Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!tut.cis.ohio-state.edu!rutgers!ucsd!ucsdhub!hp-sdd!ncr-sd!ncrlnk!uunet!mcvax!kth!enea!maxim!prc
From: prc@maxim.ERBE.SE (Robert Claeson)
Newsgroups: comp.databases
Subject: Re: Strange problems with Informix Perform
Keywords: Informix perform nohangup?
Message-ID: <482@maxim.ERBE.SE>
Date: 4 Feb 89 17:19:28 GMT
References: <34891@codas.att.com> <4723@sfsup.UUCP>
Organization: ERBE DATA AB
Lines: 41

In article <4723@sfsup.UUCP>, jdt@sfsup.UUCP (J Tais) writes:
> In article <34891@codas.att.com>, jaa@codas.att.com (James Anderson) writes:
> > A user brings up the perform screen, enters the Add or Update mode.
> > Something happens to the dialin port they are on and they get disconnected
> > from the system. The Informix process starts to run wild grabbing as much
> > usr and sys tics it can get (can tell by reading sar).
> > a Who shows the user still logged in with no idle time. A stat on the tty
> > usualy shows 0 idle read time and a write idle time (sometimes no write idle).

> I remember a similar problem on my last project, but it happened when users
> were rlogin'ed over TCP/IP and running perform on the remote machine.  We
> had a persistent problem with rogue perform processes grabbing all kinds of
> cpu time when users disconnected in abnormal ways or were terminated by the
> idle-line watcher.

I've seen this behaviour in much too many software packages -- Informix
and Oracle is just a few of them. What I think happens is that these
packages ignores SIGHUP and relies on the return code from the write()
and read() system calls to determine when a user has been disconnected.

On many machines, the return code is 0 when the disconnect occurs on a
dialup port (meaning "0 characters read/written") and -1 when a network
connection is disconnected (meaning "error"; errno is set to some reasonable
value). So these packages examines the return code, sees a 0  or a -1
and the  program logic decides  "heck, sumthin' went wrong, let's try
it again".  And off we go.

Some packages interprets the 0 return code as a hangup indication, while
they thinks that -1 is some kind of error and the fix is to retry the
operation until it succeeds.

Note that I don't say that this is what happens in all packages. I just
happens to know that this is the way it happens in some packages. In fact,
I haven't got the faintest idea about what Oracle and Informix does.
I've just seen it happen to both of them, but in Oracle's case only when
the disconnect occured on a TELNET/rlogin connection.
-- 
Robert Claeson, ERBE DATA AB, P.O. Box 77, S-175 22 Jarfalla, Sweden
"No problems." -- Alf
Tel: +46 758-202 50  EUnet:    rclaeson@ERBE.SE  uucp:   uunet!erbe.se!rclaeson
Fax: +46 758-197 20  Internet: rclaeson@ERBE.SE  BITNET: rclaeson@ERBE.SE