Re: Out of memory crashes ssh

[Andrew Reid <reidac@bellatlantic.net>]

On Wednesday 20 May 2009 17:17:35 Maicon Faria wrote:
> Package: OpenSSH_5.1p1 Debian-5
>
> When I run a program, as user, that uses more then the memory
> avaliable the OOM(OUT-OF-MEMORY) kill this job.
> After that, ssh crashes and must be restarted by the root.
>
> Anyone has experienced something like that ?
> This happens in computers in a cluster where the nodes system comes
> from a nfs service. The kernel is a vanilla 2.6.26 one.

  I have hacked around a similar problem, also on a cluster,
but not really solved it.

  I have a cron job that looks for the SSH process (by trying
to ssh to localhost and run "echo"), and if it ever fails,
it assumes sshd has crashed, and restarts it.

  The problem with this approach is that, when the system 
runs out of memory, services tend to fail unpredictably, and
it sometimes happens that both cron *and* sshd fail.  That's
much more rare, of course.


  Somewhere on my "todo" list is to figure out memory 
management, and try to set up sshd so that it doesn't
get sacrificed when resources run low.

				-- A.
-- 
Andrew Reid / reidac@bellatlantic.net