… | |
… | |
12 | ->require ("MyModule") |
12 | ->require ("MyModule") |
13 | ->AnyEvent::Fork::RPC::run ( |
13 | ->AnyEvent::Fork::RPC::run ( |
14 | "MyModule::server", |
14 | "MyModule::server", |
15 | ); |
15 | ); |
16 | |
16 | |
|
|
17 | use AnyEvent; |
|
|
18 | |
17 | my $cv = AE::cv; |
19 | my $cv = AE::cv; |
18 | |
20 | |
19 | $rpc->(1, 2, 3, sub { |
21 | $rpc->(1, 2, 3, sub { |
20 | print "MyModule::server returned @_\n"; |
22 | print "MyModule::server returned @_\n"; |
21 | $cv->send; |
23 | $cv->send; |
… | |
… | |
49 | silly, but illustrates the use of events. |
51 | silly, but illustrates the use of events. |
50 | |
52 | |
51 | First the parent process: |
53 | First the parent process: |
52 | |
54 | |
53 | use AnyEvent; |
55 | use AnyEvent; |
54 | use AnyEvent::Fork; |
|
|
55 | use AnyEvent::Fork::RPC; |
56 | use AnyEvent::Fork::RPC; |
56 | |
57 | |
57 | my $done = AE::cv; |
58 | my $done = AE::cv; |
58 | |
59 | |
59 | my $rpc = AnyEvent::Fork |
60 | my $rpc = AnyEvent::Fork |
… | |
… | |
174 | child process may exit at any time, so you should call C<$done> only when |
175 | child process may exit at any time, so you should call C<$done> only when |
175 | you really I<are> done. |
176 | you really I<are> done. |
176 | |
177 | |
177 | =head2 Example 2: Asynchronous Backend |
178 | =head2 Example 2: Asynchronous Backend |
178 | |
179 | |
179 | #TODO |
180 | This example implements multiple count-downs in the child, using |
|
|
181 | L<AnyEvent> timers. While this is a bit silly (one could use timers in te |
|
|
182 | parent just as well), it illustrates the ability to use AnyEvent in the |
|
|
183 | child and the fact that responses can arrive in a different order then the |
|
|
184 | requests. |
|
|
185 | |
|
|
186 | It also shows how to embed the actual child code into a C<__DATA__> |
|
|
187 | section, so it doesn't need any external files at all. |
|
|
188 | |
|
|
189 | And when your parent process is often busy, and you have stricter timing |
|
|
190 | requirements, then running timers in a child process suddenly doesn't look |
|
|
191 | so silly anymore. |
|
|
192 | |
|
|
193 | Without further ado, here is the code: |
|
|
194 | |
|
|
195 | use AnyEvent; |
|
|
196 | use AnyEvent::Fork::RPC; |
|
|
197 | |
|
|
198 | my $done = AE::cv; |
|
|
199 | |
|
|
200 | my $rpc = AnyEvent::Fork |
|
|
201 | ->new |
|
|
202 | ->require ("AnyEvent::Fork::RPC::Async") |
|
|
203 | ->eval (do { local $/; <DATA> }) |
|
|
204 | ->AnyEvent::Fork::RPC::run ("run", |
|
|
205 | async => 1, |
|
|
206 | on_error => sub { warn "FATAL: $_[0]"; exit 1 }, |
|
|
207 | on_event => sub { print $_[0] }, |
|
|
208 | on_destroy => $done, |
|
|
209 | ); |
|
|
210 | |
|
|
211 | for my $count (3, 2, 1) { |
|
|
212 | $rpc->($count, sub { |
|
|
213 | warn "job $count finished\n"; |
|
|
214 | }); |
|
|
215 | } |
|
|
216 | |
|
|
217 | undef $rpc; |
|
|
218 | |
|
|
219 | $done->recv; |
|
|
220 | |
|
|
221 | __DATA__ |
|
|
222 | |
|
|
223 | # this ends up in main, as we don't use a package declaration |
|
|
224 | |
|
|
225 | use AnyEvent; |
|
|
226 | |
|
|
227 | sub run { |
|
|
228 | my ($done, $count) = @_; |
|
|
229 | |
|
|
230 | my $n; |
|
|
231 | |
|
|
232 | AnyEvent::Fork::RPC::event "starting to count up to $count\n"; |
|
|
233 | |
|
|
234 | my $w; $w = AE::timer 1, 1, sub { |
|
|
235 | ++$n; |
|
|
236 | |
|
|
237 | AnyEvent::Fork::RPC::event "count $n of $count\n"; |
|
|
238 | |
|
|
239 | if ($n == $count) { |
|
|
240 | undef $w; |
|
|
241 | $done->(); |
|
|
242 | } |
|
|
243 | }; |
|
|
244 | } |
|
|
245 | |
|
|
246 | The parent part (the one before the C<__DATA__> section) isn't very |
|
|
247 | different from the earlier examples. It sets async mode, preloads |
|
|
248 | the backend module (so the C<AnyEvent::Fork::RPC::event> function is |
|
|
249 | declared), uses a slightly different C<on_event> handler (which we use |
|
|
250 | simply for logging purposes) and then, instead of loading a module with |
|
|
251 | the actual worker code, it C<eval>'s the code from the data section in the |
|
|
252 | child process. |
|
|
253 | |
|
|
254 | It then starts three countdowns, from 3 to 1 seconds downwards, destroys |
|
|
255 | the rpc object so the example finishes eventually, and then just waits for |
|
|
256 | the stuff to trickle in. |
|
|
257 | |
|
|
258 | The worker code uses the event function to log some progress messages, but |
|
|
259 | mostly just creates a recurring one-second timer. |
|
|
260 | |
|
|
261 | The timer callback increments a counter, logs a message, and eventually, |
|
|
262 | when the count has been reached, calls the finish callback. |
|
|
263 | |
|
|
264 | On my system, this results in the following output. Since all timers fire |
|
|
265 | at roughly the same time, the actual order isn't guaranteed, but the order |
|
|
266 | shown is very likely what you would get, too. |
|
|
267 | |
|
|
268 | starting to count up to 3 |
|
|
269 | starting to count up to 2 |
|
|
270 | starting to count up to 1 |
|
|
271 | count 1 of 3 |
|
|
272 | count 1 of 2 |
|
|
273 | count 1 of 1 |
|
|
274 | job 1 finished |
|
|
275 | count 2 of 2 |
|
|
276 | job 2 finished |
|
|
277 | count 2 of 3 |
|
|
278 | count 3 of 3 |
|
|
279 | job 3 finished |
|
|
280 | |
|
|
281 | While the overall ordering isn't guaranteed, the async backend still |
|
|
282 | guarantees that events and responses are delivered to the parent process |
|
|
283 | in the exact same ordering as they were generated in the child process. |
|
|
284 | |
|
|
285 | And unless your system is I<very> busy, it should clearly show that the |
|
|
286 | job started last will finish first, as it has the lowest count. |
|
|
287 | |
|
|
288 | This concludes the async example. Since L<AnyEvent::Fork> does not |
|
|
289 | actually fork, you are free to use about any module in the child, not just |
|
|
290 | L<AnyEvent>, but also L<IO::AIO>, or L<Tk> for example. |
180 | |
291 | |
181 | =head1 PARENT PROCESS USAGE |
292 | =head1 PARENT PROCESS USAGE |
182 | |
293 | |
183 | This module exports nothing, and only implements a single function: |
294 | This module exports nothing, and only implements a single function: |
184 | |
295 | |
… | |
… | |
263 | |
374 | |
264 | The default server used in the child does all I/O blockingly, and only |
375 | The default server used in the child does all I/O blockingly, and only |
265 | allows a single RPC call to execute concurrently. |
376 | allows a single RPC call to execute concurrently. |
266 | |
377 | |
267 | Setting C<async> to a true value switches to another implementation that |
378 | Setting C<async> to a true value switches to another implementation that |
268 | uses L<AnyEvent> in the child and allows multiple concurrent RPC calls. |
379 | uses L<AnyEvent> in the child and allows multiple concurrent RPC calls (it |
|
|
380 | does not support recursion in the event loop however, blocking condvar |
|
|
381 | calls will fail). |
269 | |
382 | |
270 | The actual API in the child is documented in the section that describes |
383 | The actual API in the child is documented in the section that describes |
271 | the calling semantics of the returned C<$rpc> function. |
384 | the calling semantics of the returned C<$rpc> function. |
272 | |
385 | |
273 | If you want to pre-load the actual back-end modules to enable memory |
386 | If you want to pre-load the actual back-end modules to enable memory |
… | |
… | |
275 | synchronous, and C<AnyEvent::Fork::RPC::Async> for asynchronous mode. |
388 | synchronous, and C<AnyEvent::Fork::RPC::Async> for asynchronous mode. |
276 | |
389 | |
277 | If you use a template process and want to fork both sync and async |
390 | If you use a template process and want to fork both sync and async |
278 | children, then it is permissible to load both modules. |
391 | children, then it is permissible to load both modules. |
279 | |
392 | |
280 | =item serialiser => $string (default: '(sub { pack "(w/a*)*", @_ }, sub { unpack "(w/a*)*", shift })') |
393 | =item serialiser => $string (default: $AnyEvent::Fork::RPC::STRING_SERIALISER) |
281 | |
394 | |
282 | All arguments, result data and event data have to be serialised to be |
395 | All arguments, result data and event data have to be serialised to be |
283 | transferred between the processes. For this, they have to be frozen and |
396 | transferred between the processes. For this, they have to be frozen and |
284 | thawed in both parent and child processes. |
397 | thawed in both parent and child processes. |
285 | |
398 | |
286 | By default, only octet strings can be passed between the processes, which |
399 | By default, only octet strings can be passed between the processes, which |
287 | is reasonably fast and efficient. |
400 | is reasonably fast and efficient and requires no extra modules. |
288 | |
401 | |
289 | For more complicated use cases, you can provide your own freeze and thaw |
402 | For more complicated use cases, you can provide your own freeze and thaw |
290 | functions, by specifying a string with perl source code. It's supposed to |
403 | functions, by specifying a string with perl source code. It's supposed to |
291 | return two code references when evaluated: the first receives a list of |
404 | return two code references when evaluated: the first receives a list of |
292 | perl values and must return an octet string. The second receives the octet |
405 | perl values and must return an octet string. The second receives the octet |
… | |
… | |
294 | |
407 | |
295 | If you need an external module for serialisation, then you can either |
408 | If you need an external module for serialisation, then you can either |
296 | pre-load it into your L<AnyEvent::Fork> process, or you can add a C<use> |
409 | pre-load it into your L<AnyEvent::Fork> process, or you can add a C<use> |
297 | or C<require> statement into the serialiser string. Or both. |
410 | or C<require> statement into the serialiser string. Or both. |
298 | |
411 | |
|
|
412 | Here are some examples - some of them are also available as global |
|
|
413 | variables that make them easier to use. |
|
|
414 | |
|
|
415 | =over 4 |
|
|
416 | |
|
|
417 | =item octet strings - C<$AnyEvent::Fork::RPC::STRING_SERIALISER> |
|
|
418 | |
|
|
419 | This serialiser concatenates length-prefixes octet strings, and is the |
|
|
420 | default. |
|
|
421 | |
|
|
422 | Implementation: |
|
|
423 | |
|
|
424 | ( |
|
|
425 | sub { pack "(w/a*)*", @_ }, |
|
|
426 | sub { unpack "(w/a*)*", shift } |
|
|
427 | ) |
|
|
428 | |
|
|
429 | =item json - C<$AnyEvent::Fork::RPC::JSON_SERIALISER> |
|
|
430 | |
|
|
431 | This serialiser creates JSON arrays - you have to make sure the L<JSON> |
|
|
432 | module is installed for this serialiser to work. It can be beneficial for |
|
|
433 | sharing when you preload the L<JSON> module in a template process. |
|
|
434 | |
|
|
435 | L<JSON> (with L<JSON::XS> installed) is slower than the octet string |
|
|
436 | serialiser, but usually much faster than L<Storable>, unless big chunks of |
|
|
437 | binary data need to be transferred. |
|
|
438 | |
|
|
439 | Implementation: |
|
|
440 | |
|
|
441 | use JSON (); |
|
|
442 | ( |
|
|
443 | sub { JSON::encode_json \@_ }, |
|
|
444 | sub { @{ JSON::decode_json shift } } |
|
|
445 | ) |
|
|
446 | |
|
|
447 | =item storable - C<$AnyEvent::Fork::RPC::STORABLE_SERIALISER> |
|
|
448 | |
|
|
449 | This serialiser uses L<Storable>, which means it has high chance of |
|
|
450 | serialising just about anything you throw at it, at the cost of having |
|
|
451 | very high overhead per operation. It also comes with perl. |
|
|
452 | |
|
|
453 | Implementation: |
|
|
454 | |
|
|
455 | use Storable (); |
|
|
456 | ( |
|
|
457 | sub { Storable::freeze \@_ }, |
|
|
458 | sub { @{ Storable::thaw shift } } |
|
|
459 | ) |
|
|
460 | |
|
|
461 | =back |
|
|
462 | |
299 | =back |
463 | =back |
300 | |
464 | |
301 | See the examples section earlier in this document for some actual |
465 | See the examples section earlier in this document for some actual |
302 | examples. |
466 | examples. |
303 | |
467 | |
304 | =cut |
468 | =cut |
305 | |
469 | |
306 | our $STRING_SERIALISER = '(sub { pack "(w/a*)*", @_ }, sub { unpack "(w/a*)*", shift })'; |
470 | our $STRING_SERIALISER = '(sub { pack "(w/a*)*", @_ }, sub { unpack "(w/a*)*", shift })'; |
|
|
471 | our $JSON_SERIALISER = 'use JSON (); (sub { JSON::encode_json \@_ }, sub { @{ JSON::decode_json shift } })'; |
|
|
472 | our $STORABLE_SERIALISER = 'use Storable (); (sub { Storable::freeze \@_ }, sub { @{ Storable::thaw shift } })'; |
307 | |
473 | |
308 | sub run { |
474 | sub run { |
309 | my ($self, $function, %arg) = @_; |
475 | my ($self, $function, %arg) = @_; |
310 | |
476 | |
311 | my $serialiser = delete $arg{serialiser} || $STRING_SERIALISER; |
477 | my $serialiser = delete $arg{serialiser} || $STRING_SERIALISER; |
… | |
… | |
380 | } |
546 | } |
381 | } elsif (defined $len) { |
547 | } elsif (defined $len) { |
382 | undef $rw; undef $ww; # it ends here |
548 | undef $rw; undef $ww; # it ends here |
383 | |
549 | |
384 | if (@rcb || %rcb) { |
550 | if (@rcb || %rcb) { |
385 | use Data::Dump;ddx[\@rcb,\%rcb];#d# |
|
|
386 | $on_error->("unexpected eof"); |
551 | $on_error->("unexpected eof"); |
387 | } else { |
552 | } else { |
388 | $on_destroy->(); |
553 | $on_destroy->(); |
389 | } |
554 | } |
390 | } elsif ($! != Errno::EAGAIN && $! != Errno::EWOULDBLOCK) { |
555 | } elsif ($! != Errno::EAGAIN && $! != Errno::EWOULDBLOCK) { |
… | |
… | |
470 | See the examples section earlier in this document for some actual |
635 | See the examples section earlier in this document for some actual |
471 | examples. |
636 | examples. |
472 | |
637 | |
473 | =back |
638 | =back |
474 | |
639 | |
|
|
640 | =head1 ADVANCED TOPICS |
|
|
641 | |
|
|
642 | =head2 Choosing a backend |
|
|
643 | |
|
|
644 | So how do you decide which backend to use? Well, that's your problem to |
|
|
645 | solve, but here are some thoughts on the matter: |
|
|
646 | |
|
|
647 | =over 4 |
|
|
648 | |
|
|
649 | =item Synchronous |
|
|
650 | |
|
|
651 | The synchronous backend does not rely on any external modules (well, |
|
|
652 | except L<common::sense>, which works around a bug in how perl's warning |
|
|
653 | system works). This keeps the process very small, for example, on my |
|
|
654 | system, an empty perl interpreter uses 1492kB RSS, which becomes 2020kB |
|
|
655 | after C<use warnings; use strict> (for people who grew up with C64s around |
|
|
656 | them this is probably shocking every single time they see it). The worker |
|
|
657 | process in the first example in this document uses 1792kB. |
|
|
658 | |
|
|
659 | Since the calls are done synchronously, slow jobs will keep newer jobs |
|
|
660 | from executing. |
|
|
661 | |
|
|
662 | The synchronous backend also has no overhead due to running an event loop |
|
|
663 | - reading requests is therefore very efficient, while writing responses is |
|
|
664 | less so, as every response results in a write syscall. |
|
|
665 | |
|
|
666 | If the parent process is busy and a bit slow reading responses, the child |
|
|
667 | waits instead of processing further requests. This also limits the amount |
|
|
668 | of memory needed for buffering, as never more than one response has to be |
|
|
669 | buffered. |
|
|
670 | |
|
|
671 | The API in the child is simple - you just have to define a function that |
|
|
672 | does something and returns something. |
|
|
673 | |
|
|
674 | It's hard to use modules or code that relies on an event loop, as the |
|
|
675 | child cannot execute anything while it waits for more input. |
|
|
676 | |
|
|
677 | =item Asynchronous |
|
|
678 | |
|
|
679 | The asynchronous backend relies on L<AnyEvent>, which tries to be small, |
|
|
680 | but still comes at a price: On my system, the worker from example 1a uses |
|
|
681 | 3420kB RSS (for L<AnyEvent>, which loads L<EV>, which needs L<XSLoader> |
|
|
682 | which in turn loads a lot of other modules such as L<warnings>, L<strict>, |
|
|
683 | L<vars>, L<Exporter>...). |
|
|
684 | |
|
|
685 | It batches requests and responses reasonably efficiently, doing only as |
|
|
686 | few reads and writes as needed, but needs to poll for events via the event |
|
|
687 | loop. |
|
|
688 | |
|
|
689 | Responses are queued when the parent process is busy. This means the child |
|
|
690 | can continue to execute any queued requests. It also means that a child |
|
|
691 | might queue a lot of responses in memory when it generates them and the |
|
|
692 | parent process is slow accepting them. |
|
|
693 | |
|
|
694 | The API is not a straightforward RPC pattern - you have to call a |
|
|
695 | "done" callback to pass return values and signal completion. Also, more |
|
|
696 | importantly, the API starts jobs as fast as possible - when 1000 jobs |
|
|
697 | are queued and the jobs are slow, they will all run concurrently. The |
|
|
698 | child must implement some queueing/limiting mechanism if this causes |
|
|
699 | problems. Alternatively, the parent could limit the amount of rpc calls |
|
|
700 | that are outstanding. |
|
|
701 | |
|
|
702 | Blocking use of condvars is not supported. |
|
|
703 | |
|
|
704 | Using event-based modules such as L<IO::AIO>, L<Gtk2>, L<Tk> and so on is |
|
|
705 | easy. |
|
|
706 | |
|
|
707 | =back |
|
|
708 | |
|
|
709 | =head2 Passing file descriptors |
|
|
710 | |
|
|
711 | Unlike L<AnyEvent::Fork>, this module has no in-built file handle or file |
|
|
712 | descriptor passing abilities. |
|
|
713 | |
|
|
714 | The reason is that passing file descriptors is extraordinary tricky |
|
|
715 | business, and conflicts with efficient batching of messages. |
|
|
716 | |
|
|
717 | There still is a method you can use: Create a |
|
|
718 | C<AnyEvent::Util::portable_socketpair> and C<send_fh> one half of it to |
|
|
719 | the process before you pass control to C<AnyEvent::Fork::RPC::run>. |
|
|
720 | |
|
|
721 | Whenever you want to pass a file descriptor, send an rpc request to the |
|
|
722 | child process (so it expects the descriptor), then send it over the other |
|
|
723 | half of the socketpair. The child should fetch the descriptor from the |
|
|
724 | half it has passed earlier. |
|
|
725 | |
|
|
726 | Here is some (untested) pseudocode to that effect: |
|
|
727 | |
|
|
728 | use AnyEvent::Util; |
|
|
729 | use AnyEvent::Fork::RPC; |
|
|
730 | use IO::FDPass; |
|
|
731 | |
|
|
732 | my ($s1, $s2) = AnyEvent::Util::portable_socketpair; |
|
|
733 | |
|
|
734 | my $rpc = AnyEvent::Fork |
|
|
735 | ->new |
|
|
736 | ->send_fh ($s2) |
|
|
737 | ->require ("MyWorker") |
|
|
738 | ->AnyEvent::Fork::RPC::run ("MyWorker::run" |
|
|
739 | init => "MyWorker::init", |
|
|
740 | ); |
|
|
741 | |
|
|
742 | undef $s2; # no need to keep it around |
|
|
743 | |
|
|
744 | # pass an fd |
|
|
745 | $rpc->("i'll send some fd now, please expect it!", my $cv = AE::cv); |
|
|
746 | |
|
|
747 | IO::FDPass fileno $s1, fileno $handle_to_pass; |
|
|
748 | |
|
|
749 | $cv->recv; |
|
|
750 | |
|
|
751 | The MyWorker module could look like this: |
|
|
752 | |
|
|
753 | package MyWorker; |
|
|
754 | |
|
|
755 | use IO::FDPass; |
|
|
756 | |
|
|
757 | my $s2; |
|
|
758 | |
|
|
759 | sub init { |
|
|
760 | $s2 = $_[0]; |
|
|
761 | } |
|
|
762 | |
|
|
763 | sub run { |
|
|
764 | if ($_[0] eq "i'll send some fd now, please expect it!") { |
|
|
765 | my $fd = IO::FDPass::recv fileno $s2; |
|
|
766 | ... |
|
|
767 | } |
|
|
768 | } |
|
|
769 | |
|
|
770 | Of course, this might be blocking if you pass a lot of file descriptors, |
|
|
771 | so you might want to look into L<AnyEvent::FDpasser> which can handle the |
|
|
772 | gory details. |
|
|
773 | |
475 | =head1 SEE ALSO |
774 | =head1 SEE ALSO |
476 | |
775 | |
477 | L<AnyEvent::Fork> (to create the processes in the first place), |
776 | L<AnyEvent::Fork>, to create the processes in the first place. |
|
|
777 | |
478 | L<AnyEvent::Fork::Pool> (to manage whole pools of processes). |
778 | L<AnyEvent::Fork::Pool>, to manage whole pools of processes. |
479 | |
779 | |
480 | =head1 AUTHOR AND CONTACT INFORMATION |
780 | =head1 AUTHOR AND CONTACT INFORMATION |
481 | |
781 | |
482 | Marc Lehmann <schmorp@schmorp.de> |
782 | Marc Lehmann <schmorp@schmorp.de> |
483 | http://software.schmorp.de/pkg/AnyEvent-Fork-RPC |
783 | http://software.schmorp.de/pkg/AnyEvent-Fork-RPC |