On Mon, 12 May 1997, John Nowack wrote:
> >pmmon has this nasty little problem of kicking off users, then not
> >allowing them on for anywhere from 5-15 minutes. I've checked all
> >the settings, and can't figure out how to fix it. I've got enough
> >other shit todo without having to debug someone else's code too.
When dealing polling multiple items, you encounter many race-conditions
on some of these issues. Some of them took me longer to find than other
'cuase not everyone sends me a message about a problem they may have
encountered. This above problem comes from the 2.1 "Beta"
version that is out. It was a problem I didn't think of right off. This
problem usually occurs when pmmon loses connection to a terminal server
(for whatever reason). It will then attempt to reconnect. When this
occurs, it left the users in the usertable that it maintains. When the
user logs back in, else where it then got a "false" multiple login.
This problem is now fixed. It now doesn't count users that are
connected to a non connected terminal server. I'm sure you could have
this same type of problem with a hacked radius version that lost an
accounting packet that indicated a user logged out. The advantage of
polling is that when it does get back in, it updates its state to the current
state, where-as with a lost accounting packet, when will you determine the
logout. When a new user logs into that port, and you get a collision in
your usertable?
Why wasn't this a problem in the 2.0, well in the 2.0 it blocked on
connect, so pmmon would just sit at connect until it succeeded. With the
2.1 it is non-blocking on connect, making it monitor and connect to all
term-servers concurrently instead of just monitoring concurrently.
> It seems that when pmmon kicks off a user, it wants to see them NOT on
> the next time it checks. If it drops a user, then 10 minutes later checks
> again and find that user connected on the same PM, same port (very likely
> if they immeadiatly redialed), then it seems to assume that they did not
> get booted and boots them again. If they hit a different port, or do not
> reconnect for more than a few minutes they are fine.
>
I don't know what version you are running here on pmmon, but when the
user is reset, it clears out the information about the user. If the user
is on the next time it should count them as a new login. This may be
related to other race conditions. The longer the time between polls
(configurable) the more likely you are to encounter them. Also the more
items that you are polling then they are more likely.
FYI: there will be a new 2.1 Beta out in the next few days. Pending
anything major this will be pretty much the relase.
> John
>
> --
> John Nowack
> john@dpc.net
> DPConsultants, Inc.
> Sys Admin -- dpc.net
> 309-925-2451
>
----------------------------------------------------------------
Brad Owens
Product Development
thrasher@squashduck.com
http://www.squashduck.com/~thrasher/jwz/
Squashed Duck, Inc
4015 N. Woodlawn Suite 1
Wichita, KS 67220
Phone: (316) 652-8907 Fax: (316) 651-0076
Port Master Monitor Utility
http://www.pmmon.com/2.1/
http://www.squashduck.com/~thrasher/jwz/pmmon-2.0/
---------------------------------------------------------------