view release on metacpan or  search on metacpan

RPC.pm  view on Meta::CPAN

Note that the parentheses in the call to C<AnyEvent::Fork::RPC::event> are
not optional. That is because the function isn't defined when the code is
compiled. You can make sure it is visible by pre-loading the correct
backend module in the call to C<require>:

      ->require ("AnyEvent::Fork::RPC::Sync", "MyWorker")

Since the backend module declares the C<event> function, loading it first
ensures that perl will correctly interpret calls to it.

And as a final remark, there is a fine module on CPAN that can
asynchronously C<rmdir> and C<unlink> and a lot more, and more efficiently
than this example, namely L<IO::AIO>.

=head3 Example 1a: the same with the asynchronous backend

This example only shows what needs to be changed to use the async backend
instead. Doing this is not very useful, the purpose of this example is
to show the minimum amount of change that is required to go from the
synchronous to the asynchronous backend.

To use the async backend in the previous example, you need to add the
C<async> parameter to the C<AnyEvent::Fork::RPC::run> call:

      ->AnyEvent::Fork::RPC::run ("MyWorker::run",
         async      => 1,
         ...

And since the function call protocol is now changed, you need to adopt
C<MyWorker::run> to the async API.

First, you need to accept the extra initial C<$done> callback:

   sub run {
      my ($done, $cmd, $path) = @_;

And since a response is now generated when C<$done> is called, as opposed
to when the function returns, we need to call the C<$done> function with
the status:

      $done->($status or (0, "$!"));

A few remarks are in order. First, it's quite pointless to use the async
backend for this example - but it I<is> possible. Second, you can call
C<$done> before or after returning from the function. Third, having both
returned from the function and having called the C<$done> callback, the
child process may exit at any time, so you should call C<$done> only when
you really I<are> done.

=head2 Example 2: Asynchronous Backend

This example implements multiple count-downs in the child, using
L<AnyEvent> timers. While this is a bit silly (one could use timers in the
parent just as well), it illustrates the ability to use AnyEvent in the
child and the fact that responses can arrive in a different order then the
requests.

It also shows how to embed the actual child code into a C<__DATA__>
section, so it doesn't need any external files at all.

And when your parent process is often busy, and you have stricter timing
requirements, then running timers in a child process suddenly doesn't look
so silly anymore.

Without further ado, here is the code:

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

   my $done = AE::cv;

   my $rpc = AnyEvent::Fork
      ->new
      ->require ("AnyEvent::Fork::RPC::Async")
      ->eval (do { local $/; <DATA> })
      ->AnyEvent::Fork::RPC::run ("run",
         async      => 1,
         on_error   => sub { warn "ERROR: $_[0]"; exit 1 },
         on_event   => sub { print $_[0] },
         on_destroy => $done,
      );

   for my $count (3, 2, 1) {
      $rpc->($count, sub {
         warn "job $count finished\n";
      });
   }

   undef $rpc;

   $done->recv;

   __DATA__

   # this ends up in main, as we don't use a package declaration

   use AnyEvent;

   sub run {
      my ($done, $count) = @_;

      my $n;

      AnyEvent::Fork::RPC::event "starting to count up to $count\n";

      my $w; $w = AE::timer 1, 1, sub {
         ++$n;

         AnyEvent::Fork::RPC::event "count $n of $count\n";

         if ($n == $count) {
            undef $w;
            $done->();
         }
      };
   }

The parent part (the one before the C<__DATA__> section) isn't very
different from the earlier examples. It sets async mode, preloads
the backend module (so the C<AnyEvent::Fork::RPC::event> function is
declared), uses a slightly different C<on_event> handler (which we use
simply for logging purposes) and then, instead of loading a module with
the actual worker code, it C<eval>'s the code from the data section in the
child process.

It then starts three countdowns, from 3 to 1 seconds downwards, destroys
the rpc object so the example finishes eventually, and then just waits for
the stuff to trickle in.

The worker code uses the event function to log some progress messages, but
mostly just creates a recurring one-second timer.

The timer callback increments a counter, logs a message, and eventually,
when the count has been reached, calls the finish callback.

On my system, this results in the following output. Since all timers fire
at roughly the same time, the actual order isn't guaranteed, but the order
shown is very likely what you would get, too.

   starting to count up to 3
   starting to count up to 2
   starting to count up to 1
   count 1 of 3
   count 1 of 2
   count 1 of 1
   job 1 finished
   count 2 of 2
   job 2 finished
   count 2 of 3
   count 3 of 3
   job 3 finished

While the overall ordering isn't guaranteed, the async backend still
guarantees that events and responses are delivered to the parent process
in the exact same ordering as they were generated in the child process.

And unless your system is I<very> busy, it should clearly show that the
job started last will finish first, as it has the lowest count.

This concludes the async example. Since L<AnyEvent::Fork> does not
actually fork, you are free to use about any module in the child, not just
L<AnyEvent>, but also L<IO::AIO>, or L<Tk> for example.

=head2 Example 3: Asynchronous backend with Coro

With L<Coro> you can create a nice asynchronous backend implementation by
defining an rpc server function that creates a new Coro thread for every
request that calls a function "normally", i.e. the parameters from the
parent process are passed to it, and any return values are returned to the
parent process, e.g.:

   package My::Arith;

   sub add {
      return $_[0] + $_[1];
   }

   sub mul {
      return $_[0] * $_[1];
   }

   sub run {
      my ($done, $func, @arg) = @_;

      Coro::async_pool {
         $done->($func->(@arg));
      };
   }

The C<run> function creates a new thread for every invocation, using the
first argument as function name, and calls the C<$done> callback on it's
return values. This makes it quite natural to define the C<add> and C<mul>
functions to add or multiply two numbers and return the result.

Since this is the asynchronous backend, it's quite possible to define RPC
function that do I/O or wait for external events - their execution will
overlap as needed.

The above could be used like this:

   my $rpc = AnyEvent::Fork
      ->new
      ->require ("MyWorker")
      ->AnyEvent::Fork::RPC::run ("My::Arith::run",
         on_error => ..., on_event => ..., on_destroy => ...,
      );

   $rpc->(add => 1, 3, Coro::rouse_cb); say Coro::rouse_wait;
   $rpc->(mul => 3, 2, Coro::rouse_cb); say Coro::rouse_wait;

The C<say>'s will print C<4> and C<6>.

=head2 Example 4: Forward AnyEvent::Log messages using C<on_event>

This partial example shows how to use the C<event> function to forward
L<AnyEvent::Log> messages to the parent.

RPC.pm  view on Meta::CPAN

That means that if your parent process exits, the RPC process will usually
exit as well, either because it is idle anyway, or because it executes a
request. In the latter case, you will likely get an error when the RPc
process tries to send the results to the parent (because agruably, you
shouldn't exit your parent while there are still outstanding requests).

The process is usually quiescent when it happens, so it should rarely be a
problem, and C<END> handlers can be used to clean up.

=item Asynchronous Backend

For the asynchronous backend, things are more complicated: Whenever it
listens for another request by the parent, it might detect that the socket
was closed (e.g. because the parent exited). It will sotp listening for
new requests and instead try to write out any remaining data (if any) or
simply check whether the socket can be written to. After this, the RPC
process is effectively done - no new requests are incoming, no outstanding
request data can be written back.

Since chances are high that there are event watchers that the RPC server
knows nothing about (why else would one use the async backend if not for
the ability to register watchers?), the event loop would often happily
continue.

This is why the asynchronous backend explicitly calls C<CORE::exit> when
it is done (under other circumstances, such as when there is an I/O error
and there is outstanding data to write, it will log a fatal message via
L<AnyEvent::Log>, also causing the program to exit).

You can override this by specifying a function name to call via the C<done>
parameter instead.

=back

=head1 ADVANCED TOPICS

=head2 Choosing a backend

So how do you decide which backend to use? Well, that's your problem to
solve, but here are some thoughts on the matter:

=over 4

=item Synchronous

The synchronous backend does not rely on any external modules (well,
except L<common::sense>, which works around a bug in how perl's warning
system works). This keeps the process very small, for example, on my
system, an empty perl interpreter uses 1492kB RSS, which becomes 2020kB
after C<use warnings; use strict> (for people who grew up with C64s around
them this is probably shocking every single time they see it). The worker
process in the first example in this document uses 1792kB.

Since the calls are done synchronously, slow jobs will keep newer jobs
from executing.

The synchronous backend also has no overhead due to running an event loop
- reading requests is therefore very efficient, while writing responses is
less so, as every response results in a write syscall.

If the parent process is busy and a bit slow reading responses, the child
waits instead of processing further requests. This also limits the amount
of memory needed for buffering, as never more than one response has to be
buffered.

The API in the child is simple - you just have to define a function that
does something and returns something.

It's hard to use modules or code that relies on an event loop, as the
child cannot execute anything while it waits for more input.

=item Asynchronous

The asynchronous backend relies on L<AnyEvent>, which tries to be small,
but still comes at a price: On my system, the worker from example 1a uses
3420kB RSS (for L<AnyEvent>, which loads L<EV>, which needs L<XSLoader>
which in turn loads a lot of other modules such as L<warnings>, L<strict>,
L<vars>, L<Exporter>...).

It batches requests and responses reasonably efficiently, doing only as
few reads and writes as needed, but needs to poll for events via the event
loop.

Responses are queued when the parent process is busy. This means the child
can continue to execute any queued requests. It also means that a child
might queue a lot of responses in memory when it generates them and the
parent process is slow accepting them.

The API is not a straightforward RPC pattern - you have to call a
"done" callback to pass return values and signal completion. Also, more
importantly, the API starts jobs as fast as possible - when 1000 jobs
are queued and the jobs are slow, they will all run concurrently. The
child must implement some queueing/limiting mechanism if this causes
problems. Alternatively, the parent could limit the amount of rpc calls
that are outstanding.

Blocking use of condvars is not supported (in the main thread, outside of
e.g. L<Coro> threads).

Using event-based modules such as L<IO::AIO>, L<Gtk2>, L<Tk> and so on is
easy.

=back

=head2 Passing file descriptors

Unlike L<AnyEvent::Fork>, this module has no in-built file handle or file
descriptor passing abilities.

The reason is that passing file descriptors is extraordinary tricky
business, and conflicts with efficient batching of messages.

There still is a method you can use: Create a
C<AnyEvent::Util::portable_socketpair> and C<send_fh> one half of it to
the process before you pass control to C<AnyEvent::Fork::RPC::run>.

Whenever you want to pass a file descriptor, send an rpc request to the
child process (so it expects the descriptor), then send it over the other
half of the socketpair. The child should fetch the descriptor from the
half it has passed earlier.

Here is some (untested) pseudocode to that effect:

   use AnyEvent::Util;
   use AnyEvent::Fork;
   use AnyEvent::Fork::RPC;
   use IO::FDPass;

   my ($s1, $s2) = AnyEvent::Util::portable_socketpair;

   my $rpc = AnyEvent::Fork
      ->new
      ->send_fh ($s2)
      ->require ("MyWorker")
      ->AnyEvent::Fork::RPC::run ("MyWorker::run"
           init => "MyWorker::init",
        );

   undef $s2; # no need to keep it around

   # pass an fd
   $rpc->("i'll send some fd now, please expect it!", my $cv = AE::cv);

   IO::FDPass fileno $s1, fileno $handle_to_pass;