Grex is now running sendmail 8.6.9

Nov 26 20:56 UTC 1994

   If you encounter anything strange with mail, please put it here.
I was tempted to ask people to mail staff about problems, but it
occured to me that if we're having a problem with mail, mail to staff
might not be such a swift way to communicate...

   We're running a new mail system called sendmail.  This replaces the
smail program that we've used for a while now.  There are multiple
reasons for making the switch, and one of the more important is that
sendmail will impact on the systems load less, when we get several
peices of mail at once.

   But anything as complicated as sendmail takes some doing, and there
is the possibility that its going to need some tweaking along the way
to being completely stable.

   Its looking good right now--its handled several hundred messages
so far, and took off site mail from myself just fine, so I think things
are running pretty good.  The real test of course, is after Grex has
been down for several hours, and gets pounced upon by various sites
trying to mail us.

   Thanks to Marcus for compiling this and getting it moved over.
That took some effort to do.  In the process of installing sendmail,
the original smail files were tucked away, such that we can drop
back to smail if we have to.

response 2 of 106:

Nov 27 00:54 UTC 1994

general 95 linking as coop item 41...done.

response 4 of 106:

Nov 27 10:07 UTC 1994

can I say I haven't had any problems, or should I save it? :)

response 6 of 106:

Nov 28 01:44 UTC 1994

what are the differences between the old and the new?

response 8 of 106:

Nov 28 01:46 UTC 1994

   Re #6: Bascially, as far as you the user goes, no difference at all.
If you look closely at your mail now, compared with a week ago, the
headers will look at little different.  There is a signature of the
mailer, so that is different.
   I should say there is one difference that you should see, which is
that (hopefully) sendmail will be less of a pig here, and mail will get
to folks faster than before!

response 10 of 106:

Nov 28 07:50 UTC 1994

   Yup.

   Well, we had a pounding  earlier tonight, where the load average
hovered between 9 and 19, with the average about 12.  Now, back in "the
old days" of smail, we'd have likely have gotten load averages in the
40's, and we'd have stayed there for a while.

   As sendmail is currently configured, it starts queueing up
incomming mail once the load average hits 8.  This one of the smart
things that sendmail does, in comparison to smail--it won't merrily
send off nnn copies of itself clogging the system up.  If queueing
doesn't being the load average down, or if the system is just plain
busy, then when the load average hits 12, sendmail refuses new
connections.

   This does not mean that we're losing mail--we're just shutting
down 'till the load average gets back down below 12, before we start
taking new mail in.  So what happens when Grex refuses mail?  Well,
actually, it simply stops acknoledging incomming SMTP connects, so
other mailers think that Grex is down, and try connecting with us
again at a later time.

   It will be interesting to see what happens when Grex has been
off the net a while, with sendmail in operation.  I believe what will
happen is that we'll accept the horde of mail that always seems ready
to pounce on Grex after its been down for more than 5 minutes, and
the load average will hit 12 after a bit.  We'll then deal with mail
thats been queued up, and then start accepting mail again.  This cycling
will likely continue a couple of times until we get caught up.

   Quite a bit better than load averages of 70+ ...

response 12 of 106:

Nov 29 05:28 UTC 1994

I just got the following "bounced" mail.

This is a MIME-encapsulated message

--XAA04347.786084790/grex.cyberspace.org

    **********************************************
    **      THIS IS A WARNING MESSAGE ONLY      **
    **  YOU DO NOT NEED TO RESEND YOUR MESSAGE  **
    **********************************************

The original message was received at Mon, 28 Nov 1994 19:21:41 -0500
from raven@localhost

   ----- The following addresses had delivery problems -----                   
   
XXXXX@XXX.XXX  (transient failure)              
                                            
   ----- Transcript of session follows -----    
Warning: message still undelivered after 4 hours
Will keep trying until message is 5 days old

        I know the address is correct. Is this just because Grex is off
the net, or is it a problem with the new mail program.  I changed the
name of the receiopient to XXXX@XXX.XXX to protect his/her privacy.

response 14 of 106:

Nov 30 06:56 UTC 1994

Two things: (1) The mail isn't going out because the net connection
is broken. This causes it to  back up in the mail queue.
(2) Sendmail (unlike smail) is configured to generate these warning
messages to let you know that the mail is being slow getting to its
destination. THis is a nice feature, I think.

response 16 of 106:

Nov 30 08:26 UTC 1994

Oh, 'nother thing - - - I see no value in turning off the
+reception+ of new email from anywhere. I would presume that
the spool of new mail can be increased to the capacity of
free disk space. Why, then, cut off access? The spawning of
new processes to handle the spool INTERNALLY is the key to
the untoward load.

response 18 of 106:

Nov 30 12:56 UTC 1994

   There most certaily is value in turning off the ability of
receving new mail--that is part of the mechanism that will keep
Grex from getting load averages of 70+.
   To always accept new mail and then store it means that you have
to catch it, and catching it means a copy of sendmail running to
do this.  It is far better to sit there and reject messages for the
quarter of an hour (or whatever it is) than to say you'll accept
messages, but be so slow that you run the risk of having the other
system time out becuase you were so slow in responding.

response 20 of 106:

Dec 1 06:55 UTC 1994

   It's been an interesting hour and a half.  We got back on the
net at 12:15AM Thursday.  Within seconds we had 5 people on.  By
the time I noticed that the link was up, the load average had
climbed to 12; 30+ SMTP connections had been established to Grex!

   At 12:23, becuase of the *deluge* of smtp connections, the
load average shot up to a peak of 51.  By 12:26, after sendmail
had stopped taking new mail, the load average had dropped down to
12.5.  In the last 80 minutes or so, the load average has hovered
around 12, with peaks at 24 and valleys of 8.

   Remember now, this is an *extreme* condition--the electronic hordes
have been waiting to pounce on Grex for 50 hours, and they're all doing
it now.  So, I'd say sendmail is working *wonderfully*.

   When we get pounded on all over the world, the load average *is*
going to go up--its just a plain fact of life.  But, becuase sendmail
knows how to throttle itself back and not try to server the entire
worlds requests at once, the load average is liveable.

   I'm very *very* happy with this.  Getting sendmail up and running
came at just the right time, considering our 54 hour period of downtime.
I can only shudder if I think of what would have happened had smail
been in the picture: we'd have had hundreds (and I do mean hundreds)
of smails all trying to run, getting nowhere, running Grex completely
out of swap, and probably crashing the system after load averages of
100+ for an hour or two.

response 22 of 106:

Dec 1 10:57 UTC 1994

I'm very pleased that the load averages aren't in the stratosphere this time
around.  Very **very** pleased.

I'm also somewhat puzzled at the moment.  I subscribe to a couple of pretty
active mailing lists, & after some 4 hours (based on the time STeve entered
his notice) I still have *no* mail from either of them.  I wonder if messages
are queued up somewhere or lost, & whether I've been dropped from the lists
because of notices being continually passed back that Grex wasn't accepting
mail (or something like that).

At any rate: Thanks one more time, STeve (and anyone else who got in on this
one).  One more big time sink on top of all the others.

response 24 of 106:

Dec 2 00:25 UTC 1994

I got a phone call this evening from a user (I forgot to ask her name)
who says she hasn't received mail from the outside world in several days,
and mail she's sent out hasn't arrived.  She's gotten both the warning
message and genuine bounce messages.

My own mail doesn't seem to be coming and going properly either: I still
haven't received a message that should have been sent last weekend, and
mail I sent this morning to groups I'm in (eg. staff or cfadm) still hasn't
arrived.