1 |
=head1 NAME |
2 |
|
3 |
AnyEvent::Fork - everything you wanted to use fork() for, but couldn't |
4 |
|
5 |
=head1 SYNOPSIS |
6 |
|
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; |
15 |
|
16 |
=head1 DESCRIPTION |
17 |
|
18 |
This module allows you to create new processes, without actually forking |
19 |
them from your current process (avoiding the problems of forking), but |
20 |
preserving most of the advantages of fork. |
21 |
|
22 |
It can be used to create new worker processes or new independent |
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. |
27 |
|
28 |
Special care has been taken to make this module useful from other modules, |
29 |
while still supporting specialised environments such as L<App::Staticperl> |
30 |
or L<PAR::Packer>. |
31 |
|
32 |
=head1 WHAT THIS MODULE IS NOT |
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 can either implement it yourself |
40 |
in whatever way you like, use some message-passing module such |
41 |
as L<AnyEvent::MP>, some pipe such as L<AnyEvent::ZeroMQ>, use |
42 |
L<AnyEvent::Handle> on both sides to send e.g. JSON or Storable messages, |
43 |
and so on. |
44 |
|
45 |
=head1 PROBLEM STATEMENT |
46 |
|
47 |
There are two traditional ways to implement parallel processing on UNIX |
48 |
like operating systems - fork and process, and fork+exec and process. They |
49 |
have different advantages and disadvantages that I describe below, |
50 |
together 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 |
|
56 |
A 5GB process needs 0.05s to fork on my 3.6GHz amd64 GNU/Linux box. This |
57 |
overhead is often shared with exec (because you have to fork first), but |
58 |
in some circumstances (e.g. when vfork is used), fork+exec can be much |
59 |
faster. |
60 |
|
61 |
This module can help here by telling a small(er) helper process to fork, |
62 |
which is faster then forking the main process, and also uses vfork where |
63 |
possible. 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 |
66 |
process. |
67 |
|
68 |
For example, modules or data files that are loaded will not use additional |
69 |
memory after a fork. When exec'ing a new process, modules and data files |
70 |
might need to be loaded again, at extra CPU and memory cost. But when |
71 |
forking, literally all data structures are copied - if the program frees |
72 |
them and replaces them by new data, the child processes will retain the |
73 |
old version even if it isn't used, which can suddenly and unexpectedly |
74 |
increase memory usage when freeing memory. |
75 |
|
76 |
The trade-off is between more sharing with fork (which can be good or |
77 |
bad), and no sharing with exec. |
78 |
|
79 |
This module allows the main program to do a controlled fork, and allows |
80 |
modules to exec processes safely at any time. When creating a custom |
81 |
process pool you can take advantage of data sharing via fork without |
82 |
risking to share large dynamic data structures that will blow up child |
83 |
memory usage. |
84 |
|
85 |
In other words, this module puts you into control over what is being |
86 |
shared and what isn't, at all times. |
87 |
|
88 |
=item Exec'ing a new perl process might be difficult. |
89 |
|
90 |
For example, it is not easy to find the correct path to the perl |
91 |
interpreter - C<$^X> might not be a perl interpreter at all. |
92 |
|
93 |
This module tries hard to identify the correct path to the perl |
94 |
interpreter. With a cooperative main program, exec'ing the interpreter |
95 |
might not even be necessary, but even without help from the main program, |
96 |
it will still work when used from a module. |
97 |
|
98 |
=item Exec'ing a new perl process might be slow, as all necessary modules |
99 |
have to be loaded from disk again, with no guarantees of success. |
100 |
|
101 |
Long running processes might run into problems when perl is upgraded |
102 |
and modules are no longer loadable because they refer to a different |
103 |
perl version, or parts of a distribution are newer than the ones already |
104 |
loaded. |
105 |
|
106 |
This module supports creating pre-initialised perl processes to be used as |
107 |
a template for new processes. |
108 |
|
109 |
=item Forking might be impossible when a program is running. |
110 |
|
111 |
For example, POSIX makes it almost impossible to fork from a |
112 |
multi-threaded program while doing anything useful in the child - in |
113 |
fact, if your perl program uses POSIX threads (even indirectly via |
114 |
e.g. L<IO::AIO> or L<threads>), you cannot call fork on the perl level |
115 |
anymore without risking corruption issues on a number of operating |
116 |
systems. |
117 |
|
118 |
This module can safely fork helper processes at any time, by calling |
119 |
fork+exec in C, in a POSIX-compatible way (via L<Proc::FastSpawn>). |
120 |
|
121 |
=item Parallel processing with fork might be inconvenient or difficult |
122 |
to implement. Modules might not work in both parent and child. |
123 |
|
124 |
For example, when a program uses an event loop and creates watchers it |
125 |
becomes very hard to use the event loop from a child program, as the |
126 |
watchers already exist but are only meaningful in the parent. Worse, a |
127 |
module might want to use such a module, not knowing whether another module |
128 |
or the main program also does, leading to problems. |
129 |
|
130 |
Apart from event loops, graphical toolkits also commonly fall into the |
131 |
"unsafe module" category, or just about anything that communicates with |
132 |
the external world, such as network libraries and file I/O modules, which |
133 |
usually don't like being copied and then allowed to continue in two |
134 |
processes. |
135 |
|
136 |
With this module only the main program is allowed to create new processes |
137 |
by forking (because only the main program can know when it is still safe |
138 |
to do so) - all other processes are created via fork+exec, which makes it |
139 |
possible to use modules such as event loops or window interfaces safely. |
140 |
|
141 |
=back |
142 |
|
143 |
=head1 EXAMPLES |
144 |
|
145 |
=head2 Create a single new process, tell it to run your worker function. |
146 |
|
147 |
AnyEvent::Fork |
148 |
->new |
149 |
->require ("MyModule") |
150 |
->run ("MyModule::worker, sub { |
151 |
my ($master_filehandle) = @_; |
152 |
|
153 |
# now $master_filehandle is connected to the |
154 |
# $slave_filehandle in the new process. |
155 |
}); |
156 |
|
157 |
C<MyModule> might look like this: |
158 |
|
159 |
package MyModule; |
160 |
|
161 |
sub worker { |
162 |
my ($slave_filehandle) = @_; |
163 |
|
164 |
# now $slave_filehandle is connected to the $master_filehandle |
165 |
# in the original prorcess. have fun! |
166 |
} |
167 |
|
168 |
=head2 Create a pool of server processes all accepting on the same socket. |
169 |
|
170 |
# create listener socket |
171 |
my $listener = ...; |
172 |
|
173 |
# create a pool template, initialise it and give it the socket |
174 |
my $pool = AnyEvent::Fork |
175 |
->new |
176 |
->require ("Some::Stuff", "My::Server") |
177 |
->send_fh ($listener); |
178 |
|
179 |
# now create 10 identical workers |
180 |
for my $id (1..10) { |
181 |
$pool |
182 |
->fork |
183 |
->send_arg ($id) |
184 |
->run ("My::Server::run"); |
185 |
} |
186 |
|
187 |
# now do other things - maybe use the filehandle provided by run |
188 |
# to wait for the processes to die. or whatever. |
189 |
|
190 |
C<My::Server> might look like this: |
191 |
|
192 |
package My::Server; |
193 |
|
194 |
sub run { |
195 |
my ($slave, $listener, $id) = @_; |
196 |
|
197 |
close $slave; # we do not use the socket, so close it to save resources |
198 |
|
199 |
# we could go ballistic and use e.g. AnyEvent here, or IO::AIO, |
200 |
# or anything we usually couldn't do in a process forked normally. |
201 |
while (my $socket = $listener->accept) { |
202 |
# do sth. with new socket |
203 |
} |
204 |
} |
205 |
|
206 |
=head2 use AnyEvent::Fork as a faster fork+exec |
207 |
|
208 |
This runs C</bin/echo hi>, with stdandard output redirected to /tmp/log |
209 |
and standard error redirected to the communications socket. It is usually |
210 |
faster than fork+exec, but still lets you prepare the environment. |
211 |
|
212 |
open my $output, ">/tmp/log" or die "$!"; |
213 |
|
214 |
AnyEvent::Fork |
215 |
->new |
216 |
->eval (' |
217 |
sub run { |
218 |
my ($fh, $output, @cmd) = @_; |
219 |
|
220 |
# perl will clear close-on-exec on STDOUT/STDERR |
221 |
open STDOUT, ">&", $output or die; |
222 |
open STDERR, ">&", $fh or die; |
223 |
|
224 |
exec @cmd; |
225 |
} |
226 |
') |
227 |
->send_fh ($output) |
228 |
->send_arg ("/bin/echo", "hi") |
229 |
->run ("run", my $cv = AE::cv); |
230 |
|
231 |
my $stderr = $cv->recv; |
232 |
|
233 |
=head1 CONCEPTS |
234 |
|
235 |
This module can create new processes either by executing a new perl |
236 |
process, or by forking from an existing "template" process. |
237 |
|
238 |
Each such process comes with its own file handle that can be used to |
239 |
communicate with it (it's actually a socket - one end in the new process, |
240 |
one end in the main process), and among the things you can do in it are |
241 |
load modules, fork new processes, send file handles to it, and execute |
242 |
functions. |
243 |
|
244 |
There are multiple ways to create additional processes to execute some |
245 |
jobs: |
246 |
|
247 |
=over 4 |
248 |
|
249 |
=item fork a new process from the "default" template process, load code, |
250 |
run it |
251 |
|
252 |
This module has a "default" template process which it executes when it is |
253 |
needed the first time. Forking from this process shares the memory used |
254 |
for the perl interpreter with the new process, but loading modules takes |
255 |
time, and the memory is not shared with anything else. |
256 |
|
257 |
This is ideal for when you only need one extra process of a kind, with the |
258 |
option of starting and stopping it on demand. |
259 |
|
260 |
Example: |
261 |
|
262 |
AnyEvent::Fork |
263 |
->new |
264 |
->require ("Some::Module") |
265 |
->run ("Some::Module::run", sub { |
266 |
my ($fork_fh) = @_; |
267 |
}); |
268 |
|
269 |
=item fork a new template process, load code, then fork processes off of |
270 |
it and run the code |
271 |
|
272 |
When you need to have a bunch of processes that all execute the same (or |
273 |
very similar) tasks, then a good way is to create a new template process |
274 |
for them, loading all the modules you need, and then create your worker |
275 |
processes from this new template process. |
276 |
|
277 |
This way, all code (and data structures) that can be shared (e.g. the |
278 |
modules you loaded) is shared between the processes, and each new process |
279 |
consumes relatively little memory of its own. |
280 |
|
281 |
The disadvantage of this approach is that you need to create a template |
282 |
process for the sole purpose of forking new processes from it, but if you |
283 |
only need a fixed number of processes you can create them, and then destroy |
284 |
the template process. |
285 |
|
286 |
Example: |
287 |
|
288 |
my $template = AnyEvent::Fork->new->require ("Some::Module"); |
289 |
|
290 |
for (1..10) { |
291 |
$template->fork->run ("Some::Module::run", sub { |
292 |
my ($fork_fh) = @_; |
293 |
}); |
294 |
} |
295 |
|
296 |
# at this point, you can keep $template around to fork new processes |
297 |
# later, or you can destroy it, which causes it to vanish. |
298 |
|
299 |
=item execute a new perl interpreter, load some code, run it |
300 |
|
301 |
This is relatively slow, and doesn't allow you to share memory between |
302 |
multiple processes. |
303 |
|
304 |
The only advantage is that you don't have to have a template process |
305 |
hanging around all the time to fork off some new processes, which might be |
306 |
an advantage when there are long time spans where no extra processes are |
307 |
needed. |
308 |
|
309 |
Example: |
310 |
|
311 |
AnyEvent::Fork |
312 |
->new_exec |
313 |
->require ("Some::Module") |
314 |
->run ("Some::Module::run", sub { |
315 |
my ($fork_fh) = @_; |
316 |
}); |
317 |
|
318 |
=back |
319 |
|
320 |
=head1 THE C<AnyEvent::Fork> CLASS |
321 |
|
322 |
This module exports nothing, and only implements a single class - |
323 |
C<AnyEvent::Fork>. |
324 |
|
325 |
There are two class constructors that both create new processes - C<new> |
326 |
and C<new_exec>. The C<fork> method creates a new process by forking an |
327 |
existing one and could be considered a third constructor. |
328 |
|
329 |
Most of the remaining methods deal with preparing the new process, by |
330 |
loading code, evaluating code and sending data to the new process. They |
331 |
usually return the process object, so you can chain method calls. |
332 |
|
333 |
If a process object is destroyed before calling its C<run> method, then |
334 |
the process simply exits. After C<run> is called, all responsibility is |
335 |
passed to the specified function. |
336 |
|
337 |
As long as there is any outstanding work to be done, process objects |
338 |
resist being destroyed, so there is no reason to store them unless you |
339 |
need them later - configure and forget works just fine. |
340 |
|
341 |
=over 4 |
342 |
|
343 |
=cut |
344 |
|
345 |
package AnyEvent::Fork; |
346 |
|
347 |
use common::sense; |
348 |
|
349 |
use Errno (); |
350 |
|
351 |
use AnyEvent; |
352 |
use AnyEvent::Util (); |
353 |
|
354 |
use IO::FDPass; |
355 |
|
356 |
our $VERSION = 0.5; |
357 |
|
358 |
our $PERL; # the path to the perl interpreter, deduces with various forms of magic |
359 |
|
360 |
=over 4 |
361 |
|
362 |
=back |
363 |
|
364 |
=cut |
365 |
|
366 |
# the early fork template process |
367 |
our $EARLY; |
368 |
|
369 |
# the empty template process |
370 |
our $TEMPLATE; |
371 |
|
372 |
sub _cmd { |
373 |
my $self = shift; |
374 |
|
375 |
# ideally, we would want to use "a (w/a)*" as format string, but perl |
376 |
# versions from at least 5.8.9 to 5.16.3 are all buggy and can't unpack |
377 |
# it. |
378 |
push @{ $self->[2] }, pack "a L/a*", $_[0], $_[1]; |
379 |
|
380 |
$self->[3] ||= AE::io $self->[1], 1, sub { |
381 |
do { |
382 |
# send the next "thing" in the queue - either a reference to an fh, |
383 |
# or a plain string. |
384 |
|
385 |
if (ref $self->[2][0]) { |
386 |
# send fh |
387 |
unless (IO::FDPass::send fileno $self->[1], fileno ${ $self->[2][0] }) { |
388 |
return if $! == Errno::EAGAIN || $! == Errno::EWOULDBLOCK; |
389 |
undef $self->[3]; |
390 |
die "AnyEvent::Fork: file descriptor send failure: $!"; |
391 |
} |
392 |
|
393 |
shift @{ $self->[2] }; |
394 |
|
395 |
} else { |
396 |
# send string |
397 |
my $len = syswrite $self->[1], $self->[2][0]; |
398 |
|
399 |
unless ($len) { |
400 |
return if $! == Errno::EAGAIN || $! == Errno::EWOULDBLOCK; |
401 |
undef $self->[3]; |
402 |
die "AnyEvent::Fork: command write failure: $!"; |
403 |
} |
404 |
|
405 |
substr $self->[2][0], 0, $len, ""; |
406 |
shift @{ $self->[2] } unless length $self->[2][0]; |
407 |
} |
408 |
} while @{ $self->[2] }; |
409 |
|
410 |
# everything written |
411 |
undef $self->[3]; |
412 |
|
413 |
# invoke run callback, if any |
414 |
$self->[4]->($self->[1]) if $self->[4]; |
415 |
}; |
416 |
|
417 |
() # make sure we don't leak the watcher |
418 |
} |
419 |
|
420 |
sub _new { |
421 |
my ($self, $fh, $pid) = @_; |
422 |
|
423 |
AnyEvent::Util::fh_nonblocking $fh, 1; |
424 |
|
425 |
$self = bless [ |
426 |
$pid, |
427 |
$fh, |
428 |
[], # write queue - strings or fd's |
429 |
undef, # AE watcher |
430 |
], $self; |
431 |
|
432 |
$self |
433 |
} |
434 |
|
435 |
# fork template from current process, used by AnyEvent::Fork::Early/Template |
436 |
sub _new_fork { |
437 |
my ($fh, $slave) = AnyEvent::Util::portable_socketpair; |
438 |
my $parent = $$; |
439 |
|
440 |
my $pid = fork; |
441 |
|
442 |
if ($pid eq 0) { |
443 |
require AnyEvent::Fork::Serve; |
444 |
$AnyEvent::Fork::Serve::OWNER = $parent; |
445 |
close $fh; |
446 |
$0 = "$_[1] of $parent"; |
447 |
$SIG{CHLD} = 'IGNORE'; |
448 |
AnyEvent::Fork::Serve::serve ($slave); |
449 |
exit 0; |
450 |
} elsif (!$pid) { |
451 |
die "AnyEvent::Fork::Early/Template: unable to fork template process: $!"; |
452 |
} |
453 |
|
454 |
AnyEvent::Fork->_new ($fh, $pid) |
455 |
} |
456 |
|
457 |
=item my $proc = new AnyEvent::Fork |
458 |
|
459 |
Create a new "empty" perl interpreter process and returns its process |
460 |
object for further manipulation. |
461 |
|
462 |
The new process is forked from a template process that is kept around |
463 |
for this purpose. When it doesn't exist yet, it is created by a call to |
464 |
C<new_exec> first and then stays around for future calls. |
465 |
|
466 |
=cut |
467 |
|
468 |
sub new { |
469 |
my $class = shift; |
470 |
|
471 |
$TEMPLATE ||= $class->new_exec; |
472 |
$TEMPLATE->fork |
473 |
} |
474 |
|
475 |
=item $new_proc = $proc->fork |
476 |
|
477 |
Forks C<$proc>, creating a new process, and returns the process object |
478 |
of the new process. |
479 |
|
480 |
If any of the C<send_> functions have been called before fork, then they |
481 |
will be cloned in the child. For example, in a pre-forked server, you |
482 |
might C<send_fh> the listening socket into the template process, and then |
483 |
keep calling C<fork> and C<run>. |
484 |
|
485 |
=cut |
486 |
|
487 |
sub fork { |
488 |
my ($self) = @_; |
489 |
|
490 |
my ($fh, $slave) = AnyEvent::Util::portable_socketpair; |
491 |
|
492 |
$self->send_fh ($slave); |
493 |
$self->_cmd ("f"); |
494 |
|
495 |
AnyEvent::Fork->_new ($fh) |
496 |
} |
497 |
|
498 |
=item my $proc = new_exec AnyEvent::Fork |
499 |
|
500 |
Create a new "empty" perl interpreter process and returns its process |
501 |
object for further manipulation. |
502 |
|
503 |
Unlike the C<new> method, this method I<always> spawns a new perl process |
504 |
(except in some cases, see L<AnyEvent::Fork::Early> for details). This |
505 |
reduces the amount of memory sharing that is possible, and is also slower. |
506 |
|
507 |
You should use C<new> whenever possible, except when having a template |
508 |
process around is unacceptable. |
509 |
|
510 |
The path to the perl interpreter is divined using various methods - first |
511 |
C<$^X> is investigated to see if the path ends with something that sounds |
512 |
as if it were the perl interpreter. Failing this, the module falls back to |
513 |
using C<$Config::Config{perlpath}>. |
514 |
|
515 |
=cut |
516 |
|
517 |
sub new_exec { |
518 |
my ($self) = @_; |
519 |
|
520 |
return $EARLY->fork |
521 |
if $EARLY; |
522 |
|
523 |
# first find path of perl |
524 |
my $perl = $; |
525 |
|
526 |
# first we try $^X, but the path must be absolute (always on win32), and end in sth. |
527 |
# that looks like perl. this obviously only works for posix and win32 |
528 |
unless ( |
529 |
($^O eq "MSWin32" || $perl =~ m%^/%) |
530 |
&& $perl =~ m%[/\\]perl(?:[0-9]+(\.[0-9]+)+)?(\.exe)?$%i |
531 |
) { |
532 |
# if it doesn't look perlish enough, try Config |
533 |
require Config; |
534 |
$perl = $Config::Config{perlpath}; |
535 |
$perl =~ s/(?:\Q$Config::Config{_exe}\E)?$/$Config::Config{_exe}/; |
536 |
} |
537 |
|
538 |
require Proc::FastSpawn; |
539 |
|
540 |
my ($fh, $slave) = AnyEvent::Util::portable_socketpair; |
541 |
Proc::FastSpawn::fd_inherit (fileno $slave); |
542 |
|
543 |
# new fh's should always be set cloexec (due to $^F), |
544 |
# but hey, not on win32, so we always clear the inherit flag. |
545 |
Proc::FastSpawn::fd_inherit (fileno $fh, 0); |
546 |
|
547 |
# quick. also doesn't work in win32. of course. what did you expect |
548 |
#local $ENV{PERL5LIB} = join ":", grep !ref, @INC; |
549 |
my %env = %ENV; |
550 |
$env{PERL5LIB} = join +($^O eq "MSWin32" ? ";" : ":"), grep !ref, @INC; |
551 |
|
552 |
my $pid = Proc::FastSpawn::spawn ( |
553 |
$perl, |
554 |
["perl", "-MAnyEvent::Fork::Serve", "-e", "AnyEvent::Fork::Serve::me", fileno $slave, $$], |
555 |
[map "$_=$env{$_}", keys %env], |
556 |
) or die "unable to spawn AnyEvent::Fork server: $!"; |
557 |
|
558 |
$self->_new ($fh, $pid) |
559 |
} |
560 |
|
561 |
=item $pid = $proc->pid |
562 |
|
563 |
Returns the process id of the process I<iff it is a direct child of the |
564 |
process running AnyEvent::Fork>, and C<undef> otherwise. |
565 |
|
566 |
Normally, only processes created via C<< AnyEvent::Fork->new_exec >> and |
567 |
L<AnyEvent::Fork::Template> are direct children, and you are responsible |
568 |
to clean up their zombies when they die. |
569 |
|
570 |
All other processes are not direct children, and will be cleaned up by |
571 |
AnyEvent::Fork itself. |
572 |
|
573 |
=cut |
574 |
|
575 |
sub pid { |
576 |
$_[0][0] |
577 |
} |
578 |
|
579 |
=item $proc = $proc->eval ($perlcode, @args) |
580 |
|
581 |
Evaluates the given C<$perlcode> as ... perl code, while setting C<@_> to |
582 |
the strings specified by C<@args>, in the "main" package. |
583 |
|
584 |
This call is meant to do any custom initialisation that might be required |
585 |
(for example, the C<require> method uses it). It's not supposed to be used |
586 |
to completely take over the process, use C<run> for that. |
587 |
|
588 |
The code will usually be executed after this call returns, and there is no |
589 |
way to pass anything back to the calling process. Any evaluation errors |
590 |
will be reported to stderr and cause the process to exit. |
591 |
|
592 |
If you want to execute some code (that isn't in a module) to take over the |
593 |
process, you should compile a function via C<eval> first, and then call |
594 |
it via C<run>. This also gives you access to any arguments passed via the |
595 |
C<send_xxx> methods, such as file handles. See the L<use AnyEvent::Fork as |
596 |
a faster fork+exec> example to see it in action. |
597 |
|
598 |
Returns the process object for easy chaining of method calls. |
599 |
|
600 |
=cut |
601 |
|
602 |
sub eval { |
603 |
my ($self, $code, @args) = @_; |
604 |
|
605 |
$self->_cmd (e => pack "(w/a*)*", $code, @args); |
606 |
|
607 |
$self |
608 |
} |
609 |
|
610 |
=item $proc = $proc->require ($module, ...) |
611 |
|
612 |
Tries to load the given module(s) into the process |
613 |
|
614 |
Returns the process object for easy chaining of method calls. |
615 |
|
616 |
=cut |
617 |
|
618 |
sub require { |
619 |
my ($self, @modules) = @_; |
620 |
|
621 |
s%::%/%g for @modules; |
622 |
$self->eval ('require "$_.pm" for @_', @modules); |
623 |
|
624 |
$self |
625 |
} |
626 |
|
627 |
=item $proc = $proc->send_fh ($handle, ...) |
628 |
|
629 |
Send one or more file handles (I<not> file descriptors) to the process, |
630 |
to prepare a call to C<run>. |
631 |
|
632 |
The process object keeps a reference to the handles until they have |
633 |
been passed over to the process, so you must not explicitly close the |
634 |
handles. This is most easily accomplished by simply not storing the file |
635 |
handles anywhere after passing them to this method - when AnyEvent::Fork |
636 |
is finished using them, perl will automatically close them. |
637 |
|
638 |
Returns the process object for easy chaining of method calls. |
639 |
|
640 |
Example: pass a file handle to a process, and release it without |
641 |
closing. It will be closed automatically when it is no longer used. |
642 |
|
643 |
$proc->send_fh ($my_fh); |
644 |
undef $my_fh; # free the reference if you want, but DO NOT CLOSE IT |
645 |
|
646 |
=cut |
647 |
|
648 |
sub send_fh { |
649 |
my ($self, @fh) = @_; |
650 |
|
651 |
for my $fh (@fh) { |
652 |
$self->_cmd ("h"); |
653 |
push @{ $self->[2] }, \$fh; |
654 |
} |
655 |
|
656 |
$self |
657 |
} |
658 |
|
659 |
=item $proc = $proc->send_arg ($string, ...) |
660 |
|
661 |
Send one or more argument strings to the process, to prepare a call to |
662 |
C<run>. The strings can be any octet strings. |
663 |
|
664 |
The protocol is optimised to pass a moderate number of relatively short |
665 |
strings - while you can pass up to 4GB of data in one go, this is more |
666 |
meant to pass some ID information or other startup info, not big chunks of |
667 |
data. |
668 |
|
669 |
Returns the process object for easy chaining of method calls. |
670 |
|
671 |
=cut |
672 |
|
673 |
sub send_arg { |
674 |
my ($self, @arg) = @_; |
675 |
|
676 |
$self->_cmd (a => pack "(w/a*)*", @arg); |
677 |
|
678 |
$self |
679 |
} |
680 |
|
681 |
=item $proc->run ($func, $cb->($fh)) |
682 |
|
683 |
Enter the function specified by the function name in C<$func> in the |
684 |
process. The function is called with the communication socket as first |
685 |
argument, followed by all file handles and string arguments sent earlier |
686 |
via C<send_fh> and C<send_arg> methods, in the order they were called. |
687 |
|
688 |
The process object becomes unusable on return from this function - any |
689 |
further method calls result in undefined behaviour. |
690 |
|
691 |
The function name should be fully qualified, but if it isn't, it will be |
692 |
looked up in the C<main> package. |
693 |
|
694 |
If the called function returns, doesn't exist, or any error occurs, the |
695 |
process exits. |
696 |
|
697 |
Preparing the process is done in the background - when all commands have |
698 |
been sent, the callback is invoked with the local communications socket |
699 |
as argument. At this point you can start using the socket in any way you |
700 |
like. |
701 |
|
702 |
If the communication socket isn't used, it should be closed on both sides, |
703 |
to save on kernel memory. |
704 |
|
705 |
The socket is non-blocking in the parent, and blocking in the newly |
706 |
created process. The close-on-exec flag is set in both. |
707 |
|
708 |
Even if not used otherwise, the socket can be a good indicator for the |
709 |
existence of the process - if the other process exits, you get a readable |
710 |
event on it, because exiting the process closes the socket (if it didn't |
711 |
create any children using fork). |
712 |
|
713 |
Example: create a template for a process pool, pass a few strings, some |
714 |
file handles, then fork, pass one more string, and run some code. |
715 |
|
716 |
my $pool = AnyEvent::Fork |
717 |
->new |
718 |
->send_arg ("str1", "str2") |
719 |
->send_fh ($fh1, $fh2); |
720 |
|
721 |
for (1..2) { |
722 |
$pool |
723 |
->fork |
724 |
->send_arg ("str3") |
725 |
->run ("Some::function", sub { |
726 |
my ($fh) = @_; |
727 |
|
728 |
# fh is nonblocking, but we trust that the OS can accept these |
729 |
# few octets anyway. |
730 |
syswrite $fh, "hi #$_\n"; |
731 |
|
732 |
# $fh is being closed here, as we don't store it anywhere |
733 |
}); |
734 |
} |
735 |
|
736 |
# Some::function might look like this - all parameters passed before fork |
737 |
# and after will be passed, in order, after the communications socket. |
738 |
sub Some::function { |
739 |
my ($fh, $str1, $str2, $fh1, $fh2, $str3) = @_; |
740 |
|
741 |
print scalar <$fh>; # prints "hi #1\n" and "hi #2\n" in any order |
742 |
} |
743 |
|
744 |
=cut |
745 |
|
746 |
sub run { |
747 |
my ($self, $func, $cb) = @_; |
748 |
|
749 |
$self->[4] = $cb; |
750 |
$self->_cmd (r => $func); |
751 |
} |
752 |
|
753 |
=back |
754 |
|
755 |
=head1 PERFORMANCE |
756 |
|
757 |
Now for some unscientific benchmark numbers (all done on an amd64 |
758 |
GNU/Linux box). These are intended to give you an idea of the relative |
759 |
performance you can expect, they are not meant to be absolute performance |
760 |
numbers. |
761 |
|
762 |
OK, so, I ran a simple benchmark that creates a socket pair, forks, calls |
763 |
exit in the child and waits for the socket to close in the parent. I did |
764 |
load AnyEvent, EV and AnyEvent::Fork, for a total process size of 5100kB. |
765 |
|
766 |
2079 new processes per second, using manual socketpair + fork |
767 |
|
768 |
Then I did the same thing, but instead of calling fork, I called |
769 |
AnyEvent::Fork->new->run ("CORE::exit") and then again waited for the |
770 |
socket form the child to close on exit. This does the same thing as manual |
771 |
socket pair + fork, except that what is forked is the template process |
772 |
(2440kB), and the socket needs to be passed to the server at the other end |
773 |
of the socket first. |
774 |
|
775 |
2307 new processes per second, using AnyEvent::Fork->new |
776 |
|
777 |
And finally, using C<new_exec> instead C<new>, using vforks+execs to exec |
778 |
a new perl interpreter and compile the small server each time, I get: |
779 |
|
780 |
479 vfork+execs per second, using AnyEvent::Fork->new_exec |
781 |
|
782 |
So how can C<< AnyEvent->new >> be faster than a standard fork, even |
783 |
though it uses the same operations, but adds a lot of overhead? |
784 |
|
785 |
The difference is simply the process size: forking the 6MB process takes |
786 |
so much longer than forking the 2.5MB template process that the overhead |
787 |
introduced is canceled out. |
788 |
|
789 |
If the benchmark process grows, the normal fork becomes even slower: |
790 |
|
791 |
1340 new processes, manual fork in a 20MB process |
792 |
731 new processes, manual fork in a 200MB process |
793 |
235 new processes, manual fork in a 2000MB process |
794 |
|
795 |
What that means (to me) is that I can use this module without having a |
796 |
very bad conscience because of the extra overhead required to start new |
797 |
processes. |
798 |
|
799 |
=head1 TYPICAL PROBLEMS |
800 |
|
801 |
This section lists typical problems that remain. I hope by recognising |
802 |
them, most can be avoided. |
803 |
|
804 |
=over 4 |
805 |
|
806 |
=item "leaked" file descriptors for exec'ed processes |
807 |
|
808 |
POSIX systems inherit file descriptors by default when exec'ing a new |
809 |
process. While perl itself laudably sets the close-on-exec flags on new |
810 |
file handles, most C libraries don't care, and even if all cared, it's |
811 |
often not possible to set the flag in a race-free manner. |
812 |
|
813 |
That means some file descriptors can leak through. And since it isn't |
814 |
possible to know which file descriptors are "good" and "necessary" (or |
815 |
even to know which file descriptors are open), there is no good way to |
816 |
close the ones that might harm. |
817 |
|
818 |
As an example of what "harm" can be done consider a web server that |
819 |
accepts connections and afterwards some module uses AnyEvent::Fork for the |
820 |
first time, causing it to fork and exec a new process, which might inherit |
821 |
the network socket. When the server closes the socket, it is still open |
822 |
in the child (which doesn't even know that) and the client might conclude |
823 |
that the connection is still fine. |
824 |
|
825 |
For the main program, there are multiple remedies available - |
826 |
L<AnyEvent::Fork::Early> is one, creating a process early and not using |
827 |
C<new_exec> is another, as in both cases, the first process can be exec'ed |
828 |
well before many random file descriptors are open. |
829 |
|
830 |
In general, the solution for these kind of problems is to fix the |
831 |
libraries or the code that leaks those file descriptors. |
832 |
|
833 |
Fortunately, most of these leaked descriptors do no harm, other than |
834 |
sitting on some resources. |
835 |
|
836 |
=item "leaked" file descriptors for fork'ed processes |
837 |
|
838 |
Normally, L<AnyEvent::Fork> does start new processes by exec'ing them, |
839 |
which closes file descriptors not marked for being inherited. |
840 |
|
841 |
However, L<AnyEvent::Fork::Early> and L<AnyEvent::Fork::Template> offer |
842 |
a way to create these processes by forking, and this leaks more file |
843 |
descriptors than exec'ing them, as there is no way to mark descriptors as |
844 |
"close on fork". |
845 |
|
846 |
An example would be modules like L<EV>, L<IO::AIO> or L<Gtk2>. Both create |
847 |
pipes for internal uses, and L<Gtk2> might open a connection to the X |
848 |
server. L<EV> and L<IO::AIO> can deal with fork, but Gtk2 might have |
849 |
trouble with a fork. |
850 |
|
851 |
The solution is to either not load these modules before use'ing |
852 |
L<AnyEvent::Fork::Early> or L<AnyEvent::Fork::Template>, or to delay |
853 |
initialising them, for example, by calling C<init Gtk2> manually. |
854 |
|
855 |
=item exit runs destructors |
856 |
|
857 |
This only applies to users of Lc<AnyEvent::Fork:Early> and |
858 |
L<AnyEvent::Fork::Template>. |
859 |
|
860 |
When a process created by AnyEvent::Fork exits, it might do so by calling |
861 |
exit, or simply letting perl reach the end of the program. At which point |
862 |
Perl runs all destructors. |
863 |
|
864 |
Not all destructors are fork-safe - for example, an object that represents |
865 |
the connection to an X display might tell the X server to free resources, |
866 |
which is inconvenient when the "real" object in the parent still needs to |
867 |
use them. |
868 |
|
869 |
This is obviously not a problem for L<AnyEvent::Fork::Early>, as you used |
870 |
it as the very first thing, right? |
871 |
|
872 |
It is a problem for L<AnyEvent::Fork::Template> though - and the solution |
873 |
is to not create objects with nontrivial destructors that might have an |
874 |
effect outside of Perl. |
875 |
|
876 |
=back |
877 |
|
878 |
=head1 PORTABILITY NOTES |
879 |
|
880 |
Native win32 perls are somewhat supported (AnyEvent::Fork::Early is a nop, |
881 |
and ::Template is not going to work), and it cost a lot of blood and sweat |
882 |
to make it so, mostly due to the bloody broken perl that nobody seems to |
883 |
care about. The fork emulation is a bad joke - I have yet to see something |
884 |
useful that you can do with it without running into memory corruption |
885 |
issues or other braindamage. Hrrrr. |
886 |
|
887 |
Cygwin perl is not supported at the moment, as it should implement fd |
888 |
passing, but doesn't, and rolling my own is hard, as cygwin doesn't |
889 |
support enough functionality to do it. |
890 |
|
891 |
=head1 SEE ALSO |
892 |
|
893 |
L<AnyEvent::Fork::Early> (to avoid executing a perl interpreter), |
894 |
L<AnyEvent::Fork::Template> (to create a process by forking the main |
895 |
program at a convenient time). |
896 |
|
897 |
=head1 AUTHOR |
898 |
|
899 |
Marc Lehmann <schmorp@schmorp.de> |
900 |
http://home.schmorp.de/ |
901 |
|
902 |
=cut |
903 |
|
904 |
1 |
905 |
|