| … | |
… | |
| 489 | } |
489 | } |
| 490 | } elsif (defined $len) { |
490 | } elsif (defined $len) { |
| 491 | undef $rw; undef $ww; # it ends here |
491 | undef $rw; undef $ww; # it ends here |
| 492 | |
492 | |
| 493 | if (@rcb || %rcb) { |
493 | if (@rcb || %rcb) { |
| 494 | use Data::Dump;ddx[\@rcb,\%rcb];#d# |
|
|
| 495 | $on_error->("unexpected eof"); |
494 | $on_error->("unexpected eof"); |
| 496 | } else { |
495 | } else { |
| 497 | $on_destroy->(); |
496 | $on_destroy->(); |
| 498 | } |
497 | } |
| 499 | } elsif ($! != Errno::EAGAIN && $! != Errno::EWOULDBLOCK) { |
498 | } elsif ($! != Errno::EAGAIN && $! != Errno::EWOULDBLOCK) { |
| … | |
… | |
| 579 | See the examples section earlier in this document for some actual |
578 | See the examples section earlier in this document for some actual |
| 580 | examples. |
579 | examples. |
| 581 | |
580 | |
| 582 | =back |
581 | =back |
| 583 | |
582 | |
|
|
583 | =head1 ADVANCED TOPICS |
|
|
584 | |
|
|
585 | =head2 Choosing a backend |
|
|
586 | |
|
|
587 | So how do you decide which backend to use? Well, that's your problem to |
|
|
588 | solve, but here are some thoughts on the matter: |
|
|
589 | |
|
|
590 | =over 4 |
|
|
591 | |
|
|
592 | =item Synchronous |
|
|
593 | |
|
|
594 | The synchronous backend does not rely on any external modules (well, |
|
|
595 | except L<common::sense>, which works around a bug in how perl's warning |
|
|
596 | system works). This keeps the process very small, for example, on my |
|
|
597 | system, an empty perl interpreter uses 1492kB RSS, which becomes 2020kB |
|
|
598 | after C<use warnings; use strict> (for people who grew up with C64s around |
|
|
599 | them this is probably shocking every single time they see it). The worker |
|
|
600 | process in the first example in this document uses 1792kB. |
|
|
601 | |
|
|
602 | Since the calls are done synchronously, slow jobs will keep newer jobs |
|
|
603 | from executing. |
|
|
604 | |
|
|
605 | The synchronous backend also has no overhead due to running an event loop |
|
|
606 | - reading requests is therefore very efficient, while writing responses is |
|
|
607 | less so, as every response results in a write syscall. |
|
|
608 | |
|
|
609 | If the parent process is busy and a bit slow reading responses, the child |
|
|
610 | waits instead of processing further requests. This also limits the amount |
|
|
611 | of memory needed for buffering, as never more than one response has to be |
|
|
612 | buffered. |
|
|
613 | |
|
|
614 | The API in the child is simple - you just have to define a function that |
|
|
615 | does something and returns something. |
|
|
616 | |
|
|
617 | It's hard to use modules or code that relies on an event loop, as the |
|
|
618 | child cannot execute anything while it waits for more input. |
|
|
619 | |
|
|
620 | =item Asynchronous |
|
|
621 | |
|
|
622 | The asynchronous backend relies on L<AnyEvent>, which tries to be small, |
|
|
623 | but still comes at a price: On my system, the worker from example 1a uses |
|
|
624 | 3420kB RSS (for L<AnyEvent>, which loads L<EV>, which needs L<XSLoader> |
|
|
625 | which in turn loads a lot of other modules such as L<warnings>, L<strict>, |
|
|
626 | L<vars>, L<Exporter>...). |
|
|
627 | |
|
|
628 | It batches requests and responses reasonably efficiently, doing only as |
|
|
629 | few reads and writes as needed, but needs to poll for events via the event |
|
|
630 | loop. |
|
|
631 | |
|
|
632 | Responses are queued when the parent process is busy. This means the child |
|
|
633 | can continue to execute any queued requests. It also means that a child |
|
|
634 | might queue a lot of responses in memory when it generates them and the |
|
|
635 | parent process is slow accepting them. |
|
|
636 | |
|
|
637 | The API is not a straightforward RPC pattern - you have to call a |
|
|
638 | "done" callback to pass return values and signal completion. Also, more |
|
|
639 | importantly, the API starts jobs as fast as possible - when 1000 jobs |
|
|
640 | are queued and the jobs are slow, they will all run concurrently. The |
|
|
641 | child must implement some queueing/limiting mechanism if this causes |
|
|
642 | problems. Alternatively, the parent could limit the amount of rpc calls |
|
|
643 | that are outstanding. |
|
|
644 | |
|
|
645 | Using event-based modules such as L<IO::AIO>, L<Gtk2>, L<Tk> and so on is |
|
|
646 | easy. |
|
|
647 | |
|
|
648 | =back |
|
|
649 | |
|
|
650 | =head2 Passing file descriptors |
|
|
651 | |
|
|
652 | Unlike L<AnyEvent::Fork>, this module has no in-built file handle or file |
|
|
653 | descriptor passing abilities. |
|
|
654 | |
|
|
655 | The reason is that passing file descriptors is extraordinary tricky |
|
|
656 | business, and conflicts with efficient batching of messages. |
|
|
657 | |
|
|
658 | There still is a method you can use: Create a |
|
|
659 | C<AnyEvent::Util::portable_socketpair> and C<send_fh> one half of it to |
|
|
660 | the process before you pass control to C<AnyEvent::Fork::RPC::run>. |
|
|
661 | |
|
|
662 | Whenever you want to pass a file descriptor, send an rpc request to the |
|
|
663 | child process (so it expects the descriptor), then send it over the other |
|
|
664 | half of the socketpair. The child should fetch the descriptor from the |
|
|
665 | half it has passed earlier. |
|
|
666 | |
|
|
667 | Here is some (untested) pseudocode to that effect: |
|
|
668 | |
|
|
669 | use AnyEvent::Util; |
|
|
670 | use AnyEvent::Fork::RPC; |
|
|
671 | use IO::FDPass; |
|
|
672 | |
|
|
673 | my ($s1, $s2) = AnyEvent::Util::portable_socketpair; |
|
|
674 | |
|
|
675 | my $rpc = AnyEvent::Fork |
|
|
676 | ->new |
|
|
677 | ->send_fh ($s2) |
|
|
678 | ->require ("MyWorker") |
|
|
679 | ->AnyEvent::Fork::RPC::run ("MyWorker::run" |
|
|
680 | init => "MyWorker::init", |
|
|
681 | ); |
|
|
682 | |
|
|
683 | undef $s2; # no need to keep it around |
|
|
684 | |
|
|
685 | # pass an fd |
|
|
686 | $rpc->("i'll send some fd now, please expect it!", my $cv = AE::cv); |
|
|
687 | |
|
|
688 | IO::FDPass fileno $s1, fileno $handle_to_pass; |
|
|
689 | |
|
|
690 | $cv->recv; |
|
|
691 | |
|
|
692 | The MyWorker module could look like this: |
|
|
693 | |
|
|
694 | package MyWorker; |
|
|
695 | |
|
|
696 | use IO::FDPass; |
|
|
697 | |
|
|
698 | my $s2; |
|
|
699 | |
|
|
700 | sub init { |
|
|
701 | $s2 = $_[0]; |
|
|
702 | } |
|
|
703 | |
|
|
704 | sub run { |
|
|
705 | if ($_[0] eq "i'll send some fd now, please expect it!") { |
|
|
706 | my $fd = IO::FDPass::recv fileno $s2; |
|
|
707 | ... |
|
|
708 | } |
|
|
709 | } |
|
|
710 | |
|
|
711 | Of course, this might be blocking if you pass a lot of file descriptors, |
|
|
712 | so you might want to look into L<AnyEvent::FDpasser> which can handle the |
|
|
713 | gory details. |
|
|
714 | |
| 584 | =head1 SEE ALSO |
715 | =head1 SEE ALSO |
| 585 | |
716 | |
| 586 | L<AnyEvent::Fork> (to create the processes in the first place), |
717 | L<AnyEvent::Fork> (to create the processes in the first place), |
| 587 | L<AnyEvent::Fork::Pool> (to manage whole pools of processes). |
718 | L<AnyEvent::Fork::Pool> (to manage whole pools of processes). |
| 588 | |
719 | |
