[ViewVC] Diff of: cvs/AnyEvent-Fork-RPC/RPC.pm

Comparing AnyEvent-Fork-RPC/RPC.pm (file contents):
Revision 1.10 by root, Wed Apr 17 22:04:49 2013 UTC vs.
Revision 1.16 by root, Thu Apr 18 14:07:15 2013 UTC

…		…
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

Diff Legend

-–
+Removed lines
-+
+Added lines
-<
+Changed lines
->
+Changed lines

Comparing AnyEvent-Fork-RPC/RPC.pm (file contents): Revision 1.10 by root, Wed Apr 17 22:04:49 2013 UTC vs. Revision 1.16 by root, Thu Apr 18 14:07:15 2013 UTC

Diff Legend

Comparing AnyEvent-Fork-RPC/RPC.pm (file contents):
Revision 1.10 by root, Wed Apr 17 22:04:49 2013 UTC vs.
Revision 1.16 by root, Thu Apr 18 14:07:15 2013 UTC