Re: Bug#97336: ITP: dsh -- dancer's shell, or a distributed shell.

["Eray Ozkural \(exa\)" <erayo@cs.bilkent.edu.tr>]

Hi Rob,

Rob Latham wrote:
> 
> it's a bit more fun (though harder to sort the output) if you do
> 
> for a in $(seq 1 100); do rsh -n note-$a w & ; done
> 
> and the obvious refinement is to prefix each line with the node name,
> then use sort with whatever parameters make it sort on the first
> column with no regard for other columns.
> 
> it's those "whatever parameters" that keep me from backgrounding N rsh
> jobs on a regular basis.  but for lots of tasks i can get away with it
> ( for example, the emulex fiber channel driver takes 10 seconds to
> load.  on a 33 node cluster, that adds up fast )

Here's the script I use for sequential execution. It's the most robust
I've come up with. :P

borg:exa$ cat bin/doborg
#! /bin/sh

#for x in `cat ~/bin/nodes`; do
#  rsh node$x $1;
#done

# mass ping all nodes

gen-nodes | strobe -p 514 -t 1 | grep "514 shell" | awk '{print $1;}' \
>/tmp/borg-upnodes-$USER

#trap "killall rsh" INTERRUPT

for x in `cat /tmp/borg-upnodes-$USER`; do
  echo "remote execution on" $x
  rsh $x $@;
done

I use another script called gen-nodes to generate a sequence of nodes, I don't
know. Kind of felt better abstraction that way. Using strobe works quite
fine if you'd try for yourself.

One thing: how would you do this fault tolerant with a shell script. I could
write a very robust cluster shell frontend in c/c++ but have no idea how you
could do that with a simple shell script.

Perhaps writing it in perl or tcl/tk would be a better idea.

Though I know that you could do quite complicated stuff just with a bash
script (mklibs.sh) :)

Thanks,

-- 
Eray Ozkural (exa)
Comp. Sci. Dept., Bilkent University, Ankara
e-mail: erayo@cs.bilkent.edu.tr
www: http://www.cs.bilkent.edu.tr/~erayo