ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/AnyEvent-Fork/Fork.pm
(Generate patch)

Comparing AnyEvent-Fork/Fork.pm (file contents):
Revision 1.10 by root, Thu Apr 4 06:09:15 2013 UTC vs.
Revision 1.27 by root, Sat Apr 6 09:05:50 2013 UTC

1=head1 NAME 1=head1 NAME
2 2
3AnyEvent::Fork - everything you wanted to use fork() for, but couldn't 3AnyEvent::Fork - everything you wanted to use fork() for, but couldn't
4 4
5ATTENTION, this is a very early release, and very untested. Consider it a
6technology preview.
7
8=head1 SYNOPSIS 5=head1 SYNOPSIS
9 6
10 use AnyEvent::Fork; 7 use AnyEvent::Fork;
11 8
12 ################################################################## 9 AnyEvent::Fork
10 ->new
11 ->require ("MyModule")
12 ->run ("MyModule::server", my $cv = AE::cv);
13
14 my $fh = $cv->recv;
15
16=head1 DESCRIPTION
17
18This module allows you to create new processes, without actually forking
19them from your current process (avoiding the problems of forking), but
20preserving most of the advantages of fork.
21
22It can be used to create new worker processes or new independent
23subprocesses for short- and long-running jobs, process pools (e.g. for use
24in pre-forked servers) but also to spawn new external processes (such as
25CGI scripts from a web server), which can be faster (and more well behaved)
26than using fork+exec in big processes.
27
28Special care has been taken to make this module useful from other modules,
29while still supporting specialised environments such as L<App::Staticperl>
30or L<PAR::Packer>.
31
32=head1 WHAT THIS MODULE IS NOT
33
34This module only creates processes and lets you pass file handles and
35strings to it, and run perl code. It does not implement any kind of RPC -
36there is no back channel from the process back to you, and there is no RPC
37or message passing going on.
38
39If you need some form of RPC, you can either implement it yourself
40in whatever way you like, use some message-passing module such
41as L<AnyEvent::MP>, some pipe such as L<AnyEvent::ZeroMQ>, use
42L<AnyEvent::Handle> on both sides to send e.g. JSON or Storable messages,
43and so on.
44
45=head1 PROBLEM STATEMENT
46
47There are two traditional ways to implement parallel processing on UNIX
48like operating systems - fork and process, and fork+exec and process. They
49have different advantages and disadvantages that I describe below,
50together with how this module tries to mitigate the disadvantages.
51
52=over 4
53
54=item Forking from a big process can be very slow.
55
56A 5GB process needs 0.05s to fork on my 3.6GHz amd64 GNU/Linux box. This
57overhead is often shared with exec (because you have to fork first), but
58in some circumstances (e.g. when vfork is used), fork+exec can be much
59faster.
60
61This module can help here by telling a small(er) helper process to fork,
62which is faster then forking the main process, and also uses vfork where
63possible. This gives the speed of vfork, with the flexibility of fork.
64
65=item Forking usually creates a copy-on-write copy of the parent
66process.
67
68For example, modules or data files that are loaded will not use additional
69memory after a fork. When exec'ing a new process, modules and data files
70might need to be loaded again, at extra CPU and memory cost. But when
71forking, literally all data structures are copied - if the program frees
72them and replaces them by new data, the child processes will retain the
73old version even if it isn't used, which can suddenly and unexpectedly
74increase memory usage when freeing memory.
75
76The trade-off is between more sharing with fork (which can be good or
77bad), and no sharing with exec.
78
79This module allows the main program to do a controlled fork, and allows
80modules to exec processes safely at any time. When creating a custom
81process pool you can take advantage of data sharing via fork without
82risking to share large dynamic data structures that will blow up child
83memory usage.
84
85In other words, this module puts you into control over what is being
86shared and what isn't, at all times.
87
88=item Exec'ing a new perl process might be difficult.
89
90For example, it is not easy to find the correct path to the perl
91interpreter - C<$^X> might not be a perl interpreter at all.
92
93This module tries hard to identify the correct path to the perl
94interpreter. With a cooperative main program, exec'ing the interpreter
95might not even be necessary, but even without help from the main program,
96it will still work when used from a module.
97
98=item Exec'ing a new perl process might be slow, as all necessary modules
99have to be loaded from disk again, with no guarantees of success.
100
101Long running processes might run into problems when perl is upgraded
102and modules are no longer loadable because they refer to a different
103perl version, or parts of a distribution are newer than the ones already
104loaded.
105
106This module supports creating pre-initialised perl processes to be used as
107a template for new processes.
108
109=item Forking might be impossible when a program is running.
110
111For example, POSIX makes it almost impossible to fork from a
112multi-threaded program while doing anything useful in the child - in
113fact, if your perl program uses POSIX threads (even indirectly via
114e.g. L<IO::AIO> or L<threads>), you cannot call fork on the perl level
115anymore without risking corruption issues on a number of operating
116systems.
117
118This module can safely fork helper processes at any time, by calling
119fork+exec in C, in a POSIX-compatible way (via L<Proc::FastSpawn>).
120
121=item Parallel processing with fork might be inconvenient or difficult
122to implement. Modules might not work in both parent and child.
123
124For example, when a program uses an event loop and creates watchers it
125becomes very hard to use the event loop from a child program, as the
126watchers already exist but are only meaningful in the parent. Worse, a
127module might want to use such a module, not knowing whether another module
128or the main program also does, leading to problems.
129
130Apart from event loops, graphical toolkits also commonly fall into the
131"unsafe module" category, or just about anything that communicates with
132the external world, such as network libraries and file I/O modules, which
133usually don't like being copied and then allowed to continue in two
134processes.
135
136With this module only the main program is allowed to create new processes
137by forking (because only the main program can know when it is still safe
138to do so) - all other processes are created via fork+exec, which makes it
139possible to use modules such as event loops or window interfaces safely.
140
141=back
142
143=head1 EXAMPLES
144
13 # create a single new process, tell it to run your worker function 145=head2 Create a single new process, tell it to run your worker function.
14 146
15 AnyEvent::Fork 147 AnyEvent::Fork
16 ->new 148 ->new
17 ->require ("MyModule") 149 ->require ("MyModule")
18 ->run ("MyModule::worker, sub { 150 ->run ("MyModule::worker, sub {
… …
28 160
29 # now $slave_filehandle is connected to the $master_filehandle 161 # now $slave_filehandle is connected to the $master_filehandle
30 # in the original prorcess. have fun! 162 # in the original prorcess. have fun!
31 } 163 }
32 164
33 ##################################################################
34 # create a pool of server processes all accepting on the same socket 165=head2 Create a pool of server processes all accepting on the same socket.
35 166
36 # create listener socket 167 # create listener socket
37 my $listener = ...; 168 my $listener = ...;
38 169
39 # create a pool template, initialise it and give it the socket 170 # create a pool template, initialise it and give it the socket
… …
64 while (my $socket = $listener->accept) { 195 while (my $socket = $listener->accept) {
65 # do sth. with new socket 196 # do sth. with new socket
66 } 197 }
67 } 198 }
68 199
69=head1 DESCRIPTION 200=head2 use AnyEvent::Fork as a faster fork+exec
70 201
71This module allows you to create new processes, without actually forking 202This runs /bin/echo hi, with stdout redirected to /tmp/log and stderr to
72them from your current process (avoiding the problems of forking), but 203the communications socket. It is usually faster than fork+exec, but still
73preserving most of the advantages of fork. 204let's you prepare the environment.
74 205
75It can be used to create new worker processes or new independent 206 open my $output, ">/tmp/log" or die "$!";
76subprocesses for short- and long-running jobs, process pools (e.g. for use
77in pre-forked servers) but also to spawn new external processes (such as
78CGI scripts from a webserver), which can be faster (and more well behaved)
79than using fork+exec in big processes.
80 207
81Special care has been taken to make this module useful from other modules, 208 AnyEvent::Fork
82while still supporting specialised environments such as L<App::Staticperl> 209 ->new
83or L<PAR::Packer>. 210 ->eval ('
211 sub run {
212 my ($fh, $output, @cmd) = @_;
84 213
85=head1 PROBLEM STATEMENT 214 # perl will clear close-on-exec on STDOUT/STDERR
215 open STDOUT, ">&", $output or die;
216 open STDERR, ">&", $fh or die;
86 217
87There are two ways to implement parallel processing on UNIX like operating 218 exec @cmd;
88systems - fork and process, and fork+exec and process. They have different 219 }
89advantages and disadvantages that I describe below, together with how this 220 ')
90module tries to mitigate the disadvantages. 221 ->send_fh ($output)
222 ->send_arg ("/bin/echo", "hi")
223 ->run ("run", my $cv = AE::cv);
91 224
92=over 4 225 my $stderr = $cv->recv;
93
94=item Forking from a big process can be very slow (a 5GB process needs
950.05s to fork on my 3.6GHz amd64 GNU/Linux box for example). This overhead
96is often shared with exec (because you have to fork first), but in some
97circumstances (e.g. when vfork is used), fork+exec can be much faster.
98
99This module can help here by telling a small(er) helper process to fork,
100or fork+exec instead.
101
102=item Forking usually creates a copy-on-write copy of the parent
103process. Memory (for example, modules or data files that have been
104will not take additional memory). When exec'ing a new process, modules
105and data files might need to be loaded again, at extra cpu and memory
106cost. Likewise when forking, all data structures are copied as well - if
107the program frees them and replaces them by new data, the child processes
108will retain the memory even if it isn't used.
109
110This module allows the main program to do a controlled fork, and allows
111modules to exec processes safely at any time. When creating a custom
112process pool you can take advantage of data sharing via fork without
113risking to share large dynamic data structures that will blow up child
114memory usage.
115
116=item Exec'ing a new perl process might be difficult and slow. For
117example, it is not easy to find the correct path to the perl interpreter,
118and all modules have to be loaded from disk again. Long running processes
119might run into problems when perl is upgraded for example.
120
121This module supports creating pre-initialised perl processes to be used
122as template, and also tries hard to identify the correct path to the perl
123interpreter. With a cooperative main program, exec'ing the interpreter
124might not even be necessary.
125
126=item Forking might be impossible when a program is running. For example,
127POSIX makes it almost impossible to fork from a multithreaded program and
128do anything useful in the child - strictly speaking, if your perl program
129uses posix threads (even indirectly via e.g. L<IO::AIO> or L<threads>),
130you cannot call fork on the perl level anymore, at all.
131
132This module can safely fork helper processes at any time, by caling
133fork+exec in C, in a POSIX-compatible way.
134
135=item Parallel processing with fork might be inconvenient or difficult
136to implement. For example, when a program uses an event loop and creates
137watchers it becomes very hard to use the event loop from a child
138program, as the watchers already exist but are only meaningful in the
139parent. Worse, a module might want to use such a system, not knowing
140whether another module or the main program also does, leading to problems.
141
142This module only lets the main program create pools by forking (because
143only the main program can know when it is still safe to do so) - all other
144pools are created by fork+exec, after which such modules can again be
145loaded.
146
147=back
148 226
149=head1 CONCEPTS 227=head1 CONCEPTS
150 228
151This module can create new processes either by executing a new perl 229This module can create new processes either by executing a new perl
152process, or by forking from an existing "template" process. 230process, or by forking from an existing "template" process.
… …
169needed the first time. Forking from this process shares the memory used 247needed the first time. Forking from this process shares the memory used
170for the perl interpreter with the new process, but loading modules takes 248for the perl interpreter with the new process, but loading modules takes
171time, and the memory is not shared with anything else. 249time, and the memory is not shared with anything else.
172 250
173This is ideal for when you only need one extra process of a kind, with the 251This is ideal for when you only need one extra process of a kind, with the
174option of starting and stipping it on demand. 252option of starting and stopping it on demand.
175 253
176Example: 254Example:
177 255
178 AnyEvent::Fork 256 AnyEvent::Fork
179 ->new 257 ->new
… …
194modules you loaded) is shared between the processes, and each new process 272modules you loaded) is shared between the processes, and each new process
195consumes relatively little memory of its own. 273consumes relatively little memory of its own.
196 274
197The disadvantage of this approach is that you need to create a template 275The disadvantage of this approach is that you need to create a template
198process for the sole purpose of forking new processes from it, but if you 276process for the sole purpose of forking new processes from it, but if you
199only need a fixed number of proceses you can create them, and then destroy 277only need a fixed number of processes you can create them, and then destroy
200the template process. 278the template process.
201 279
202Example: 280Example:
203 281
204 my $template = AnyEvent::Fork->new->require ("Some::Module"); 282 my $template = AnyEvent::Fork->new->require ("Some::Module");
… …
231 my ($fork_fh) = @_; 309 my ($fork_fh) = @_;
232 }); 310 });
233 311
234=back 312=back
235 313
236=head1 FUNCTIONS 314=head1 THE C<AnyEvent::Fork> CLASS
315
316This module exports nothing, and only implements a single class -
317C<AnyEvent::Fork>.
318
319There are two constructors that both create new processes - C<new> and
320C<new_exec>. The C<fork> method creates a new process by forking an
321existing one and could be considered a third constructor.
322
323Most of the remaining methods deal with preparing the new process, by
324loading code, evaluating code and sending data to the new process. They
325usually return the process object, so you can chain method calls.
326
327If a process object is destroyed before calling its C<run> method, then
328the process simply exits. After C<run> is called, all responsibility is
329passed to the specified function.
237 330
238=over 4 331=over 4
239 332
240=cut 333=cut
241 334
242package AnyEvent::Fork; 335package AnyEvent::Fork;
243 336
244use common::sense; 337use common::sense;
245 338
246use Socket (); 339use Errno ();
247 340
248use AnyEvent; 341use AnyEvent;
249use AnyEvent::Fork::Util;
250use AnyEvent::Util (); 342use AnyEvent::Util ();
343
344use IO::FDPass;
345
346our $VERSION = 0.5;
251 347
252our $PERL; # the path to the perl interpreter, deduces with various forms of magic 348our $PERL; # the path to the perl interpreter, deduces with various forms of magic
253 349
254=item my $pool = new AnyEvent::Fork key => value... 350=item my $pool = new AnyEvent::Fork key => value...
255 351
… …
268our $TEMPLATE; 364our $TEMPLATE;
269 365
270sub _cmd { 366sub _cmd {
271 my $self = shift; 367 my $self = shift;
272 368
273 #TODO: maybe append the packet to any existing string command already in the queue
274
275 # ideally, we would want to use "a (w/a)*" as format string, but perl versions 369 # ideally, we would want to use "a (w/a)*" as format string, but perl
276 # from at least 5.8.9 to 5.16.3 are all buggy and can't unpack it. 370 # versions from at least 5.8.9 to 5.16.3 are all buggy and can't unpack
277 push @{ $self->[2] }, pack "N/a", pack "(w/a)*", @_; 371 # it.
372 push @{ $self->[2] }, pack "a L/a*", $_[0], $_[1];
278 373
279 $self->[3] ||= AE::io $self->[1], 1, sub { 374 $self->[3] ||= AE::io $self->[1], 1, sub {
375 do {
280 # send the next "thing" in the queue - either a reference to an fh, 376 # send the next "thing" in the queue - either a reference to an fh,
281 # or a plain string. 377 # or a plain string.
282 378
283 if (ref $self->[2][0]) { 379 if (ref $self->[2][0]) {
284 # send fh 380 # send fh
285 AnyEvent::Fork::Util::fd_send fileno $self->[1], fileno ${ $self->[2][0] } 381 unless (IO::FDPass::send fileno $self->[1], fileno ${ $self->[2][0] }) {
382 return if $! == Errno::EAGAIN || $! == Errno::EWOULDBLOCK;
383 undef $self->[3];
384 die "AnyEvent::Fork: file descriptor send failure: $!";
385 }
386
286 and shift @{ $self->[2] }; 387 shift @{ $self->[2] };
287 388
288 } else { 389 } else {
289 # send string 390 # send string
290 my $len = syswrite $self->[1], $self->[2][0] 391 my $len = syswrite $self->[1], $self->[2][0];
392
393 unless ($len) {
394 return if $! == Errno::EAGAIN || $! == Errno::EWOULDBLOCK;
395 undef $self->[3];
291 or do { undef $self->[3]; die "AnyEvent::Fork: command write failure: $!" }; 396 die "AnyEvent::Fork: command write failure: $!";
397 }
292 398
293 substr $self->[2][0], 0, $len, ""; 399 substr $self->[2][0], 0, $len, "";
294 shift @{ $self->[2] } unless length $self->[2][0]; 400 shift @{ $self->[2] } unless length $self->[2][0];
295 } 401 }
402 } while @{ $self->[2] };
296 403
297 unless (@{ $self->[2] }) { 404 # everything written
298 undef $self->[3]; 405 undef $self->[3];
406
299 # invoke run callback 407 # invoke run callback, if any
300 $self->[0]->($self->[1]) if $self->[0]; 408 $self->[4]->($self->[1]) if $self->[4];
301 }
302 }; 409 };
410
411 () # make sure we don't leak the watcher
303} 412}
304 413
305sub _new { 414sub _new {
306 my ($self, $fh) = @_; 415 my ($self, $fh, $pid) = @_;
307 416
308 AnyEvent::Util::fh_nonblocking $fh, 1; 417 AnyEvent::Util::fh_nonblocking $fh, 1;
309 418
310 $self = bless [ 419 $self = bless [
311 undef, # run callback 420 $pid,
312 $fh, 421 $fh,
313 [], # write queue - strings or fd's 422 [], # write queue - strings or fd's
314 undef, # AE watcher 423 undef, # AE watcher
315 ], $self; 424 ], $self;
316 425
… …
327 if ($pid eq 0) { 436 if ($pid eq 0) {
328 require AnyEvent::Fork::Serve; 437 require AnyEvent::Fork::Serve;
329 $AnyEvent::Fork::Serve::OWNER = $parent; 438 $AnyEvent::Fork::Serve::OWNER = $parent;
330 close $fh; 439 close $fh;
331 $0 = "$_[1] of $parent"; 440 $0 = "$_[1] of $parent";
441 $SIG{CHLD} = 'IGNORE';
332 AnyEvent::Fork::Serve::serve ($slave); 442 AnyEvent::Fork::Serve::serve ($slave);
333 AnyEvent::Fork::Util::_exit 0; 443 exit 0;
334 } elsif (!$pid) { 444 } elsif (!$pid) {
335 die "AnyEvent::Fork::Early/Template: unable to fork template process: $!"; 445 die "AnyEvent::Fork::Early/Template: unable to fork template process: $!";
336 } 446 }
337 447
338 AnyEvent::Fork->_new ($fh) 448 AnyEvent::Fork->_new ($fh, $pid)
339} 449}
340 450
341=item my $proc = new AnyEvent::Fork 451=item my $proc = new AnyEvent::Fork
342 452
343Create a new "empty" perl interpreter process and returns its process 453Create a new "empty" perl interpreter process and returns its process
… …
394reduces the amount of memory sharing that is possible, and is also slower. 504reduces the amount of memory sharing that is possible, and is also slower.
395 505
396You should use C<new> whenever possible, except when having a template 506You should use C<new> whenever possible, except when having a template
397process around is unacceptable. 507process around is unacceptable.
398 508
399The path to the perl interpreter is divined usign various methods - first 509The path to the perl interpreter is divined using various methods - first
400C<$^X> is investigated to see if the path ends with something that sounds 510C<$^X> is investigated to see if the path ends with something that sounds
401as if it were the perl interpreter. Failing this, the module falls back to 511as if it were the perl interpreter. Failing this, the module falls back to
402using C<$Config::Config{perlpath}>. 512using C<$Config::Config{perlpath}>.
403 513
404=cut 514=cut
… …
413 my $perl = $; 523 my $perl = $;
414 524
415 # first we try $^X, but the path must be absolute (always on win32), and end in sth. 525 # first we try $^X, but the path must be absolute (always on win32), and end in sth.
416 # that looks like perl. this obviously only works for posix and win32 526 # that looks like perl. this obviously only works for posix and win32
417 unless ( 527 unless (
418 (AnyEvent::Fork::Util::WIN32 || $perl =~ m%^/%) 528 ($^O eq "MSWin32" || $perl =~ m%^/%)
419 && $perl =~ m%[/\\]perl(?:[0-9]+(\.[0-9]+)+)?(\.exe)?$%i 529 && $perl =~ m%[/\\]perl(?:[0-9]+(\.[0-9]+)+)?(\.exe)?$%i
420 ) { 530 ) {
421 # if it doesn't look perlish enough, try Config 531 # if it doesn't look perlish enough, try Config
422 require Config; 532 require Config;
423 $perl = $Config::Config{perlpath}; 533 $perl = $Config::Config{perlpath};
… …
434 Proc::FastSpawn::fd_inherit (fileno $fh, 0); 544 Proc::FastSpawn::fd_inherit (fileno $fh, 0);
435 545
436 # quick. also doesn't work in win32. of course. what did you expect 546 # quick. also doesn't work in win32. of course. what did you expect
437 #local $ENV{PERL5LIB} = join ":", grep !ref, @INC; 547 #local $ENV{PERL5LIB} = join ":", grep !ref, @INC;
438 my %env = %ENV; 548 my %env = %ENV;
439 $env{PERL5LIB} = join +(AnyEvent::Fork::Util::WIN32 ? ";" : ":"), grep !ref, @INC; 549 $env{PERL5LIB} = join +($^O eq "MSWin32" ? ";" : ":"), grep !ref, @INC;
440 550
441 Proc::FastSpawn::spawn ( 551 my $pid = Proc::FastSpawn::spawn (
442 $perl, 552 $perl,
443 ["perl", "-MAnyEvent::Fork::Serve", "-e", "AnyEvent::Fork::Serve::me", fileno $slave, $$], 553 ["perl", "-MAnyEvent::Fork::Serve", "-e", "AnyEvent::Fork::Serve::me", fileno $slave, $$],
444 [map "$_=$env{$_}", keys %env], 554 [map "$_=$env{$_}", keys %env],
445 ) or die "unable to spawn AnyEvent::Fork server: $!"; 555 ) or die "unable to spawn AnyEvent::Fork server: $!";
446 556
447 $self->_new ($fh) 557 $self->_new ($fh, $pid)
558}
559
560=item $pid = $proc->pid
561
562Returns the process id of the process I<iff it is a direct child of the
563process> running AnyEvent::Fork, and C<undef> otherwise.
564
565Normally, only processes created via C<< AnyEvent::Fork->new_exec >> and
566L<AnyEvent::Fork::Template> are direct children, and you are responsible
567to clean up their zombies when they die.
568
569All other processes are not direct children, and will be cleaned up by
570AnyEvent::Fork.
571
572=cut
573
574sub pid {
575 $_[0][0]
448} 576}
449 577
450=item $proc = $proc->eval ($perlcode, @args) 578=item $proc = $proc->eval ($perlcode, @args)
451 579
452Evaluates the given C<$perlcode> as ... perl code, while setting C<@_> to 580Evaluates the given C<$perlcode> as ... perl code, while setting C<@_> to
453the strings specified by C<@args>. 581the strings specified by C<@args>, in the "main" package.
454 582
455This call is meant to do any custom initialisation that might be required 583This call is meant to do any custom initialisation that might be required
456(for example, the C<require> method uses it). It's not supposed to be used 584(for example, the C<require> method uses it). It's not supposed to be used
457to completely take over the process, use C<run> for that. 585to completely take over the process, use C<run> for that.
458 586
459The code will usually be executed after this call returns, and there is no 587The code will usually be executed after this call returns, and there is no
460way to pass anything back to the calling process. Any evaluation errors 588way to pass anything back to the calling process. Any evaluation errors
461will be reported to stderr and cause the process to exit. 589will be reported to stderr and cause the process to exit.
462 590
591If you want to execute some code to take over the process (see the
592"fork+exec" example in the SYNOPSIS), you should compile a function via
593C<eval> first, and then call it via C<run>. This also gives you access to
594any arguments passed via the C<send_xxx> methods, such as file handles.
595
463Returns the process object for easy chaining of method calls. 596Returns the process object for easy chaining of method calls.
464 597
465=cut 598=cut
466 599
467sub eval { 600sub eval {
468 my ($self, $code, @args) = @_; 601 my ($self, $code, @args) = @_;
469 602
470 $self->_cmd (e => $code, @args); 603 $self->_cmd (e => pack "(w/a*)*", $code, @args);
471 604
472 $self 605 $self
473} 606}
474 607
475=item $proc = $proc->require ($module, ...) 608=item $proc = $proc->require ($module, ...)
… …
499accomplished by simply not storing the file handles anywhere after passing 632accomplished by simply not storing the file handles anywhere after passing
500them to this method. 633them to this method.
501 634
502Returns the process object for easy chaining of method calls. 635Returns the process object for easy chaining of method calls.
503 636
504Example: pass an fh to a process, and release it without closing. it will 637Example: pass a file handle to a process, and release it without
505be closed automatically when it is no longer used. 638closing. It will be closed automatically when it is no longer used.
506 639
507 $proc->send_fh ($my_fh); 640 $proc->send_fh ($my_fh);
508 undef $my_fh; # free the reference if you want, but DO NOT CLOSE IT 641 undef $my_fh; # free the reference if you want, but DO NOT CLOSE IT
509 642
510=cut 643=cut
… …
523=item $proc = $proc->send_arg ($string, ...) 656=item $proc = $proc->send_arg ($string, ...)
524 657
525Send one or more argument strings to the process, to prepare a call to 658Send one or more argument strings to the process, to prepare a call to
526C<run>. The strings can be any octet string. 659C<run>. The strings can be any octet string.
527 660
661The protocol is optimised to pass a moderate number of relatively short
662strings - while you can pass up to 4GB of data in one go, this is more
663meant to pass some ID information or other startup info, not big chunks of
664data.
665
528Returns the process object for easy chaining of emthod calls. 666Returns the process object for easy chaining of method calls.
529 667
530=cut 668=cut
531 669
532sub send_arg { 670sub send_arg {
533 my ($self, @arg) = @_; 671 my ($self, @arg) = @_;
534 672
535 $self->_cmd (a => @arg); 673 $self->_cmd (a => pack "(w/a*)*", @arg);
536 674
537 $self 675 $self
538} 676}
539 677
540=item $proc->run ($func, $cb->($fh)) 678=item $proc->run ($func, $cb->($fh))
541 679
542Enter the function specified by the fully qualified name in C<$func> in 680Enter the function specified by the function name in C<$func> in the
543the process. The function is called with the communication socket as first 681process. The function is called with the communication socket as first
544argument, followed by all file handles and string arguments sent earlier 682argument, followed by all file handles and string arguments sent earlier
545via C<send_fh> and C<send_arg> methods, in the order they were called. 683via C<send_fh> and C<send_arg> methods, in the order they were called.
546 684
547If the called function returns, the process exits. 685The function name should be fully qualified, but if it isn't, it will be
686looked up in the main package.
548 687
549Preparing the process can take time - when the process is ready, the 688If the called function returns, doesn't exist, or any error occurs, the
689process exits.
690
691Preparing the process is done in the background - when all commands have
550callback is invoked with the local communications socket as argument. 692been sent, the callback is invoked with the local communications socket
693as argument. At this point you can start using the socket in any way you
694like.
551 695
552The process object becomes unusable on return from this function. 696The process object becomes unusable on return from this function - any
697further method calls result in undefined behaviour.
553 698
554If the communication socket isn't used, it should be closed on both sides, 699If the communication socket isn't used, it should be closed on both sides,
555to save on kernel memory. 700to save on kernel memory.
556 701
557The socket is non-blocking in the parent, and blocking in the newly 702The socket is non-blocking in the parent, and blocking in the newly
558created process. The close-on-exec flag is set on both. Even if not used 703created process. The close-on-exec flag is set in both.
704
559otherwise, the socket can be a good indicator for the existance of the 705Even if not used otherwise, the socket can be a good indicator for the
560process - if the other process exits, you get a readable event on it, 706existence of the process - if the other process exits, you get a readable
561because exiting the process closes the socket (if it didn't create any 707event on it, because exiting the process closes the socket (if it didn't
562children using fork). 708create any children using fork).
563 709
564Example: create a template for a process pool, pass a few strings, some 710Example: create a template for a process pool, pass a few strings, some
565file handles, then fork, pass one more string, and run some code. 711file handles, then fork, pass one more string, and run some code.
566 712
567 my $pool = AnyEvent::Fork 713 my $pool = AnyEvent::Fork
… …
575 ->send_arg ("str3") 721 ->send_arg ("str3")
576 ->run ("Some::function", sub { 722 ->run ("Some::function", sub {
577 my ($fh) = @_; 723 my ($fh) = @_;
578 724
579 # fh is nonblocking, but we trust that the OS can accept these 725 # fh is nonblocking, but we trust that the OS can accept these
580 # extra 3 octets anyway. 726 # few octets anyway.
581 syswrite $fh, "hi #$_\n"; 727 syswrite $fh, "hi #$_\n";
582 728
583 # $fh is being closed here, as we don't store it anywhere 729 # $fh is being closed here, as we don't store it anywhere
584 }); 730 });
585 } 731 }
… …
587 # Some::function might look like this - all parameters passed before fork 733 # Some::function might look like this - all parameters passed before fork
588 # and after will be passed, in order, after the communications socket. 734 # and after will be passed, in order, after the communications socket.
589 sub Some::function { 735 sub Some::function {
590 my ($fh, $str1, $str2, $fh1, $fh2, $str3) = @_; 736 my ($fh, $str1, $str2, $fh1, $fh2, $str3) = @_;
591 737
592 print scalar <$fh>; # prints "hi 1\n" and "hi 2\n" 738 print scalar <$fh>; # prints "hi #1\n" and "hi #2\n" in any order
593 } 739 }
594 740
595=cut 741=cut
596 742
597sub run { 743sub run {
598 my ($self, $func, $cb) = @_; 744 my ($self, $func, $cb) = @_;
599 745
600 $self->[0] = $cb; 746 $self->[4] = $cb;
601 $self->_cmd (r => $func); 747 $self->_cmd (r => $func);
602} 748}
749
750=back
751
752=head1 PERFORMANCE
753
754Now for some unscientific benchmark numbers (all done on an amd64
755GNU/Linux box). These are intended to give you an idea of the relative
756performance you can expect, they are not meant to be absolute performance
757numbers.
758
759OK, so, I ran a simple benchmark that creates a socket pair, forks, calls
760exit in the child and waits for the socket to close in the parent. I did
761load AnyEvent, EV and AnyEvent::Fork, for a total process size of 5100kB.
762
763 2079 new processes per second, using manual socketpair + fork
764
765Then I did the same thing, but instead of calling fork, I called
766AnyEvent::Fork->new->run ("CORE::exit") and then again waited for the
767socket form the child to close on exit. This does the same thing as manual
768socket pair + fork, except that what is forked is the template process
769(2440kB), and the socket needs to be passed to the server at the other end
770of the socket first.
771
772 2307 new processes per second, using AnyEvent::Fork->new
773
774And finally, using C<new_exec> instead C<new>, using vforks+execs to exec
775a new perl interpreter and compile the small server each time, I get:
776
777 479 vfork+execs per second, using AnyEvent::Fork->new_exec
778
779So how can C<< AnyEvent->new >> be faster than a standard fork, even
780though it uses the same operations, but adds a lot of overhead?
781
782The difference is simply the process size: forking the 6MB process takes
783so much longer than forking the 2.5MB template process that the overhead
784introduced is canceled out.
785
786If the benchmark process grows, the normal fork becomes even slower:
787
788 1340 new processes, manual fork in a 20MB process
789 731 new processes, manual fork in a 200MB process
790 235 new processes, manual fork in a 2000MB process
791
792What that means (to me) is that I can use this module without having a
793very bad conscience because of the extra overhead required to start new
794processes.
795
796=head1 TYPICAL PROBLEMS
797
798This section lists typical problems that remain. I hope by recognising
799them, most can be avoided.
800
801=over 4
802
803=item "leaked" file descriptors for exec'ed processes
804
805POSIX systems inherit file descriptors by default when exec'ing a new
806process. While perl itself laudably sets the close-on-exec flags on new
807file handles, most C libraries don't care, and even if all cared, it's
808often not possible to set the flag in a race-free manner.
809
810That means some file descriptors can leak through. And since it isn't
811possible to know which file descriptors are "good" and "necessary" (or
812even to know which file descriptors are open), there is no good way to
813close the ones that might harm.
814
815As an example of what "harm" can be done consider a web server that
816accepts connections and afterwards some module uses AnyEvent::Fork for the
817first time, causing it to fork and exec a new process, which might inherit
818the network socket. When the server closes the socket, it is still open
819in the child (which doesn't even know that) and the client might conclude
820that the connection is still fine.
821
822For the main program, there are multiple remedies available -
823L<AnyEvent::Fork::Early> is one, creating a process early and not using
824C<new_exec> is another, as in both cases, the first process can be exec'ed
825well before many random file descriptors are open.
826
827In general, the solution for these kind of problems is to fix the
828libraries or the code that leaks those file descriptors.
829
830Fortunately, most of these leaked descriptors do no harm, other than
831sitting on some resources.
832
833=item "leaked" file descriptors for fork'ed processes
834
835Normally, L<AnyEvent::Fork> does start new processes by exec'ing them,
836which closes file descriptors not marked for being inherited.
837
838However, L<AnyEvent::Fork::Early> and L<AnyEvent::Fork::Template> offer
839a way to create these processes by forking, and this leaks more file
840descriptors than exec'ing them, as there is no way to mark descriptors as
841"close on fork".
842
843An example would be modules like L<EV>, L<IO::AIO> or L<Gtk2>. Both create
844pipes for internal uses, and L<Gtk2> might open a connection to the X
845server. L<EV> and L<IO::AIO> can deal with fork, but Gtk2 might have
846trouble with a fork.
847
848The solution is to either not load these modules before use'ing
849L<AnyEvent::Fork::Early> or L<AnyEvent::Fork::Template>, or to delay
850initialising them, for example, by calling C<init Gtk2> manually.
851
852=item exit runs destructors
853
854This only applies to users of Lc<AnyEvent::Fork:Early> and
855L<AnyEvent::Fork::Template>.
856
857When a process created by AnyEvent::Fork exits, it might do so by calling
858exit, or simply letting perl reach the end of the program. At which point
859Perl runs all destructors.
860
861Not all destructors are fork-safe - for example, an object that represents
862the connection to an X display might tell the X server to free resources,
863which is inconvenient when the "real" object in the parent still needs to
864use them.
865
866This is obviously not a problem for L<AnyEvent::Fork::Early>, as you used
867it as the very first thing, right?
868
869It is a problem for L<AnyEvent::Fork::Template> though - and the solution
870is to not create objects with nontrivial destructors that might have an
871effect outside of Perl.
603 872
604=back 873=back
605 874
606=head1 PORTABILITY NOTES 875=head1 PORTABILITY NOTES
607 876
608Native win32 perls are somewhat supported (AnyEvent::Fork::Early is a nop, 877Native win32 perls are somewhat supported (AnyEvent::Fork::Early is a nop,
609and ::Template is not going to work), and it cost a lot of blood and sweat 878and ::Template is not going to work), and it cost a lot of blood and sweat
610to make it so, mostly due to the bloody broken perl that nobody seems to 879to make it so, mostly due to the bloody broken perl that nobody seems to
611care about. The fork emulation is a bad joke - I have yet to see something 880care about. The fork emulation is a bad joke - I have yet to see something
612useful that you cna do with it without running into memory corruption 881useful that you can do with it without running into memory corruption
613issues or other braindamage. Hrrrr. 882issues or other braindamage. Hrrrr.
614 883
615Cygwin perl is not supported at the moment, as it should implement fd 884Cygwin perl is not supported at the moment, as it should implement fd
616passing, but doesn't, and rolling my own is hard, as cygwin doesn't 885passing, but doesn't, and rolling my own is hard, as cygwin doesn't
617support enough functionality to do it. 886support enough functionality to do it.
618 887
888=head1 SEE ALSO
889
890L<AnyEvent::Fork::Early> (to avoid executing a perl interpreter),
891L<AnyEvent::Fork::Template> (to create a process by forking the main
892program at a convenient time).
893
619=head1 AUTHOR 894=head1 AUTHOR
620 895
621 Marc Lehmann <schmorp@schmorp.de> 896 Marc Lehmann <schmorp@schmorp.de>
622 http://home.schmorp.de/ 897 http://home.schmorp.de/
623 898

Diff Legend

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