strange inn "crash"
paul at atgi.net
Sat Apr 15 03:26:17 UTC 2000
ah. as the kid's say these days, 'my bad'. yeah, i had the hard limit
cranked way up (so high i'm embarrassed to admit the value!).
It's strange though, at almost the exact same time (according to an email
exchange), one of my newspeers in the midwest U.S. experienced an almost
identical sort of failure - syslog bailed, innd essentially hung, with
virtually no traffic passing down the pipe. For a while there, we were both
wondering whether this might be some new kind of DoS of nntp.
At 05:37 PM 4/14/00, Russ Allbery wrote:
>Paul Theodoropoulos <paul at atgi.net> writes:
> > my transit server locked up this morning, with the following:
> > Apr 14 10:50:00 HSNX innd: 220.127.116.11 connected 1023 streaming allowed
> > Apr 14 10:50:00 HSNX innd: 18.104.22.168 connected 1024 streaming allowed
> > Apr 14 10:50:00 HSNX innd: SERVER cant select Invalid argument
>You went over 1024 file descriptors (those numbers are the file descriptor
>number). Are you running Solaris? Don't ever set the hard file
>descriptor limit on Solaris over 1024 so long as innd uses select instead
>of poll, or you'll get the above problem.
>There should probably be some sanity checks put into innd for this, but I
>don't know of a good recovery method beyond just throttling the server.
>Russ Allbery (rra at stanford.edu) <http://www.eyrie.org/~eagle/>
Paul Theodoropoulos Advanced TelCom Group, Inc.
Senior Unix Systems Administrator Internet Services
Work: http://www.atgi.net Play: http://www.anastrophe.com
"Reliability means never having to say you're sorry." -D.J. Bernstein
More information about the inn-workers