On Thu, 24 Nov 2016 11:33:19 +0100 Axb wrote: > On 11/24/2016 11:23 AM, Geoff Soper wrote: > > For a few weeks I've been suffering spam messages with attachments > > getting through with a suspicious score of 0.0. Upon inspection, > > they all had the following lines in the header: > > > > ... > > X-Spam-Status: No, score=0.0 required=3.0 tests=UNPARSEABLE_RELAY > > autolearn=unavailable version=3.3.2
Do you normally have a BAYES_* result in X-Spam-Status? I think that autolearn=unavailable implies that Bayes is configured to be on. Try running one of these through spamassassin -D bayes If you haven't already done it, set "bayes_auto_expire 0" and instead run "sa-learn --force-expire" from cron (as the correct user). > Try this in your local.cf > > clear_headers > add_header all Relays-External _RELAYSEXTERNAL_ > add_header all Relays-Untrusted _RELAYSUNTRUSTED_ UNPARSEABLE_RELAY is only a significant problem if it's on the edge of your internal or trusted network. These relays should be the first--force-expire entries in the above two headers - check that they are sensible.