NetNews Usenet Archive 1992 #18/NN_1992_18.iso/spool/comp/unix/aix/8919

Internet Message Format | 1992-08-19 | 3.7 KB

Path: sparky!uunet!charon.amdahl.com!pacbell.com!mips!sdd.hp.com!uakari.primate.wisc.edu!ames!riacs!pioneer.arc.nasa.gov!lamaster
From: [e-mail] (Hugh LaMaster)
Newsgroups: comp.unix.aix
Subject: Re: runaway processes (was Re: Vi is still broken)
Message-ID: <[e-mail]>
Date: 19 Aug 92 16:32:18 GMT
References: <[e-mail]> <[e-mail]> <[e-mail]> <[e-mail]>
Sender: [e-mail]
Organization: RIACS, NASA Ames Research Center
Lines: 65

In article <[e-mail]>, [e-mail] (Steve Losen) writes:
|> [ complaints about runaway processes eating up cpu time ]
|>
|> In article <[e-mail]>,
|> |>
|> |> Yup.  We see this kind of thing frequently.  Since we are about to start
|> |> chargeback accounting, this will be a severe pain.
|> |>
|> |> Our most recent method of producing the problem is to close a window in
|> |> which a process which is on the receiving end of a pipe is running.
|> |>
|> |> This has been going on for years but has lately become intolerable.  I
|> |> must now try to come up with a reproducible example and phone up IBM
|> |> to see if it's a 'supported defect' :-}
|>
|> We've had this problem since day one way back at AIX 3.1.  No AIX upgrade
|> has fixed it yet.  In our case, the runaways are all interactive jobs such
|> as editors, mail readers, news readers, etc., and they all seem to happen
|> when a telnet session ends abnormally.


I assumed that this was a problem since way back, probably in 4.2 BSD.
I have seen it for years.  Mail readers seem especially
vulnerable.  I have experienced the same or a similar problem on SunOS and
Ultrix as well, including very recently (SunOS 4.1.1 and Ultrix 4.2).

|>
|> Very early on I wrote a perl script that runs "ps caux" every few minutes
|> and looks for runaways.  I have a "hit list" of interactive commands that
|> are known to runaway, including vi, jove, more, less, telnetd, rlogind,
|> mail, mush, etc.  The script kills off any of these commands if ps
|> indicates that it is using over 9% of the cpu and has accumulated 2
|> minutes of cpu time.  I just pulled these heuristics out of thin air, but
|> they have worked well on several loaded 540s and 550s.
|>
|> Sure beats getting called up several times a day to kill these things off.
|>
|> I would post the perl script, but it has grown very large because it does
|> a whole lot of other stuff such as renicing long running cpu burners,
|> detecting when a user is running >1 cpu burner at a time, etc.  Also, I
|> will have to fix the script to run under 3.2.  IBM has changed the output
|> format of ps.  Thankfully the new format is easier to parse.  Under 3.1.5,
|> some of the fields can run together.  I think under 3.2, you are always
|> guaranteed at least one space of separation.

This sounds useful.  If you are feeling generous, you might post it to
comp.sources.something one of these days.  A systematic, generic problem
like this ought to be cleaned up by the vendors if they expect to be able
to market clusters of Unix boxes as alternatives to mainframes.  In a
production shop you can't very well just go killing off processes which
happen to exceed the confines of a heuristic, but for now, it could be
very useful.


|>
|> --
|> Steve Losen                     [e-mail]
|>
|> University of Virginia Academic Computing Center

--
  Hugh LaMaster, M/S 233-9,     UUCP:      ames!lamaster
  NASA Ames Research Center     Internet:  [e-mail]
  Moffett Field, CA 94035-1000  Or:        [e-mail]
  Phone:  [phone]                     #include <usenet/std_disclaimer.h>