Very Computer

I'm still fighting processor usage on one of our internet servers,
5.0.4 with Fast Start. All email clients are POP3 virtual domain
types.

cpuhog shows that popper is what's sucking cycles, and they hang
around forever until they have the whole machine. sar shows idle
of 0% after the machine has been up for 2-3 hours, which is obviously
wrong, as it runs fine and is very quick. It does slow down some,
but nothing like sar says it should be - 97% sys usage. (!)

This is having secondary effects, in that sendmail requires me to keep
upping the queue trigger levels, and sometimes wedges and as to be
restarted. I mentioned more detail on this in an earlier message.

In any case, popper's man page mentions a -d option which pertains to
time-to-livem but it neglects ranges or what the default might be. It
says it gets it from a bvariable in a config file, but I can't find
it.

Most of these users are PPP-dialup, their ping rates are in the low 200
ms range from server-to-client, so WHAT should popper -d be set to so
the damned things will die before they take over the system?

Is there some special tuning I should do to 5.0.4 since I'm running
Fast Start on it? Regular (5.0.2 based) Fast Start 1.0 ran like a
bear- never had these weird little problems with it.

Please do point me somewhere; right now we're dealing with this by
watching it and fixing it by hand, 24/7

'Shot

Bill
--

UUCP:              camco!bill   2835 82nd Avenue S.E. S-100
FAX:           (206) 232-9186   Mercer Island, WA 98040; (206) 947-5591

"The strongest reason for the people to retain the right to keep and
 bear arms is, as a last resort, to protect themselves against tyranny
 in government." --1 Thomas Jefferson Papers, 334 (C.J.Boyd, Ed.,1950)

Quote:> This is having secondary effects, in that sendmail requires me to keep
> upping the queue trigger levels, and sometimes wedges and as to be
> restarted. I mentioned more detail on this in an earlier message.

> In any case, popper's man page mentions a -d option which pertains to
> time-to-livem but it neglects ranges or what the default might be. It
> says it gets it from a bvariable in a config file, but I can't find
> it.

  # cd /etc
  # mv popper popper.real
  # cat > popper
  #!/bin/ksh
  ulimit -St 10    # die after 10 seconds of CPU time

(I'm assuming that /etc/popper is the path of the binary you're
running...)

I can also suggest a diagnostic, which is likely to be effective only if
you have not put the above workaround in place.  Get scotruss from
ftp://ftp.dgii.com/users/erics/scotruss.tar.Z; get lsof from
ftp://vic.cc.purdue.edu/pub/tools/unix/lsof/.  When you see one of these
spinning popper processes, run lsof and scotruss on it:

  lsof -p1234
  scotruss -p1234

lsof will hopefully tell you what files it has open.  scotruss will show
you what system calls it's making (you'll have to interrupt it after a
while, or scotruss will faithfully tell you about the infinite number of
system calls popper is making).  If you can't make sense of the results,
someone here probably can.

Quote:> Most of these users are PPP-dialup, their ping rates are in the low 200
> ms range from server-to-client, so WHAT should popper -d be set to so
> the damned things will die before they take over the system?

> Is there some special tuning I should do to 5.0.4 since I'm running
> Fast Start on it? Regular (5.0.2 based) Fast Start 1.0 ran like a
> bear- never had these weird little problems with it.

I've seen many cases of spinning poppers, generally caused by the pop
user rebooting his Windows machine.  That was on a LAN.  Presumably
similar effects ensue if they hang up the modem without ending their pop
session.  Whatever it is, it doesn't happen every time; there are some
other enabling conditions (which I never figured out -- the LAN
situation was fixed by using a different popper binary which,
empirically, was immune to whatever the fault conditions were).

Quote:>Bela<

What was your reason for using ipop3d instead? IMAP? Or is SCO's popper
actually broken?

I have managed to fix it, just an hour ago. At least, load is typical
and the popper isn't* out anymore than its timeout says it
should.

The fix was to add a space to the end of the command /etc/popper -T 90
in the /etc/inetd.conf.

Is this a bug or a feature? ;-)

I'll post an update later; it's almost too soon to tell. I do think
it might be fixed, though- we have about a hundred users popping mail
right now and only 3 poppers show under ps -e. Before adding the space,
we'd have 25 or so by now..

I'll check out the ipop3d, Bill, and I'd appreciate any info and
suggestions on switching to it.

'Shot

<much snipped>

Quote:

> I've seen many cases of spinning poppers, generally caused by the pop
> user rebooting his Windows machine.  That was on a LAN.  Presumably
> similar effects ensue if they hang up the modem without ending their pop
> session.  Whatever it is, it doesn't happen every time; there are some
> other enabling conditions (which I never figured out -- the LAN
> situation was fixed by using a different popper binary which,
> empirically, was immune to whatever the fault conditions were).

> >Bela<

Anyway, in my earlier messages I mentioned the -d option as being
timeout; it's actually -T and I used the correct one. Just had a
brainlock when I was writing about it. As I mentioned in an earlier
off-thread message, adding a space to the end of the popper command
in inetd.conf seems to have fixed it. Since I did this, I've seen
it spin up to 10 poppers, but they all went away as they should have.

Bug or feature? ;-)

'Shot

Quote:>I have managed to fix it, just an hour ago. At least, load is typical
>and the popper isn't* out anymore than its timeout says it
>should.

>The fix was to add a space to the end of the command /etc/popper -T 90
>in the /etc/inetd.conf.

We run all our inetd stuff through tcp_wrappers, generally with the timeout
option to have an external control to kill off dead Microsoft originated
processes.  It also allows us to limit access to things like ipop3d, imapd,
etc. to specific domains.

...

Quote:>I'll check out the ipop3d, Bill, and I'd appreciate any info and
>suggestions on switching to it.

        kill -HUP `cat /etc/inetd.pid'

The imap programs insert a dummy message in each mailbox as a place holder
to allow them to see what the last message they've read is.  This can cause
some gripes by people who read mail locally as well as from POP clients.

The imap programs understand both MMDF and standard mailbox formats.  THe
version I've compiled default to standard instead of MMDF which only has an
effect if they find an empty mailbox.

Bill
--

UUCP:               camco!bill  PO Box 820; 2835 82nd Avenue S.E. S-100
FAX:            (206) 232-9186  Mercer Island, WA 98040-0820; (206) 236-1676
URL: http://www.veryComputer.com/

It is necessary for the welfare of society that genius should be
privileged to utter sedition, to blaspheme, to outrage good taste, to
corrupt the youthful mind, and generally to scandalize one's uncles.
                -- George Bernard Shaw

Change it back and it'll keep working cleanly (unless you've by now
discovered that it isn't really fixed...)

Quote:>Bela<