AnyEvent::Fork::Remote - remote processes with AnyEvent::Fork interface



   use AnyEvent;
   use AnyEvent::Fork::Remote;

   my $rpc = AnyEvent::Fork::Remote
      ->new_execp ("ssh", "ssh", "othermachine", "perl")
      ->require ("MyModule")
      ->run ("MyModule::run", my $cv = AE::cv);

   my $fh = $cv->recv;


Despite what the name of this module might suggest, it doesn't actually create remote processes for you. But it does make it easy to use them, once you have started them.

This module implements a very similar API as AnyEvent::Fork. In fact, similar enough to require at most minor modifications to support both at the same time. For example, it works with AnyEvent::Fork::RPC and AnyEvent::Fork::Pool.

The documentation for this module will therefore only document the parts of the API that differ between the two modules.


Here is a short summary of the main differences between AnyEvent::Fork and this module:

  • send_fh is not implemented and will fail

  • the child-side run function must read from STDIN and write to STDOUT

  • fork does not actually fork, but will create a new process


This example uses a local perl (because that is likely going to work without further setup) and the AnyEvent::Fork::RPC to create simple worker process.

First load the modules we are going to use:

   use AnyEvent;
   use AnyEvent::Fork::Remote;
   use AnyEvent::Fork::RPC;

Then create, configure and run the process:

   my $rpc = AnyEvent::Fork::Remote
      ->new_execp ("perl", "perl")
      ->eval ('
           sub myrun {
              "this is process $$, and you passed <@_>"
      ->AnyEvent::Fork::RPC::run ("myrun");

We use new_execp to execute the first perl found in the PATH. You'll have to make sure there is one for this to work. The perl does not actually have to be the same perl as the one running the example, and it doesn't need to have any modules installed.

The reason we have to specify perl twice is that the first argument to new_execp (and also new_exec) is the program name or path, while the remaining ones are the arguments, and the first argument passed to a program is the program name, so it has to be specified twice.

Finally, the standard example, send some numbers to the remote function, and print whatever it returns:

   my $cv = AE::cv;

   for (1..10) {
      $rpc->($_, sub {
         print "remote function returned: $_[0]\n";


Now, executing perl in the PATH isn't very interesting - you could have done the same with AnyEvent::Fork, and it might even be more efficient.

The power of this module is that the perl doesn't need to run on the local box, you could simply substitute another command, such as ssh remotebox perl:

   my $rpc = AnyEvent::Fork::Remote
      ->new_execp ("ssh", "ssh", "remotebox", "perl")

And if you want to use a specific path for ssh, use new_exec:

   my $rpc = AnyEvent::Fork::Remote
      ->new_exec ("/usr/bin/ssh", "ssh", "remotebox", "perl")

Of course, it doesn't really matter to this module how you construct your perl processes, what matters is that somehow, you give it a file handle connected to the new perls STDIN and STDOUT.


my $proc = new_exec AnyEvent::Fork::Remote $path, @args...

Creates a new AnyEvent::Fork::Remote object. Unlike AnyEvent::Fork, processes are only created when run is called, every other method call is is simply recorded until then.

Each time a new process is needed, it executes $path with the given arguments (the first array member must be the program name, as with the exec function with explicit PROGRAM argument) and both STDIN and STDOUT connected to a communications socket. No input must be consumed by the command before perl is started, and no output should be generated.

The program must invoke perl somehow, with STDIN and STDOUT intact, without specifying anything to execute (no script file name, no -e switch etc.).

Here are some examples to give you an idea:

   # just "perl"
   $proc = new_exec AnyEvent::Fork::Remote
      "/usr/bin/perl", "perl";

   # rsh othernode exec perl
   $proc = new_exec AnyEvent::Fork::Remote
      "/usr/bin/rsh", "rsh", "othernode", "exec perl";

   # a complicated ssh command
   $proc = new_exec AnyEvent::Fork::Remote
      qw(ssh -q
         -oCheckHostIP=no -oTCPKeepAlive=yes -oStrictHostKeyChecking=no
         -oGlobalKnownHostsFile=/dev/null -oUserKnownHostsFile=/dev/null
         exec perl);
my $proc = new_execp AnyEvent::Fork::Remote $file, @args...

Just like new_exec, except that the program is searched in the $ENV{PATH} first, similarly to how the shell does it. This makes it easier to find e.g. ssh:

   $proc = new_execp AnyEvent::Fork::Remote "ssh", "ssh", "otherhost", "perl";
my $proc = new AnyEvent::Fork::Remote $create_callback

Basically the same as new_exec, but instead of a command to execute, it expects a callback which is invoked each time a process needs to be created.

The $create_callback is called with another callback as argument, and should call this callback with the file handle that is connected to a perl process. This callback can be invoked even after the $create_callback returns.

Example: emulate new_exec using new.

   use AnyEvent::Util;
   use Proc::FastSpawn;

   $proc = new AnyEvent::Fork::Remote sub {
      my $done = shift;

      my ($a, $b) = AnyEvent::Util::portable_socketpair
         or die;

      open my $oldin , "<&0" or die;
      open my $oldout, ">&1" or die;

      open STDIN , "<&" . fileno $b or die;
      open STDOUT, ">&" . fileno $b or die;

      spawn "/usr/bin/rsh", ["rsh", "othernode", "perl"];

      open STDIN , "<&" . fileno $oldin ;
      open STDOUT, ">&" . fileno $oldout;

my $proc = new_from_fh $fh

Creates an AnyEvent::Fork::Remote object from a file handle. This file handle must be connected to both STDIN and STDOUT of a perl process.

This form might be more convenient than new or new_exec when creating an AnyEvent::Fork::Remote object, but the resulting object does not support fork.

$new_proc = $proc->fork

Quite the same as the same method of AnyEvent::Fork, except that it simply clones the object without creating an actual process.

undef = $proc->pid

The pid method always returns undef and only exists for compatibility with AnyEvent::Fork.

$proc = $proc->send_fh (...)

Not supported and always croaks.

$proc = $proc->eval ($perlcode, @args)

Quite the same as the same method of AnyEvent::Fork.

$proc = $proc->require ($module, ...)

Quite the same as the same method of AnyEvent::Fork.

$proc = $proc->send_arg ($string, ...)

Quite the same as the same method of AnyEvent::Fork.

$proc->run ($func, $cb->($fh))

Very similar to the run method of AnyEvent::Fork.

On the parent side, the API is identical, except that a $cb argument of undef instead of a valid file handle signals an error.

On the child side, the "communications socket" is in fact just *STDIN, and typically can only be read from (this highly depends on how the program is created - if you just run perl locally, it will work for both reading and writing, but commands such as rsh or ssh typically only provide read-only handles for STDIN).

To be portable, if the run function wants to read data that is written to $fh in the parent, then it should read from STDIN. If the run function wants to provide data that can later be read from $fh, then it should write them to STDOUT.

You can write a run function that works with both AnyEvent::Fork and this module by checking fileno $fh. If it is 0 (meaning it is STDIN), then you should use it for reading, and STDOUT for writing. Otherwise, you should use the file handle for both:

   sub run {
      my ($rfh, ...) = @_;
      my $wfh = fileno $rfh ? $rfh : *STDOUT;

      # now use $rfh for reading and $wfh for writing


AnyEvent::Fork, the same as this module, for local processes.

AnyEvent::Fork::RPC, to talk to the created processes.

AnyEvent::Fork::Pool, to manage whole pools of processes.


 Marc Lehmann <>