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

Comparing AnyEvent-Fork/Fork.pm (file contents):
Revision 1.2 by root, Sun Mar 31 03:26:50 2013 UTC vs.
Revision 1.63 by root, Wed Nov 26 13:36:18 2014 UTC

1	=head1 NAME	1	=head1 NAME
2		2
3	AnyEvent::ProcessPool - manage pools of perl worker processes, exec'ed or fork'ed	3	AnyEvent::Fork - everything you wanted to use fork() for, but couldn't
4		4
5	=head1 SYNOPSIS	5	=head1 SYNOPSIS
6		6
7	use AnyEvent::ProcessPool;	7	use AnyEvent::Fork;
		8
		9	AnyEvent::Fork
		10	->new
		11	->require ("MyModule")
		12	->run ("MyModule::server", my $cv = AE::cv);
		13
		14	my $fh = $cv->recv;
8		15
9	=head1 DESCRIPTION	16	=head1 DESCRIPTION
10		17
11	This module allows you to create single worker processes but also worker	18	This module allows you to create new processes, without actually forking
12	pool that share memory, by forking from the main program, or exec'ing new	19	them from your current process (avoiding the problems of forking), but
13	perl interpreters from a module.	20	preserving most of the advantages of fork.
14		21
15	You create a new processes in a pool by specifying a function to call	22	It can be used to create new worker processes or new independent
16	with any combination of string values and file handles.	23	subprocesses for short- and long-running jobs, process pools (e.g. for use
		24	in pre-forked servers) but also to spawn new external processes (such as
		25	CGI scripts from a web server), which can be faster (and more well behaved)
		26	than using fork+exec in big processes.
17		27
18	A pool can have initialisation code which is executed before forking. The	28	Special care has been taken to make this module useful from other modules,
19	initialisation code is only executed once and the resulting process is	29	while still supporting specialised environments such as L<App::Staticperl>
20	cached, to be used as a template.	30	or L<PAR::Packer>.
21		31
22	Pools without such initialisation code don't cache an extra process.	32	=head2 WHAT THIS MODULE IS NOT
23		33
		34	This module only creates processes and lets you pass file handles and
		35	strings to it, and run perl code. It does not implement any kind of RPC -
		36	there is no back channel from the process back to you, and there is no RPC
		37	or message passing going on.
		38
		39	If you need some form of RPC, you could use the L<AnyEvent::Fork::RPC>
		40	companion module, which adds simple RPC/job queueing to a process created
		41	by this module.
		42
		43	And if you need some automatic process pool management on top of
		44	L<AnyEvent::Fork::RPC>, you can look at the L<AnyEvent::Fork::Pool>
		45	companion module.
		46
		47	Or you can implement it yourself in whatever way you like: use some
		48	message-passing module such as L<AnyEvent::MP>, some pipe such as
		49	L<AnyEvent::ZeroMQ>, use L<AnyEvent::Handle> on both sides to send
		50	e.g. JSON or Storable messages, and so on.
		51
		52	=head2 COMPARISON TO OTHER MODULES
		53
		54	There is an abundance of modules on CPAN that do "something fork", such as
		55	L<Parallel::ForkManager>, L<AnyEvent::ForkManager>, L<AnyEvent::Worker>
		56	or L<AnyEvent::Subprocess>. There are modules that implement their own
		57	process management, such as L<AnyEvent::DBI>.
		58
		59	The problems that all these modules try to solve are real, however, none
		60	of them (from what I have seen) tackle the very real problems of unwanted
		61	memory sharing, efficiency or not being able to use event processing, GUI
		62	toolkits or similar modules in the processes they create.
		63
		64	This module doesn't try to replace any of them - instead it tries to solve
		65	the problem of creating processes with a minimum of fuss and overhead (and
		66	also luxury). Ideally, most of these would use AnyEvent::Fork internally,
		67	except they were written before AnyEvent:Fork was available, so obviously
		68	had to roll their own.
		69
24	=head1 PROBLEM STATEMENT	70	=head2 PROBLEM STATEMENT
25		71
26	There are two ways to implement parallel processing on UNIX like operating	72	There are two traditional ways to implement parallel processing on UNIX
27	systems - fork and process, and fork+exec and process. They have different	73	like operating systems - fork and process, and fork+exec and process. They
28	advantages and disadvantages that I describe below, together with how this	74	have different advantages and disadvantages that I describe below,
29	module tries to mitigate the disadvantages.	75	together with how this module tries to mitigate the disadvantages.
30		76
31	=over 4	77	=over 4
32		78
33	=item Forking from a big process can be very slow (a 5GB process needs	79	=item Forking from a big process can be very slow.
34	0.05s to fork on my 3.6GHz amd64 GNU/Linux box for example). This overhead	80
		81	A 5GB process needs 0.05s to fork on my 3.6GHz amd64 GNU/Linux box. This
35	is often shared with exec (because you have to fork first), but in some	82	overhead is often shared with exec (because you have to fork first), but
36	circumstances (e.g. when vfork is used), fork+exec can be much faster.	83	in some circumstances (e.g. when vfork is used), fork+exec can be much
		84	faster.
37		85
38	This module can help here by telling a small(er) helper process to fork,	86	This module can help here by telling a small(er) helper process to fork,
39	or fork+exec instead.	87	which is faster then forking the main process, and also uses vfork where
		88	possible. This gives the speed of vfork, with the flexibility of fork.
40		89
41	=item Forking usually creates a copy-on-write copy of the parent	90	=item Forking usually creates a copy-on-write copy of the parent
42	process. Memory (for example, modules or data files that have been	91	process.
43	will not take additional memory). When exec'ing a new process, modules	92
		93	For example, modules or data files that are loaded will not use additional
		94	memory after a fork. Exec'ing a new process, in contrast, means modules
44	and data files might need to be loaded again, at extra cpu and memory	95	and data files might need to be loaded again, at extra CPU and memory
45	cost. Likewise when forking, all data structures are copied as well - if	96	cost.
		97
		98	But when forking, you still create a copy of your data structures - if
46	the program frees them and replaces them by new data, the child processes	99	the program frees them and replaces them by new data, the child processes
47	will retain the memory even if it isn't used.	100	will retain the old version even if it isn't used, which can suddenly and
		101	unexpectedly increase memory usage when freeing memory.
		102
		103	For example, L<Gtk2::CV> is an image viewer optimised for large
		104	directories (millions of pictures). It also forks subprocesses for
		105	thumbnail generation, which inherit the data structure that stores all
		106	file information. If the user changes the directory, it gets freed in
		107	the main process, leaving a copy in the thumbnailer processes. This can
		108	lead to many times the memory usage that would actually be required. The
		109	solution is to fork early (and being unable to dynamically generate more
		110	subprocesses or do this from a module)... or to use L<AnyEvent:Fork>.
		111
		112	There is a trade-off between more sharing with fork (which can be good or
		113	bad), and no sharing with exec.
48		114
49	This module allows the main program to do a controlled fork, and allows	115	This module allows the main program to do a controlled fork, and allows
50	modules to exec processes safely at any time. When creating a custom	116	modules to exec processes safely at any time. When creating a custom
51	process pool you can take advantage of data sharing via fork without	117	process pool you can take advantage of data sharing via fork without
52	risking to share large dynamic data structures that will blow up child	118	risking to share large dynamic data structures that will blow up child
53	memory usage.	119	memory usage.
54		120
		121	In other words, this module puts you into control over what is being
		122	shared and what isn't, at all times.
		123
55	=item Exec'ing a new perl process might be difficult and slow. For	124	=item Exec'ing a new perl process might be difficult.
		125
56	example, it is not easy to find the correct path to the perl interpreter,	126	For example, it is not easy to find the correct path to the perl
57	and all modules have to be loaded from disk again. Long running processes	127	interpreter - C<$^X> might not be a perl interpreter at all. Worse, there
58	might run into problems when perl is upgraded for example.	128	might not even be a perl binary installed on the system.
59		129
60	This module supports creating pre-initialised perl processes to be used
61	as template, and also tries hard to identify the correct path to the perl	130	This module tries hard to identify the correct path to the perl
62	interpreter. With a cooperative main program, exec'ing the interpreter	131	interpreter. With a cooperative main program, exec'ing the interpreter
63	might not even be necessary.	132	might not even be necessary, but even without help from the main program,
		133	it will still work when used from a module.
64		134
		135	=item Exec'ing a new perl process might be slow, as all necessary modules
		136	have to be loaded from disk again, with no guarantees of success.
		137
		138	Long running processes might run into problems when perl is upgraded
		139	and modules are no longer loadable because they refer to a different
		140	perl version, or parts of a distribution are newer than the ones already
		141	loaded.
		142
		143	This module supports creating pre-initialised perl processes to be used as
		144	a template for new processes at a later time, e.g. for use in a process
		145	pool.
		146
65	=item Forking might be impossible when a program is running. For example,	147	=item Forking might be impossible when a program is running.
66	POSIX makes it almost impossible to fork from a multithreaded program and
67	do anything useful in the child - strictly speaking, if your perl program
68	uses posix threads (even indirectly via e.g. L<IO::AIO> or L<threads>),
69	you cannot call fork on the perl level anymore, at all.
70		148
		149	For example, POSIX makes it almost impossible to fork from a
		150	multi-threaded program while doing anything useful in the child - in
		151	fact, if your perl program uses POSIX threads (even indirectly via
		152	e.g. L<IO::AIO> or L<threads>), you cannot call fork on the perl level
		153	anymore without risking memory corruption or worse on a number of
		154	operating systems.
		155
71	This module can safely fork helper processes at any time, by caling	156	This module can safely fork helper processes at any time, by calling
72	fork+exec in C, in a POSIX-compatible way.	157	fork+exec in C, in a POSIX-compatible way (via L<Proc::FastSpawn>).
73		158
74	=item Parallel processing with fork might be inconvenient or difficult	159	=item Parallel processing with fork might be inconvenient or difficult
		160	to implement. Modules might not work in both parent and child.
		161
75	to implement. For example, when a program uses an event loop and creates	162	For example, when a program uses an event loop and creates watchers it
76	watchers it becomes very hard to use the event loop from a child	163	becomes very hard to use the event loop from a child program, as the
77	program, as the watchers already exist but are only meaningful in the	164	watchers already exist but are only meaningful in the parent. Worse, a
78	parent. Worse, a module might want to use such a system, not knowing	165	module might want to use such a module, not knowing whether another module
79	whether another module or the main program also does, leading to problems.	166	or the main program also does, leading to problems.
80		167
81	This module only lets the main program create pools by forking (because	168	Apart from event loops, graphical toolkits also commonly fall into the
82	only the main program can know when it is still safe to do so) - all other	169	"unsafe module" category, or just about anything that communicates with
83	pools are created by fork+exec, after which such modules can again be	170	the external world, such as network libraries and file I/O modules, which
84	loaded.	171	usually don't like being copied and then allowed to continue in two
		172	processes.
		173
		174	With this module only the main program is allowed to create new processes
		175	by forking (because only the main program can know when it is still safe
		176	to do so) - all other processes are created via fork+exec, which makes it
		177	possible to use modules such as event loops or window interfaces safely.
85		178
86	=back	179	=back
87		180
		181	=head1 EXAMPLES
		182
		183	=head2 Create a single new process, tell it to run your worker function.
		184
		185	AnyEvent::Fork
		186	->new
		187	->require ("MyModule")
		188	->run ("MyModule::worker, sub {
		189	my ($master_filehandle) = @_;
		190
		191	# now $master_filehandle is connected to the
		192	# $slave_filehandle in the new process.
		193	});
		194
		195	C<MyModule> might look like this:
		196
		197	package MyModule;
		198
		199	sub worker {
		200	my ($slave_filehandle) = @_;
		201
		202	# now $slave_filehandle is connected to the $master_filehandle
		203	# in the original prorcess. have fun!
		204	}
		205
		206	=head2 Create a pool of server processes all accepting on the same socket.
		207
		208	# create listener socket
		209	my $listener = ...;
		210
		211	# create a pool template, initialise it and give it the socket
		212	my $pool = AnyEvent::Fork
		213	->new
		214	->require ("Some::Stuff", "My::Server")
		215	->send_fh ($listener);
		216
		217	# now create 10 identical workers
		218	for my $id (1..10) {
		219	$pool
		220	->fork
		221	->send_arg ($id)
		222	->run ("My::Server::run");
		223	}
		224
		225	# now do other things - maybe use the filehandle provided by run
		226	# to wait for the processes to die. or whatever.
		227
		228	C<My::Server> might look like this:
		229
		230	package My::Server;
		231
		232	sub run {
		233	my ($slave, $listener, $id) = @_;
		234
		235	close $slave; # we do not use the socket, so close it to save resources
		236
		237	# we could go ballistic and use e.g. AnyEvent here, or IO::AIO,
		238	# or anything we usually couldn't do in a process forked normally.
		239	while (my $socket = $listener->accept) {
		240	# do sth. with new socket
		241	}
		242	}
		243
		244	=head2 use AnyEvent::Fork as a faster fork+exec
		245
		246	This runs C</bin/echo hi>, with standard output redirected to F</tmp/log>
		247	and standard error redirected to the communications socket. It is usually
		248	faster than fork+exec, but still lets you prepare the environment.
		249
		250	open my $output, ">/tmp/log" or die "$!";
		251
		252	AnyEvent::Fork
		253	->new
		254	->eval ('
		255	# compile a helper function for later use
		256	sub run {
		257	my ($fh, $output, @cmd) = @_;
		258
		259	# perl will clear close-on-exec on STDOUT/STDERR
		260	open STDOUT, ">&", $output or die;
		261	open STDERR, ">&", $fh or die;
		262
		263	exec @cmd;
		264	}
		265	')
		266	->send_fh ($output)
		267	->send_arg ("/bin/echo", "hi")
		268	->run ("run", my $cv = AE::cv);
		269
		270	my $stderr = $cv->recv;
		271
		272	=head2 For stingy users: put the worker code into a C<DATA> section.
		273
		274	When you want to be stingy with files, you can put your code into the
		275	C<DATA> section of your module (or program):
		276
		277	use AnyEvent::Fork;
		278
		279	AnyEvent::Fork
		280	->new
		281	->eval (do { local $/; <DATA> })
		282	->run ("doit", sub { ... });
		283
		284	__DATA__
		285
		286	sub doit {
		287	... do something!
		288	}
		289
		290	=head2 For stingy standalone programs: do not rely on external files at
		291	all.
		292
		293	For single-file scripts it can be inconvenient to rely on external
		294	files - even when using a C<DATA> section, you still need to C<exec> an
		295	external perl interpreter, which might not be available when using
		296	L<App::Staticperl>, L<Urlader> or L<PAR::Packer> for example.
		297
		298	Two modules help here - L<AnyEvent::Fork::Early> forks a template process
		299	for all further calls to C<new_exec>, and L<AnyEvent::Fork::Template>
		300	forks the main program as a template process.
		301
		302	Here is how your main program should look like:
		303
		304	#! perl
		305
		306	# optional, as the very first thing.
		307	# in case modules want to create their own processes.
		308	use AnyEvent::Fork::Early;
		309
		310	# next, load all modules you need in your template process
		311	use Example::My::Module
		312	use Example::Whatever;
		313
		314	# next, put your run function definition and anything else you
		315	# need, but do not use code outside of BEGIN blocks.
		316	sub worker_run {
		317	my ($fh, @args) = @_;
		318	...
		319	}
		320
		321	# now preserve everything so far as AnyEvent::Fork object
		322	# in $TEMPLATE.
		323	use AnyEvent::Fork::Template;
		324
		325	# do not put code outside of BEGIN blocks until here
		326
		327	# now use the $TEMPLATE process in any way you like
		328
		329	# for example: create 10 worker processes
		330	my @worker;
		331	my $cv = AE::cv;
		332	for (1..10) {
		333	$cv->begin;
		334	$TEMPLATE->fork->send_arg ($_)->run ("worker_run", sub {
		335	push @worker, shift;
		336	$cv->end;
		337	});
		338	}
		339	$cv->recv;
		340
		341	=head1 CONCEPTS
		342
		343	This module can create new processes either by executing a new perl
		344	process, or by forking from an existing "template" process.
		345
		346	All these processes are called "child processes" (whether they are direct
		347	children or not), while the process that manages them is called the
		348	"parent process".
		349
		350	Each such process comes with its own file handle that can be used to
		351	communicate with it (it's actually a socket - one end in the new process,
		352	one end in the main process), and among the things you can do in it are
		353	load modules, fork new processes, send file handles to it, and execute
		354	functions.
		355
		356	There are multiple ways to create additional processes to execute some
		357	jobs:
		358
88	=over 4	359	=over 4
89		360
90	=cut	361	=item fork a new process from the "default" template process, load code,
		362	run it
91		363
		364	This module has a "default" template process which it executes when it is
		365	needed the first time. Forking from this process shares the memory used
		366	for the perl interpreter with the new process, but loading modules takes
		367	time, and the memory is not shared with anything else.
		368
		369	This is ideal for when you only need one extra process of a kind, with the
		370	option of starting and stopping it on demand.
		371
		372	Example:
		373
		374	AnyEvent::Fork
		375	->new
		376	->require ("Some::Module")
		377	->run ("Some::Module::run", sub {
		378	my ($fork_fh) = @_;
		379	});
		380
		381	=item fork a new template process, load code, then fork processes off of
		382	it and run the code
		383
		384	When you need to have a bunch of processes that all execute the same (or
		385	very similar) tasks, then a good way is to create a new template process
		386	for them, loading all the modules you need, and then create your worker
		387	processes from this new template process.
		388
		389	This way, all code (and data structures) that can be shared (e.g. the
		390	modules you loaded) is shared between the processes, and each new process
		391	consumes relatively little memory of its own.
		392
		393	The disadvantage of this approach is that you need to create a template
		394	process for the sole purpose of forking new processes from it, but if you
		395	only need a fixed number of processes you can create them, and then destroy
		396	the template process.
		397
		398	Example:
		399
		400	my $template = AnyEvent::Fork->new->require ("Some::Module");
		401
		402	for (1..10) {
		403	$template->fork->run ("Some::Module::run", sub {
		404	my ($fork_fh) = @_;
		405	});
		406	}
		407
		408	# at this point, you can keep $template around to fork new processes
		409	# later, or you can destroy it, which causes it to vanish.
		410
		411	=item execute a new perl interpreter, load some code, run it
		412
		413	This is relatively slow, and doesn't allow you to share memory between
		414	multiple processes.
		415
		416	The only advantage is that you don't have to have a template process
		417	hanging around all the time to fork off some new processes, which might be
		418	an advantage when there are long time spans where no extra processes are
		419	needed.
		420
		421	Example:
		422
		423	AnyEvent::Fork
		424	->new_exec
		425	->require ("Some::Module")
		426	->run ("Some::Module::run", sub {
		427	my ($fork_fh) = @_;
		428	});
		429
		430	=back
		431
		432	=head1 THE C<AnyEvent::Fork> CLASS
		433
		434	This module exports nothing, and only implements a single class -
		435	C<AnyEvent::Fork>.
		436
		437	There are two class constructors that both create new processes - C<new>
		438	and C<new_exec>. The C<fork> method creates a new process by forking an
		439	existing one and could be considered a third constructor.
		440
		441	Most of the remaining methods deal with preparing the new process, by
		442	loading code, evaluating code and sending data to the new process. They
		443	usually return the process object, so you can chain method calls.
		444
		445	If a process object is destroyed before calling its C<run> method, then
		446	the process simply exits. After C<run> is called, all responsibility is
		447	passed to the specified function.
		448
		449	As long as there is any outstanding work to be done, process objects
		450	resist being destroyed, so there is no reason to store them unless you
		451	need them later - configure and forget works just fine.
		452
		453	=over 4
		454
		455	=cut
		456
92	package AnyEvent::ProcessPool;	457	package AnyEvent::Fork;
93		458
94	use common::sense;	459	use common::sense;
95		460
96	use Socket ();	461	use Errno ();
97		462
98	use Proc::FastSpawn;
99	use AnyEvent;	463	use AnyEvent;
100	use AnyEvent::ProcessPool::Util;
101	use AnyEvent::Util ();	464	use AnyEvent::Util ();
102		465
103	BEGIN {	466	use IO::FDPass;
104	# require Exporter;
105	}
106		467
107	=item my $pool = new AnyEvent::ProcessPool key => value...	468	our $VERSION = 1.2;
108		469
109	Create a new process pool. The following named parameters are supported:	470	# the early fork template process
		471	our $EARLY;
110		472
111	=over 4
112
113	=back
114
115	=cut
116
117	# the template process	473	# the empty template process
118	our $template;	474	our $TEMPLATE;
119		475
120	sub _queue {	476	sub QUEUE() { 0 }
121	my ($pid, $fh) = @_;	477	sub FH() { 1 }
		478	sub WW() { 2 }
		479	sub PID() { 3 }
		480	sub CB() { 4 }
122		481
123	[	482	sub _new {
		483	my ($self, $fh, $pid) = @_;
		484
		485	AnyEvent::Util::fh_nonblocking $fh, 1;
		486
		487	$self = bless [
		488	[], # write queue - strings or fd's
		489	$fh,
		490	undef, # AE watcher
124	$pid,	491	$pid,
125	$fh,	492	], $self;
126	[],
127	undef
128	]
129	}
130		493
		494	$self
		495	}
		496
131	sub queue_cmd {	497	sub _cmd {
132	my ($queue, $cmd) = @_;	498	my $self = shift;
133		499
134	push @{ $queue->[2] }, pack "N/a", $cmd;	500	# ideally, we would want to use "a (w/a)*" as format string, but perl
		501	# versions from at least 5.8.9 to 5.16.3 are all buggy and can't unpack
		502	# it.
		503	push @{ $self->[QUEUE] }, pack "a L/a*", $_[0], $_[1];
135		504
136	$queue->[3] \|\|= AE::io $queue->[1], 1, sub {	505	$self->[WW] \|\|= AE::io $self->[FH], 1, sub {
		506	do {
		507	# send the next "thing" in the queue - either a reference to an fh,
		508	# or a plain string.
		509
137	if (ref $queue->[2][0]) {	510	if (ref $self->[QUEUE][0]) {
138	AnyEvent::ProcessPool::Util::fd_send fileno $queue->[1], fileno ${ $queue->[2][0] }	511	# send fh
		512	unless (IO::FDPass::send fileno $self->[FH], fileno ${ $self->[QUEUE][0] }) {
		513	return if $! == Errno::EAGAIN \|\| $! == Errno::EWOULDBLOCK;
		514	undef $self->[WW];
		515	die "AnyEvent::Fork: file descriptor send failure: $!";
		516	}
		517
139	and shift @{ $queue->[2] };	518	shift @{ $self->[QUEUE] };
		519
140	} else {	520	} else {
		521	# send string
141	my $len = syswrite $queue->[1], $queue->[2][0]	522	my $len = syswrite $self->[FH], $self->[QUEUE][0];
142	or do { undef $queue->[3]; die "AnyEvent::ProcessPool::queue write failure: $!" };	523
		524	unless ($len) {
		525	return if $! == Errno::EAGAIN \|\| $! == Errno::EWOULDBLOCK;
		526	undef $self->[WW];
		527	die "AnyEvent::Fork: command write failure: $!";
		528	}
		529
143	substr $queue->[2][0], 0, $len, "";	530	substr $self->[QUEUE][0], 0, $len, "";
144	shift @{ $queue->[2] } unless length $queue->[2][0];	531	shift @{ $self->[QUEUE] } unless length $self->[QUEUE][0];
		532	}
		533	} while @{ $self->[QUEUE] };
		534
		535	# everything written
		536	undef $self->[WW];
		537
		538	# invoke run callback, if any
		539	if ($self->[CB]) {
		540	$self->[CB]->($self->[FH]);
		541	@$self = ();
145	}	542	}
146
147	undef $queue->[3] unless @{ $queue->[2] };
148	};	543	};
149	}
150		544
151	sub run_template {	545	() # make sure we don't leak the watcher
152	return if $template;	546	}
153		547
		548	# fork template from current process, used by AnyEvent::Fork::Early/Template
		549	sub _new_fork {
154	my ($fh, $slave) = AnyEvent::Util::portable_socketpair;	550	my ($fh, $slave) = AnyEvent::Util::portable_socketpair;
155	AnyEvent::Util::fh_nonblocking $fh, 1;	551	my $parent = $$;
156	fd_inherit fileno $slave;
157		552
158	my %env = %ENV;	553	my $pid = fork;
159	$env{PERL5LIB} = join ":", grep !ref, @INC;
160		554
161	my $pid = spawn	555	if ($pid eq 0) {
162	$^X,	556	require AnyEvent::Fork::Serve;
163	["perl", "-MAnyEvent::ProcessPool::Serve", "-e", "AnyEvent::ProcessPool::Serve::me", fileno $slave],	557	$AnyEvent::Fork::Serve::OWNER = $parent;
164	[map "$_=$env{$_}", keys %env],	558	close $fh;
165	or die "unable to spawn AnyEvent::ProcessPool server: $!";	559	$0 = "$_[1] of $parent";
		560	AnyEvent::Fork::Serve::serve ($slave);
		561	exit 0;
		562	} elsif (!$pid) {
		563	die "AnyEvent::Fork::Early/Template: unable to fork template process: $!";
		564	}
166		565
167	close $slave;	566	AnyEvent::Fork->_new ($fh, $pid)
168
169	$template = _queue $pid, $fh;
170
171	my ($a, $b) = AnyEvent::Util::portable_socketpair;
172
173	queue_cmd $template, "Iabc";
174	push @{ $template->[2] }, \$b;
175
176	use Coro::AnyEvent; Coro::AnyEvent::sleep 1;
177	undef $b;
178	die "x" . <$a>;
179	}	567	}
		568
		569	=item my $proc = new AnyEvent::Fork
		570
		571	Create a new "empty" perl interpreter process and returns its process
		572	object for further manipulation.
		573
		574	The new process is forked from a template process that is kept around
		575	for this purpose. When it doesn't exist yet, it is created by a call to
		576	C<new_exec> first and then stays around for future calls.
		577
		578	=cut
180		579
181	sub new {	580	sub new {
182	my $class = shift;	581	my $class = shift;
183		582
184	my $self = bless {	583	$TEMPLATE \|\|= $class->new_exec;
		584	$TEMPLATE->fork
		585	}
		586
		587	=item $new_proc = $proc->fork
		588
		589	Forks C<$proc>, creating a new process, and returns the process object
		590	of the new process.
		591
		592	If any of the C<send_> functions have been called before fork, then they
		593	will be cloned in the child. For example, in a pre-forked server, you
		594	might C<send_fh> the listening socket into the template process, and then
		595	keep calling C<fork> and C<run>.
		596
		597	=cut
		598
		599	sub fork {
		600	my ($self) = @_;
		601
		602	my ($fh, $slave) = AnyEvent::Util::portable_socketpair;
		603
		604	$self->send_fh ($slave);
		605	$self->_cmd ("f");
		606
		607	AnyEvent::Fork->_new ($fh)
		608	}
		609
		610	=item my $proc = new_exec AnyEvent::Fork
		611
		612	Create a new "empty" perl interpreter process and returns its process
		613	object for further manipulation.
		614
		615	Unlike the C<new> method, this method I<always> spawns a new perl process
		616	(except in some cases, see L<AnyEvent::Fork::Early> for details). This
		617	reduces the amount of memory sharing that is possible, and is also slower.
		618
		619	You should use C<new> whenever possible, except when having a template
		620	process around is unacceptable.
		621
		622	The path to the perl interpreter is divined using various methods - first
		623	C<$^X> is investigated to see if the path ends with something that looks
		624	as if it were the perl interpreter. Failing this, the module falls back to
		625	using C<$Config::Config{perlpath}>.
		626
		627	The path to perl can also be overriden by setting the global variable
		628	C<$AnyEvent::Fork::PERL> - it's value will be used for all subsequent
		629	invocations.
		630
		631	=cut
		632
		633	our $PERL;
		634
		635	sub new_exec {
		636	my ($self) = @_;
		637
		638	return $EARLY->fork
		639	if $EARLY;
		640
		641	unless (defined $PERL) {
		642	# first find path of perl
		643	my $perl = $^X;
		644
		645	# first we try $^X, but the path must be absolute (always on win32), and end in sth.
		646	# that looks like perl. this obviously only works for posix and win32
		647	unless (
		648	($^O eq "MSWin32" \|\| $perl =~ m%^/%)
		649	&& $perl =~ m%[/\\]perl(?:[0-9]+(\.[0-9]+)+)?(\.exe)?$%i
		650	) {
		651	# if it doesn't look perlish enough, try Config
		652	require Config;
		653	$perl = $Config::Config{perlpath};
		654	$perl =~ s/(?:\Q$Config::Config{_exe}\E)?$/$Config::Config{_exe}/;
185	@_	655	}
186	}, $class;
187		656
188	run_template;	657	$PERL = $perl;
		658	}
		659
		660	require Proc::FastSpawn;
		661
		662	my ($fh, $slave) = AnyEvent::Util::portable_socketpair;
		663	Proc::FastSpawn::fd_inherit (fileno $slave);
		664
		665	# new fh's should always be set cloexec (due to $^F),
		666	# but hey, not on win32, so we always clear the inherit flag.
		667	Proc::FastSpawn::fd_inherit (fileno $fh, 0);
		668
		669	# quick. also doesn't work in win32. of course. what did you expect
		670	#local $ENV{PERL5LIB} = join ":", grep !ref, @INC;
		671	my %env = %ENV;
		672	$env{PERL5LIB} = join +($^O eq "MSWin32" ? ";" : ":"), grep !ref, @INC;
		673
		674	my $pid = Proc::FastSpawn::spawn (
		675	$PERL,
		676	["perl", "-MAnyEvent::Fork::Serve", "-e", "AnyEvent::Fork::Serve::me", fileno $slave, $$],
		677	[map "$_=$env{$_}", keys %env],
		678	) or die "unable to spawn AnyEvent::Fork server: $!";
		679
		680	$self->_new ($fh, $pid)
		681	}
		682
		683	=item $pid = $proc->pid
		684
		685	Returns the process id of the process I<iff it is a direct child of the
		686	process running AnyEvent::Fork>, and C<undef> otherwise. As a general
		687	rule (that you cannot rely upon), processes created via C<new_exec>,
		688	L<AnyEvent::Fork::Early> or L<AnyEvent::Fork::Template> are direct
		689	children, while all other processes are not.
		690
		691	Or in other words, you do not normally have to take care of zombies for
		692	processes created via C<new>, but when in doubt, or zombies are a problem,
		693	you need to check whether a process is a diretc child by calling this
		694	method, and possibly creating a child watcher or reap it manually.
		695
		696	=cut
		697
		698	sub pid {
		699	$_[0][PID]
		700	}
		701
		702	=item $proc = $proc->eval ($perlcode, @args)
		703
		704	Evaluates the given C<$perlcode> as ... Perl code, while setting C<@_> to
		705	the strings specified by C<@args>, in the "main" package.
		706
		707	This call is meant to do any custom initialisation that might be required
		708	(for example, the C<require> method uses it). It's not supposed to be used
		709	to completely take over the process, use C<run> for that.
		710
		711	The code will usually be executed after this call returns, and there is no
		712	way to pass anything back to the calling process. Any evaluation errors
		713	will be reported to stderr and cause the process to exit.
		714
		715	If you want to execute some code (that isn't in a module) to take over the
		716	process, you should compile a function via C<eval> first, and then call
		717	it via C<run>. This also gives you access to any arguments passed via the
		718	C<send_xxx> methods, such as file handles. See the L<use AnyEvent::Fork as
		719	a faster fork+exec> example to see it in action.
		720
		721	Returns the process object for easy chaining of method calls.
		722
		723	=cut
		724
		725	sub eval {
		726	my ($self, $code, @args) = @_;
		727
		728	$self->_cmd (e => pack "(w/a)", $code, @args);
189		729
190	$self	730	$self
191	}	731	}
192		732
		733	=item $proc = $proc->require ($module, ...)
		734
		735	Tries to load the given module(s) into the process
		736
		737	Returns the process object for easy chaining of method calls.
		738
		739	=cut
		740
		741	sub require {
		742	my ($self, @modules) = @_;
		743
		744	s%::%/%g for @modules;
		745	$self->eval ('require "$_.pm" for @_', @modules);
		746
		747	$self
		748	}
		749
		750	=item $proc = $proc->send_fh ($handle, ...)
		751
		752	Send one or more file handles (I<not> file descriptors) to the process,
		753	to prepare a call to C<run>.
		754
		755	The process object keeps a reference to the handles until they have
		756	been passed over to the process, so you must not explicitly close the
		757	handles. This is most easily accomplished by simply not storing the file
		758	handles anywhere after passing them to this method - when AnyEvent::Fork
		759	is finished using them, perl will automatically close them.
		760
		761	Returns the process object for easy chaining of method calls.
		762
		763	Example: pass a file handle to a process, and release it without
		764	closing. It will be closed automatically when it is no longer used.
		765
		766	$proc->send_fh ($my_fh);
		767	undef $my_fh; # free the reference if you want, but DO NOT CLOSE IT
		768
		769	=cut
		770
		771	sub send_fh {
		772	my ($self, @fh) = @_;
		773
		774	for my $fh (@fh) {
		775	$self->_cmd ("h");
		776	push @{ $self->[QUEUE] }, \$fh;
		777	}
		778
		779	$self
		780	}
		781
		782	=item $proc = $proc->send_arg ($string, ...)
		783
		784	Send one or more argument strings to the process, to prepare a call to
		785	C<run>. The strings can be any octet strings.
		786
		787	The protocol is optimised to pass a moderate number of relatively short
		788	strings - while you can pass up to 4GB of data in one go, this is more
		789	meant to pass some ID information or other startup info, not big chunks of
		790	data.
		791
		792	Returns the process object for easy chaining of method calls.
		793
		794	=cut
		795
		796	sub send_arg {
		797	my ($self, @arg) = @_;
		798
		799	$self->_cmd (a => pack "(w/a)", @arg);
		800
		801	$self
		802	}
		803
		804	=item $proc->run ($func, $cb->($fh))
		805
		806	Enter the function specified by the function name in C<$func> in the
		807	process. The function is called with the communication socket as first
		808	argument, followed by all file handles and string arguments sent earlier
		809	via C<send_fh> and C<send_arg> methods, in the order they were called.
		810
		811	The process object becomes unusable on return from this function - any
		812	further method calls result in undefined behaviour.
		813
		814	The function name should be fully qualified, but if it isn't, it will be
		815	looked up in the C<main> package.
		816
		817	If the called function returns, doesn't exist, or any error occurs, the
		818	process exits.
		819
		820	Preparing the process is done in the background - when all commands have
		821	been sent, the callback is invoked with the local communications socket
		822	as argument. At this point you can start using the socket in any way you
		823	like.
		824
		825	If the communication socket isn't used, it should be closed on both sides,
		826	to save on kernel memory.
		827
		828	The socket is non-blocking in the parent, and blocking in the newly
		829	created process. The close-on-exec flag is set in both.
		830
		831	Even if not used otherwise, the socket can be a good indicator for the
		832	existence of the process - if the other process exits, you get a readable
		833	event on it, because exiting the process closes the socket (if it didn't
		834	create any children using fork).
		835
		836	=over 4
		837
		838	=item Compatibility to L<AnyEvent::Fork::Remote>
		839
		840	If you want to write code that works with both this module and
		841	L<AnyEvent::Fork::Remote>, you need to write your code so that it assumes
		842	there are two file handles for communications, which might not be unix
		843	domain sockets. The C<run> function should start like this:
		844
		845	sub run {
		846	my ($rfh, @args) = @_; # @args is your normal arguments
		847	my $wfh = fileno $rfh ? $rfh : *STDOUT;
		848
		849	# now use $rfh for reading and $wfh for writing
		850	}
		851
		852	This checks whether the passed file handle is, in fact, the process
		853	C<STDIN> handle. If it is, then the function was invoked visa
		854	L<AnyEvent::Fork::Remote>, so STDIN should be used for reading and
		855	C<STDOUT> should be used for writing.
		856
		857	In all other cases, the function was called via this module, and there is
		858	only one file handle that should be sued for reading and writing.
		859
193	=back	860	=back
194		861
195	=head1 AUTHOR	862	Example: create a template for a process pool, pass a few strings, some
		863	file handles, then fork, pass one more string, and run some code.
		864
		865	my $pool = AnyEvent::Fork
		866	->new
		867	->send_arg ("str1", "str2")
		868	->send_fh ($fh1, $fh2);
		869
		870	for (1..2) {
		871	$pool
		872	->fork
		873	->send_arg ("str3")
		874	->run ("Some::function", sub {
		875	my ($fh) = @_;
		876
		877	# fh is nonblocking, but we trust that the OS can accept these
		878	# few octets anyway.
		879	syswrite $fh, "hi #$_\n";
		880
		881	# $fh is being closed here, as we don't store it anywhere
		882	});
		883	}
		884
		885	# Some::function might look like this - all parameters passed before fork
		886	# and after will be passed, in order, after the communications socket.
		887	sub Some::function {
		888	my ($fh, $str1, $str2, $fh1, $fh2, $str3) = @_;
		889
		890	print scalar <$fh>; # prints "hi #1\n" and "hi #2\n" in any order
		891	}
		892
		893	=cut
		894
		895	sub run {
		896	my ($self, $func, $cb) = @_;
		897
		898	$self->[CB] = $cb;
		899	$self->_cmd (r => $func);
		900	}
		901
		902	=back
		903
		904	=head2 EXPERIMENTAL METHODS
		905
		906	These methods might go away completely or change behaviour, at any time.
		907
		908	=over 4
		909
		910	=item $proc->to_fh ($cb->($fh)) # EXPERIMENTAL, MIGHT BE REMOVED
		911
		912	Flushes all commands out to the process and then calls the callback with
		913	the communications socket.
		914
		915	The process object becomes unusable on return from this function - any
		916	further method calls result in undefined behaviour.
		917
		918	The point of this method is to give you a file handle that you can pass
		919	to another process. In that other process, you can call C<new_from_fh
		920	AnyEvent::Fork $fh> to create a new C<AnyEvent::Fork> object from it,
		921	thereby effectively passing a fork object to another process.
		922
		923	=cut
		924
		925	sub to_fh {
		926	my ($self, $cb) = @_;
		927
		928	$self->[CB] = $cb;
		929
		930	unless ($self->[WW]) {
		931	$self->[CB]->($self->[FH]);
		932	@$self = ();
		933	}
		934	}
		935
		936	=item new_from_fh AnyEvent::Fork $fh # EXPERIMENTAL, MIGHT BE REMOVED
		937
		938	Takes a file handle originally rceeived by the C<to_fh> method and creates
		939	a new C<AnyEvent:Fork> object. The child process itself will not change in
		940	any way, i.e. it will keep all the modifications done to it before calling
		941	C<to_fh>.
		942
		943	The new object is very much like the original object, except that the
		944	C<pid> method will return C<undef> even if the process is a direct child.
		945
		946	=cut
		947
		948	sub new_from_fh {
		949	my ($class, $fh) = @_;
		950
		951	$class->_new ($fh)
		952	}
		953
		954	=back
		955
		956	=head1 PERFORMANCE
		957
		958	Now for some unscientific benchmark numbers (all done on an amd64
		959	GNU/Linux box). These are intended to give you an idea of the relative
		960	performance you can expect, they are not meant to be absolute performance
		961	numbers.
		962
		963	OK, so, I ran a simple benchmark that creates a socket pair, forks, calls
		964	exit in the child and waits for the socket to close in the parent. I did
		965	load AnyEvent, EV and AnyEvent::Fork, for a total process size of 5100kB.
		966
		967	2079 new processes per second, using manual socketpair + fork
		968
		969	Then I did the same thing, but instead of calling fork, I called
		970	AnyEvent::Fork->new->run ("CORE::exit") and then again waited for the
		971	socket from the child to close on exit. This does the same thing as manual
		972	socket pair + fork, except that what is forked is the template process
		973	(2440kB), and the socket needs to be passed to the server at the other end
		974	of the socket first.
		975
		976	2307 new processes per second, using AnyEvent::Fork->new
		977
		978	And finally, using C<new_exec> instead C<new>, using vforks+execs to exec
		979	a new perl interpreter and compile the small server each time, I get:
		980
		981	479 vfork+execs per second, using AnyEvent::Fork->new_exec
		982
		983	So how can C<< AnyEvent->new >> be faster than a standard fork, even
		984	though it uses the same operations, but adds a lot of overhead?
		985
		986	The difference is simply the process size: forking the 5MB process takes
		987	so much longer than forking the 2.5MB template process that the extra
		988	overhead is canceled out.
		989
		990	If the benchmark process grows, the normal fork becomes even slower:
		991
		992	1340 new processes, manual fork of a 20MB process
		993	731 new processes, manual fork of a 200MB process
		994	235 new processes, manual fork of a 2000MB process
		995
		996	What that means (to me) is that I can use this module without having a bad
		997	conscience because of the extra overhead required to start new processes.
		998
		999	=head1 TYPICAL PROBLEMS
		1000
		1001	This section lists typical problems that remain. I hope by recognising
		1002	them, most can be avoided.
		1003
		1004	=over 4
		1005
		1006	=item leaked file descriptors for exec'ed processes
		1007
		1008	POSIX systems inherit file descriptors by default when exec'ing a new
		1009	process. While perl itself laudably sets the close-on-exec flags on new
		1010	file handles, most C libraries don't care, and even if all cared, it's
		1011	often not possible to set the flag in a race-free manner.
		1012
		1013	That means some file descriptors can leak through. And since it isn't
		1014	possible to know which file descriptors are "good" and "necessary" (or
		1015	even to know which file descriptors are open), there is no good way to
		1016	close the ones that might harm.
		1017
		1018	As an example of what "harm" can be done consider a web server that
		1019	accepts connections and afterwards some module uses AnyEvent::Fork for the
		1020	first time, causing it to fork and exec a new process, which might inherit
		1021	the network socket. When the server closes the socket, it is still open
		1022	in the child (which doesn't even know that) and the client might conclude
		1023	that the connection is still fine.
		1024
		1025	For the main program, there are multiple remedies available -
		1026	L<AnyEvent::Fork::Early> is one, creating a process early and not using
		1027	C<new_exec> is another, as in both cases, the first process can be exec'ed
		1028	well before many random file descriptors are open.
		1029
		1030	In general, the solution for these kind of problems is to fix the
		1031	libraries or the code that leaks those file descriptors.
		1032
		1033	Fortunately, most of these leaked descriptors do no harm, other than
		1034	sitting on some resources.
		1035
		1036	=item leaked file descriptors for fork'ed processes
		1037
		1038	Normally, L<AnyEvent::Fork> does start new processes by exec'ing them,
		1039	which closes file descriptors not marked for being inherited.
		1040
		1041	However, L<AnyEvent::Fork::Early> and L<AnyEvent::Fork::Template> offer
		1042	a way to create these processes by forking, and this leaks more file
		1043	descriptors than exec'ing them, as there is no way to mark descriptors as
		1044	"close on fork".
		1045
		1046	An example would be modules like L<EV>, L<IO::AIO> or L<Gtk2>. Both create
		1047	pipes for internal uses, and L<Gtk2> might open a connection to the X
		1048	server. L<EV> and L<IO::AIO> can deal with fork, but Gtk2 might have
		1049	trouble with a fork.
		1050
		1051	The solution is to either not load these modules before use'ing
		1052	L<AnyEvent::Fork::Early> or L<AnyEvent::Fork::Template>, or to delay
		1053	initialising them, for example, by calling C<init Gtk2> manually.
		1054
		1055	=item exiting calls object destructors
		1056
		1057	This only applies to users of L<AnyEvent::Fork:Early> and
		1058	L<AnyEvent::Fork::Template>, or when initialising code creates objects
		1059	that reference external resources.
		1060
		1061	When a process created by AnyEvent::Fork exits, it might do so by calling
		1062	exit, or simply letting perl reach the end of the program. At which point
		1063	Perl runs all destructors.
		1064
		1065	Not all destructors are fork-safe - for example, an object that represents
		1066	the connection to an X display might tell the X server to free resources,
		1067	which is inconvenient when the "real" object in the parent still needs to
		1068	use them.
		1069
		1070	This is obviously not a problem for L<AnyEvent::Fork::Early>, as you used
		1071	it as the very first thing, right?
		1072
		1073	It is a problem for L<AnyEvent::Fork::Template> though - and the solution
		1074	is to not create objects with nontrivial destructors that might have an
		1075	effect outside of Perl.
		1076
		1077	=back
		1078
		1079	=head1 PORTABILITY NOTES
		1080
		1081	Native win32 perls are somewhat supported (AnyEvent::Fork::Early is a nop,
		1082	and ::Template is not going to work), and it cost a lot of blood and sweat
		1083	to make it so, mostly due to the bloody broken perl that nobody seems to
		1084	care about. The fork emulation is a bad joke - I have yet to see something
		1085	useful that you can do with it without running into memory corruption
		1086	issues or other braindamage. Hrrrr.
		1087
		1088	Since fork is endlessly broken on win32 perls (it doesn't even remotely
		1089	work within it's documented limits) and quite obviously it's not getting
		1090	improved any time soon, the best way to proceed on windows would be to
		1091	always use C<new_exec> and thus never rely on perl's fork "emulation".
		1092
		1093	Cygwin perl is not supported at the moment due to some hilarious
		1094	shortcomings of its API - see L<IO::FDPoll> for more details. If you never
		1095	use C<send_fh> and always use C<new_exec> to create processes, it should
		1096	work though.
		1097
		1098	=head1 USING AnyEvent::Fork IN SUBPROCESSES
		1099
		1100	AnyEvent::Fork itself cannot generally be used in subprocesses. As long as
		1101	only one process ever forks new processes, sharing the template processes
		1102	is possible (you could use a pipe as a lock by writing a byte into it to
		1103	unlock, and reading the byte to lock for example)
		1104
		1105	To make concurrent calls possible after fork, you should get rid of the
		1106	template and early fork processes. AnyEvent::Fork will create a new
		1107	template process as needed.
		1108
		1109	undef $AnyEvent::Fork::EARLY;
		1110	undef $AnyEvent::Fork::TEMPLATE;
		1111
		1112	It doesn't matter whether you get rid of them in the parent or child after
		1113	a fork.
		1114
		1115	=head1 SEE ALSO
		1116
		1117	L<AnyEvent::Fork::Early>, to avoid executing a perl interpreter at all
		1118	(part of this distribution).
		1119
		1120	L<AnyEvent::Fork::Template>, to create a process by forking the main
		1121	program at a convenient time (part of this distribution).
		1122
		1123	L<AnyEvent::Fork::Remote>, for another way to create processes that is
		1124	mostly compatible to this module and modules building on top of it, but
		1125	works better with remote processes.
		1126
		1127	L<AnyEvent::Fork::RPC>, for simple RPC to child processes (on CPAN).
		1128
		1129	L<AnyEvent::Fork::Pool>, for simple worker process pool (on CPAN).
		1130
		1131	=head1 AUTHOR AND CONTACT INFORMATION
196		1132
197	Marc Lehmann <schmorp@schmorp.de>	1133	Marc Lehmann <schmorp@schmorp.de>
198	http://home.schmorp.de/	1134	http://software.schmorp.de/pkg/AnyEvent-Fork
199		1135
200	=cut	1136	=cut
201		1137
202	1	1138	1
203		1139

Diff Legend

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

Comparing AnyEvent-Fork/Fork.pm (file contents): Revision 1.2 by root, Sun Mar 31 03:26:50 2013 UTC vs. Revision 1.63 by root, Wed Nov 26 13:36:18 2014 UTC

Diff Legend

Comparing AnyEvent-Fork/Fork.pm (file contents):
Revision 1.2 by root, Sun Mar 31 03:26:50 2013 UTC vs.
Revision 1.63 by root, Wed Nov 26 13:36:18 2014 UTC