[Mailman-Users] Mailman + Sendmail

[Rich West]

  Mark Sapiro wrote:

>Rich West wrote:
>
>  
>
>>I did some further analysis and discovered that the messages (with the 
>>same message ID) were showing up in the "ps" output as processing "from 
>>queue".  Digging a little further, I tried killing off the sendmail 
>>processes that were processing those requests, and I saw the following 
>>appear in the mailman log:
>>Sep 12 20:53:28 2005 (24387) Low level smtp error: (4, 'Interrupted 
>>system call'), msgid: <43262183.3010301 at wesmo.com>
>>    
>>
>
>
>This could be normal if SMTP_MAX_SESSIONS_PER_CONNECTION = 0 since in
>this case, Mailman (SMTPDirect) will never close the connection from
>its side. However, the specific error message comes out of the
>bulkdeliver() method which indicates Mailman is still involved in an
>SMTP transaction. I.e., it is probably waiting for a response from
>sendmail.
>  
>

Ahh.. now that makes sense.

>
>>With 
>>all of that in mind, I set :
>>SMTP_MAX_SESSIONS_PER_CONNECTION = 1
>>MAX_DELIVERY_THREADS = 0
>>SMTP_MAX_RCPTS = 1
>>
>>And things started working.  Odd..
>>    
>>
>
>
>Not so odd yet, I think.
>
>
>  
>
>>Now, I know that this is definitely NOT ideal or optimal.. But, what 
>>could be going wrong?  I tried reducing the 
>>SMTP_MAX_SESSIONS_PER_CONNECTION to "0" for my tests, and the problem 
>>crept back.
>>    
>>
>
>
>Now I think it's odd, but what happens in this case right after
>'mailmanctl restart'? Does it work for the first few mails and then
>the problem comes back. If so, it could be that the problem only
>occurs after a few transactions on an open connection.
>  
>

Yes, it works for the first few mails and then the problem re-appears.

>  
>
>>I moved that back to 1, then set SMTP_MAX_RCPTS to "0", and 
>>the problem crept back.  Only the config above seemed to work..
>>    
>>
>
>
>Now this I too find puzzling.
>
>
>
>Setting SMTP_MAX_SESSIONS_PER_CONNECTION = 1 is not particularly
>costly. It just means Mailman closes the SMTP connection after each
>transaction and reopens it for the next transaction. This is not too
>costly.
>
>Setting SMTP_MAX_RCPTS = 1 may impact performance, but maybe not as
>much as one would think. This list (mailman-users) uses personalized
>delivery which amounts to the same thing (at least for regular as
>opposed to digest deliveries). If you need SMTP_MAX_RCPTS = 1, then
>you've paid the price and you can VERP all deliveries (for better
>bounce processing) and personalize your lists for essentially no extra
>charge.
>

Well, I wanted to go back to using VERP anyhow, so I guess this is a 
good thing (in a way).  Everything is processing now, and, while these 
are mostly low-volume lists, there doesn't seem to be any noticable 
performance hit.

Thanks for your help!

-Rich