I noticed that a number of messages to the users list have been marked with [Spam] in the subjects since the list migration. This appears to be due to the Spamassassin settings on bastion. I think this subject munging is undesirable and would like to see it stopped. Having legitimate messages sent to list members tagged as [SPAM] serves no good purpose. If I were a list member I might wonder why my message was marked or why Fedora is forwarding on mail it thinks is spam.
At the same time, it's worth noting that if the SpamAssassin settings are marking legitimate list mail as spam, they probably ought to be tweaked a bit. If these messages had gone through bastion before hitting Mailman and were to a list with a rule to reject or discard on X-Spam-Flag (as I know several lists do, websites being one example), the messages would have been improperly discarded.
This isn't to say I don't appreciate the job that SpamAssassin does nor the hard work put in by Warren and others upstream. I just want to ensure that SpamAssassin does not act too aggressively at marking up the mail that comes through fedoraproject.org.
I've attached an example of a recent users list message which was marked as spam improperly. The relevant SpamAssassin headers:
X-Spam-Checker-Version: SpamAssassin 3.3.0 (2010-01-18) on bastion2.fedora.phx.redhat.com X-Spam-Flag: YES X-Spam-Level: ***** X-Spam-Status: Yes, score=5.6 required=5.0 tests=BAYES_50,SPOOF_COM2COM, SPOOF_COM2OTH autolearn=no version=3.3.0 X-Spam-Report: * 2.7 SPOOF_COM2OTH URI: URI contains ".com" in middle * 2.0 SPOOF_COM2COM URI: URI contains ".com" in middle and end * 0.8 BAYES_50 BODY: Bayes spam probability is 40 to 60% * [score: 0.5000]
On 02/02/10 14:32, Todd Zullinger wrote:
I noticed that a number of messages to the users list have been marked with [Spam] in the subjects since the list migration. This appears to be due to the Spamassassin settings on bastion. I think this subject munging is undesirable and would like to see it stopped. Having legitimate messages sent to list members tagged as [SPAM] serves no good purpose. If I were a list member I might wonder why my message was marked or why Fedora is forwarding on mail it thinks is spam.
At the same time, it's worth noting that if the SpamAssassin settings are marking legitimate list mail as spam, they probably ought to be tweaked a bit. If these messages had gone through bastion before hitting Mailman and were to a list with a rule to reject or discard on X-Spam-Flag (as I know several lists do, websites being one example), the messages would have been improperly discarded.
This isn't to say I don't appreciate the job that SpamAssassin does nor the hard work put in by Warren and others upstream. I just want to ensure that SpamAssassin does not act too aggressively at marking up the mail that comes through fedoraproject.org.
I've attached an example of a recent users list message which was marked as spam improperly. The relevant SpamAssassin headers:
X-Spam-Checker-Version: SpamAssassin 3.3.0 (2010-01-18) on bastion2.fedora.phx.redhat.com X-Spam-Flag: YES X-Spam-Level: ***** X-Spam-Status: Yes, score=5.6 required=5.0 tests=BAYES_50,SPOOF_COM2COM, SPOOF_COM2OTH autolearn=no version=3.3.0 X-Spam-Report: * 2.7 SPOOF_COM2OTH URI: URI contains ".com" in middle * 2.0 SPOOF_COM2COM URI: URI contains ".com" in middle and end * 0.8 BAYES_50 BODY: Bayes spam probability is 40 to 60% * [score: 0.5000]
infrastructure mailing list infrastructure@lists.fedoraproject.org https://admin.fedoraproject.org/mailman/listinfo/infrastructure
I like the spam headers, and have yet to see a single message marked as spam, which originated from a user. And I am on multiple lists. Also, without the headers of the message, it is hard to say what is causing said scoring. It could be the users end or maybe even their fault, i.e. using a SMTP they should not. As it happened to one of my clients. Also, just because it is marked [spam], does not mean it gets removed, and I believe most people can use common sense in that regard.
Regards, Tristan
On 2010-02-02 03:02:18 PM, Tristan Santore wrote:
I like the spam headers, and have yet to see a single message marked as spam, which originated from a user. And I am on multiple lists. Also, without the headers of the message, it is hard to say what is causing said scoring. It could be the users end or maybe even their fault, i.e. using a SMTP they should not. As it happened to one of my clients. Also, just because it is marked [spam], does not mean it gets removed, and I believe most people can use common sense in that regard.
Todd attached the message and full headers, and in this case, it seems that the presence of a domain name with ".com" in the middle caused a huge spam score, which definitely seems unreasonable.
I agree with Todd that the spam headers suffice without having to munge the subject.
Thanks, Ricky
On 02/02/10 16:22, Ricky Zhou wrote:
On 2010-02-02 03:02:18 PM, Tristan Santore wrote:
I like the spam headers, and have yet to see a single message marked as spam, which originated from a user. And I am on multiple lists. Also, without the headers of the message, it is hard to say what is causing said scoring. It could be the users end or maybe even their fault, i.e. using a SMTP they should not. As it happened to one of my clients. Also, just because it is marked [spam], does not mean it gets removed, and I believe most people can use common sense in that regard.
Todd attached the message and full headers, and in this case, it seems that the presence of a domain name with ".com" in the middle caused a huge spam score, which definitely seems unreasonable.
I agree with Todd that the spam headers suffice without having to munge the subject.
Thanks, Ricky
infrastructure mailing list infrastructure@lists.fedoraproject.org https://admin.fedoraproject.org/mailman/listinfo/infrastructure
If the score originates from com2com scoring, then the filter needs fixing, as it did not look for .com but anything com. This looks more like a bug that needs to be submitted to the relevant developers.
Regards,
Tristan
On 2010-02-02 05:27:22 PM, Tristan Santore wrote:
If the score originates from com2com scoring, then the filter needs fixing, as it did not look for .com but anything com. This looks more like a bug that needs to be submitted to the relevant developers.
I don't think it was necessarily a bug in the rule. The original email that Todd forwarded contained "google.com.s9b1.psmtp.com." which looks like it could have set it off.
Also, an update to this - we've disabled the [SPAM] subject in our spamassassin config now. The normal spamassasin headers will still be included in emails, but the subject just won't be modified.
Thanks, Ricky
On 2010-02-02 12:49:39 PM, Ricky Zhou wrote:
Also, an update to this - we've disabled the [SPAM] subject in our spamassassin config now. The normal spamassasin headers will still be included in emails, but the subject just won't be modified.
Another update to this, spamassassin is now disabled on bastion. Spam filtering for @fp.o email will take place on Red Hat MXs. However, this means that lists.fp.o email is no longer spam filtered at any level.
Test: a.b.com.c.d.net.e
Thanks, Ricky
On Tuesday 02 February 2010 08:32:04 am Todd Zullinger wrote:
I noticed that a number of messages to the users list have been marked with [Spam] in the subjects since the list migration. This appears to be due to the Spamassassin settings on bastion. I think this subject munging is undesirable and would like to see it stopped. Having legitimate messages sent to list members tagged as [SPAM] serves no good purpose. If I were a list member I might wonder why my message was marked or why Fedora is forwarding on mail it thinks is spam.
At the same time, it's worth noting that if the SpamAssassin settings are marking legitimate list mail as spam, they probably ought to be tweaked a bit. If these messages had gone through bastion before hitting Mailman and were to a list with a rule to reject or discard on X-Spam-Flag (as I know several lists do, websites being one example), the messages would have been improperly discarded.
This isn't to say I don't appreciate the job that SpamAssassin does nor the hard work put in by Warren and others upstream. I just want to ensure that SpamAssassin does not act too aggressively at marking up the mail that comes through fedoraproject.org.
I've attached an example of a recent users list message which was marked as spam improperly. The relevant SpamAssassin headers:
X-Spam-Checker-Version: SpamAssassin 3.3.0 (2010-01-18) on bastion2.fedora.phx.redhat.com X-Spam-Flag: YES X-Spam-Level: ***** X-Spam-Status: Yes, score=5.6 required=5.0 tests=BAYES_50,SPOOF_COM2COM, SPOOF_COM2OTH autolearn=no version=3.3.0 X-Spam-Report: * 2.7 SPOOF_COM2OTH URI: URI contains ".com" in middle * 2.0 SPOOF_COM2COM URI: URI contains ".com" in middle and end * 0.8 BAYES_50 BODY: Bayes spam probability is 40 to 60% * [score: 0.5000]
Mail only hits bastion on the way out. all mail comes in via one of the 3 smtp relays and gets sent to the mailman server. The idea of the relays is make sure we accept all mail coming in. once processed by mailman all mail hits bastion on its way out. so the lists should not reject mail improperly.
Dennis
On Tue, 2 Feb 2010, Todd Zullinger wrote:
I noticed that a number of messages to the users list have been marked with [Spam] in the subjects since the list migration. This appears to be due to the Spamassassin settings on bastion. I think this subject munging is undesirable and would like to see it stopped. Having legitimate messages sent to list members tagged as [SPAM] serves no good purpose. If I were a list member I might wonder why my message was marked or why Fedora is forwarding on mail it thinks is spam.
At the same time, it's worth noting that if the SpamAssassin settings are marking legitimate list mail as spam, they probably ought to be tweaked a bit. If these messages had gone through bastion before hitting Mailman and were to a list with a rule to reject or discard on X-Spam-Flag (as I know several lists do, websites being one example), the messages would have been improperly discarded.
Anyone know exactly when this started?
-Mike
On 02/02/2010 02:15 PM, Mike McGrath wrote:
Anyone know exactly when this started?
-Mike
My understanding is it was configured this way for a long time, only we didn't notice any problem because list mail wasn't going through the bastion filter before.
Warren
infrastructure@lists.fedoraproject.org