Trashing sort and uniq attribute on filter-spec

Francis J. Lacoste francis.lacoste at Contre.COM
Wed Jan 16 15:42:32 CET 2002


On Wed, Jan 16, 2002 at 02:48:06PM +0100, Joost van Baal wrote:
> Before picking the delivery attempts with highest delay, the duplicate
> entries about the same message are filtered out, by doing
> 
>  <lire:filter-spec>
>   <lire:uniq fields="msgid"/>
>  </lire:filter-spec>
> 
> One message can be listed more than once in the email dlf file: if it's
> got more than one recipient, it's listed for each recipient.  We don't
> want to "punish" relays which receive much messages for multiple
> recipients.

Hmm. But this isn't a "correct" solution. 

1- If a message is delivered to different recipients through
different relays, we lose some relays. This could be solved
through using <lire:uniq fields="msgid from_relay_host to_relay_host "/>
With this we would keep one message by ID.

2- Message-ID can be reused by SpamBot and Mailing-List. You can see
mailing list messages through several connections (the envelope
from_address will change to handle bounce according to the recipient). 

Finally, since we are averaging, we aren't "punishing" relays which 
receive much messages for multiple recipients. In this case, the relay
value will probably be the same for each recipient (if it was delivered
in the same run):

X+X+X+X/4 = X

> 
> BTW, i think we can get away with uniq and sort: indeed, this trick
> can be achieved by a derivedschema.
> 

OK. I'm removing them now.

-- 
Francis J. Lacoste
francis at Contre.COM
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 232 bytes
Desc: not available
Url : http://lists.logreport.org/pipermail/development/attachments/20020116/fad55c39/attachment.bin 


More information about the Development mailing list