> Date: Mon, 18 Sep 2000 09:36:54 +0200 (MET DST)
> From: Eric Luyten <[EMAIL PROTECTED]>
> To: Subscribers of Qpopper <[EMAIL PROTECTED]>
> Subject: Re: Time between pops
>
> > > From: Brad Groshok [mailto:[EMAIL PROTECTED]]
> ...
> > > Is there a way to control how often a user can read their mail.
>
>
> Be warned : I have seen Qpopper log extracts demonstrating that some
> clients perform two POP3 accesses in rapid succession. That behaviour
> may be beyond end user control.
>
The most egregious example of this is mailstart.com, which does one
complete pop connect per user action, ie the user checks their email
(pop connect) and mailstart shows them the subject lines; the person
selects a message a few seconds later (pop connect again to get that
message), etc, etc. My "-a accesstime" setting can be used to torture
users who insist on this horror.
The real problems I had with mailstart though was (a) their DNS setup, and
(b) FIN_WAIT_2 socket connections. Reverse lookups to their site
don't work, so if qpopper is doing reverse lookups then these connections
will sit there for awhile. If you have a lot of people (50+) using mailstart
to get to your pop server, then FIN_WAIT_2 connections will kill you.
Each one of those many connections that mailstart makes leaves a fin_wait_2
hanging around in your tcp/ip stack (do a "netstat -an -f inet" to see).
If your fin_wait_2 timeout is "forever" then you are in deep trouble because
they will eventually exhaust kernel resources. If you set the timeout
to 10 minutes (the Berkeley standard), but have a lot of connections
they will still stack up faster than you can get rid of them via timeout.
I once had nearly 1000 lingering FIN_WAIT_2 connections from mailstart,
cough, causing me to have to reboot my mail/pop server.
My solution: use tcpwrappers with qpopper in inetd, and then block out
pop connections from mailstart with /etc/hosts.deny, eg:
popper:207.231.76.
No connections -- no problem. My mail server is nice and perky without
them around.
--- Jeff Earickson
