[Mailman-Users] Mailman b5 lockup

[Ron Jarrell]

At 12:47 AM 9/5/00 -0700, Chuq Von Rospach wrote:
>At 3:35 AM -0400 9/5/00, Ron Jarrell wrote:
>
>>There's about 80 gazillion timeout settings in sendmail, all of which default to Big Numbers
>>if you don't tweak them; and it also defaults to dns validating every address it's given.
>>If your sendmail usage doesn't match that profile, adjust sendmail!
>
>
>Ron, how do you have yours configured? I'm curious -- I use sendmail to send out really huge lists with very little configuration or problem. But I'm always interested in new techniques. Wanna share some? Unfortunately, just saying "fix sendmail" is like telling someone who's looking for a tune-up to go rebuild their own engine, when they aren't even sure how to do a tune-up in the first place...

Mine mostly works right too.  But my main mail relay is currently set to split the queue
load amongst 5 queues, and I further split the queues into the seperate components. (qf,df, etc)
I'm running with FEATURE(nocanonify).  I don't generally have much of a problem with 
DNS load, so I have commented out, but ready if I need it a FEATURE(accept_unresolvable_domains),
which gets rid of that initial DNS lookup entirely.  I also run an MCI_CACHE_SIZE of 30,
with a timeout of 35, and have the QUEUE_SORT_ORDER set to host.  (Although I'll
go in every few days and run a special queue run sorted by time with higher timeouts,
to catch problematic hosts.  I use a hoststat directory, set at 10 minutes.

For other interesting timeouts, I lowered the IDENT timeout to 15 seconds (too many
PCs were broken w.r.t things hitting their ident ports, and would hang up for the minute
it used to be).   I cranked my ICONNECT down to 30s; if they're not answering by then,
they can just go in the queue and wait.  My initial and connects are at 2m, which I think
is lower than the default.  Mail is at 10m, and misc at 2m.  Quit 2m, rset 5m.  I left
rcpt at the default of 1h, because I rarely have a problem with that, althought I've observed
some seriously broken mailers take 15 minutes or more to figure out they have a host.
Like Yahoo, Hotmail, and Excite...  

Most of my other settings are defaulted, (at least in timeouts), but with regular DNS delays
I'd consider either accepting unresolvable domains, or looking at 11.0's new timeout.resolver.*
section... There's a retrans and a retry timeout, and they're further dividable into first 
attempt, and normal.  I usually set first attempts down really low, for fast initial entry into the
system; well connected, non overloaded machines will still respond.  If they don't well,
that's what the queue is for, and if a lot of my users are mailing there, then I get the
benefits of delivering multiple messages for the cost of one connection.

The place where this occaisionally breaks down is when yahoo, excite, or hotmail have
problems; although it's yahoo more than the others in the last couple of months.  They'll
get into a mode where they'll take a message or two, then kick out an i/o error, so the
queue builds up because it thinks the host keeps going down.  I end up having to do a 
special queue run that just pounds just them over and over and over until they finally go
away.

Our listserv machine is actually much more liberal in timeouts, and still performs
well.  But then, a large chunk of it's traffic is actually distributed outside the
normal channels, listserv to listserv...
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.python.org/pipermail/mailman-users/attachments/20000905/db83f50c/attachment.html