/rfc4871-dkimbase.html#canonicalization
[4] https://tools.ietf.org/html/rfc822#section-3.3
-- 
Jeremy Stanley
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 963 bytes
Desc: not available
URL: <http://lists.openstack.org/pipermail/openstack-discuss/attachments/20190304/4f7ff477/attachment.sig>