errors when running multiple openssh sessions

[Ben Lindstrom]

Sounds like someone is not reading the manpages. <smile>

man sshd_config
[..]
     MaxStartups
             Specifies the maximum number of concurrent unauthenticated con-
             nections to the sshd daemon.  Additional connections will be
             dropped until authentication succeeds or the LoginGraceTime ex-
             pires for a connection.  The default is 10.

             Alternatively, random early drop can be enabled by specifying the
             three colon separated values ``start:rate:full'' (e.g.,
             "10:30:60").  sshd will refuse connection attempts with a proba-
             bility of ``rate/100'' (30%) if there are currently ``start''
             (10) unauthenticated connections.  The probability increases lin-
             early and all connection attempts are refused if the number of
             unauthenticated connections reaches ``full'' (60).


Default is '10'.. I bump it up to 20 and your script works fine.

- Ben


On Mon, 16 Jun 2003 jcduell at lbl.gov wrote:

> On Mon, Jun 16, 2003 at 01:08:59PM -0700, Dan Kaminsky wrote:
> > jcduell at lbl.gov wrote:
> >
> > >   #!/bin/sh
> > >
> > >   for NUM in 0 1 2 3 4 5 6 7 8 9; do
> > >       ssh n2003 echo $NUM "$*" &
> > >   done
> > >
> > >So, we're running 10 ssh commands at once.
> i>
> > That invocation creates quite a spike in CPU usage.
>
> Thanks for replying to my fragment of a post (I tried another test that
> turned out to fork itself recursively, and I managed to accidentally
> send my mail while trying to kill the resulting job horde:  talk about a
> CPU spike ;)
>
>
> > Who knows, it might also be causing headaches for privsep.  A better
> > way to do what you describe above would be this:
> >
> > #!/bin/sh
> >
> > for NUM in 0 1 2 3 4 5 6 7 8 9; do
> >        echo $NUM "$*" \&
> > done | ssh n2003
> >
> > From what I've found, this is the best way to execute sets of commands
> > remotely without significant CPU load.
>
> As my full post (below) notes, I'm writing a compiler that needs to use
> ssh to as part of compilation.  If a user calls my front end with the
> equivalent of
>
>     gcc foo.c bar.c
>
> then my script will do only one ssh for both files, but most people
> write makefiles that compile each file separately.
>
> Here's my full post.  I'm creating a bug for this in Bugzilla, too,
> since I don't see anything resembling it in the existing bugs.
>
> ------------------------------------------------------------------------
> Openssh seems to fail sporadically if you issue lots of simultaneous
> ssh commands.  Take the following program:
>
>     #!/bin/sh
>
>     for NUM in 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15; do
>         ssh foo.bar.com echo $NUM &
>     done
>
> So, we're running 16 ssh commands at once, each of which just prints out a
> different number.
>
> When I run this program, several of the ssh commands fail with
>
>     ssh_exchange_identification: Connection closed by remote host
>
> Interestingly, when I run 10 or fewer ssh commands, they all work OK, at
> least on my linux box (I'm using OpenSSH 3.5p1-6 on Redhat Linux 9).  On
> some other platforms the number is different:  OpenSSH 3.2.3p1 on an IBM
> SP doesn't like more than 8 simultaneous ssh's in the background, while
> OpenSSH_3.6.1p1 on Tru64 does around 9 max.
>
> There doesn't seem to be any pattern in terms of which ssh's get
> killed--the first, second and third jobs (ie, those that print 0, 1, and
> 2) generally run OK, but which of the following ones die seems to be
> random.
>
> This smells like some kind of race condition.
>
> Why on earth would I want to run a dozen ssh jobs simultaneously?  I'm
> writing a compiler that needs to ship some files and run some commands
> on a remote server as part of the compilation process.  The latency for
> doing this is rather high, so I want to allow users to do a 'make -j' to
> parallelize the build, in order to hide the network latency.
>
> I guess for now I'll tell users to run 'make -j N' with N < 6 or so
> (which is probably not a bad idea anyway).  But I can't imagine I'll be
> the last person to pound on ssh like this...
> ------------------------------------------------------------------------
>
> Cheers,
>
> --
> Jason Duell             Future Technologies Group
> <jcduell at lbl.gov>       Computational Research Division
> Tel: +1-510-495-2354    Lawrence Berkeley National Laboratory
>
> _______________________________________________
> openssh-unix-dev mailing list
> openssh-unix-dev at mindrot.org
> http://www.mindrot.org/mailman/listinfo/openssh-unix-dev
>