use Net::OpenSSH::Parallel;

my @hosts = (...);

my $pssh = Net::OpenSSH::Parallel;

# register the list of hosts:
$pssh->add_host($_) for @hosts;

# declare the commands you want to run in every host:
$pssh->push('*', cmd => 'do.this');
$pssh->push('*', cmd => 'do.that');
$pssh->push('*',
            { stdout_file => '%HOST%.csv' }, # %HOST% is replaced by t
+he host name
            cmd => 'make.csv');

# let Net::OpenSSH::Parallel take care of everything:
$pssh->run;

# finally, collect the errors and the temporary CSV files.
for my $host (@hosts) {
  my $error = $pssh->get_error($host);
  if ($error) {
    print STDERR "Failed to generate CSV from host '$host': $error\n";
  }
  else {
    system cat => "$host.csv";
    unlink "$host.csv";
  }
}
[download]

update: code adjusted according to some of the comments by serf below.

Meh about:

    system 'cat $host.csv';
    system 'rm $host.csv';
[download]

You have open and print, you have unlink, and you have ways of checking that what you wanted to do worked correctly.

Perl isn't a shell script.

I would suggest that it's more reliable, more efficient, more cross-OS portable, and generally better practice to do these things natively in Perl than shelling out via system.

I get given grief at my $JOB by "scripts" that people who don't know Perl have written in Perl which should be in Perl but are full of awk and sed and grep and wc pipes and other badness. These scripts spawn many more processes than they need to, use more memory, and end up doing knarly things like jamming up, failing silently or causing excessive IO over NFS etc, bringing down our systems.

When we post code on perlmonks, people who don't know as much as we do will take what we've written without understanding it and use it thinking it's the *right* way to do it. I think we should try and be responsible and steer them in the right direction.

/me dismounts soapbox

Morning again Monks! Thank you so much for the replies. This module sounds just what im looking for. Im not too bothered about things being a little dirty aslong as it gets it done quickly and I get decent output. I should add its a windows server 2003. ralph

I'm not saying not to do this in Perl. However, there are many tools already for this sort of task.

Are you building a distributed application? Something like RabbitMQ or some other information queue or job queue may help. There's OpenMPI and the Parallel::MPI or Parallel::MPI::Simple library wrapping it if you need that sort of parallel interaction.

Are you doing configuration of the systems? Is this a one-off thing or will it grow to manage more things? Have you looked at GNU Parallel? How about multissh? CFEngine may work for you, too. On the really comprehensive end there are Saltstack, Ansible, Chef, or Puppet. They'll manage almost every aspect of every system across a whole server farm and report status back about everything they've done. Rex is a configuration management system that supports Windows, Linux, the BSDs, Solaris, and OS X and is written in Perl even. I'll admit I don't have any experience with Rex myself, but if you're wanting something in Perl you don't have to write from scratch it may be worth a long, hard look.

There's a comparison page for open source configuration management systems on Wikipedia.