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 |
=head1 DESCRIPTION |
10 |
|
11 |
This module allows you to create new processes, without actually forking |
12 |
them from your current process (avoiding the problems of forking), but |
13 |
preserving most of the advantages of fork. |
14 |
|
15 |
It can be used to create new worker processes or new independent |
16 |
subprocesses for short- and long-running jobs, process pools (e.g. for use |
17 |
in pre-forked servers) but also to spawn new external processes (such as |
18 |
CGI scripts from a webserver), which can be faster (and more well behaved) |
19 |
than using fork+exec in big processes. |
20 |
|
21 |
Special care has been taken to make this module useful from other modules, |
22 |
while still supporting specialised environments such as L<App::Staticperl> |
23 |
or L<PAR::Packer>. |
24 |
|
25 |
=head1 PROBLEM STATEMENT |
26 |
|
27 |
There are two ways to implement parallel processing on UNIX like operating |
28 |
systems - fork and process, and fork+exec and process. They have different |
29 |
advantages and disadvantages that I describe below, together with how this |
30 |
module tries to mitigate the disadvantages. |
31 |
|
32 |
=over 4 |
33 |
|
34 |
=item Forking from a big process can be very slow (a 5GB process needs |
35 |
0.05s to fork on my 3.6GHz amd64 GNU/Linux box for example). This overhead |
36 |
is often shared with exec (because you have to fork first), but in some |
37 |
circumstances (e.g. when vfork is used), fork+exec can be much faster. |
38 |
|
39 |
This module can help here by telling a small(er) helper process to fork, |
40 |
or fork+exec instead. |
41 |
|
42 |
=item Forking usually creates a copy-on-write copy of the parent |
43 |
process. Memory (for example, modules or data files that have been |
44 |
will not take additional memory). When exec'ing a new process, modules |
45 |
and data files might need to be loaded again, at extra cpu and memory |
46 |
cost. Likewise when forking, all data structures are copied as well - if |
47 |
the program frees them and replaces them by new data, the child processes |
48 |
will retain the memory even if it isn't used. |
49 |
|
50 |
This module allows the main program to do a controlled fork, and allows |
51 |
modules to exec processes safely at any time. When creating a custom |
52 |
process pool you can take advantage of data sharing via fork without |
53 |
risking to share large dynamic data structures that will blow up child |
54 |
memory usage. |
55 |
|
56 |
=item Exec'ing a new perl process might be difficult and slow. For |
57 |
example, it is not easy to find the correct path to the perl interpreter, |
58 |
and all modules have to be loaded from disk again. Long running processes |
59 |
might run into problems when perl is upgraded for example. |
60 |
|
61 |
This module supports creating pre-initialised perl processes to be used |
62 |
as template, and also tries hard to identify the correct path to the perl |
63 |
interpreter. With a cooperative main program, exec'ing the interpreter |
64 |
might not even be necessary. |
65 |
|
66 |
=item Forking might be impossible when a program is running. For example, |
67 |
POSIX makes it almost impossible to fork from a multithreaded program and |
68 |
do anything useful in the child - strictly speaking, if your perl program |
69 |
uses posix threads (even indirectly via e.g. L<IO::AIO> or L<threads>), |
70 |
you cannot call fork on the perl level anymore, at all. |
71 |
|
72 |
This module can safely fork helper processes at any time, by caling |
73 |
fork+exec in C, in a POSIX-compatible way. |
74 |
|
75 |
=item Parallel processing with fork might be inconvenient or difficult |
76 |
to implement. For example, when a program uses an event loop and creates |
77 |
watchers it becomes very hard to use the event loop from a child |
78 |
program, as the watchers already exist but are only meaningful in the |
79 |
parent. Worse, a module might want to use such a system, not knowing |
80 |
whether another module or the main program also does, leading to problems. |
81 |
|
82 |
This module only lets the main program create pools by forking (because |
83 |
only the main program can know when it is still safe to do so) - all other |
84 |
pools are created by fork+exec, after which such modules can again be |
85 |
loaded. |
86 |
|
87 |
=back |
88 |
|
89 |
=head1 CONCEPTS |
90 |
|
91 |
This module can create new processes either by executing a new perl |
92 |
process, or by forking from an existing "template" process. |
93 |
|
94 |
Each such process comes with its own file handle that can be used to |
95 |
communicate with it (it's actually a socket - one end in the new process, |
96 |
one end in the main process), and among the things you can do in it are |
97 |
load modules, fork new processes, send file handles to it, and execute |
98 |
functions. |
99 |
|
100 |
There are multiple ways to create additional processes to execute some |
101 |
jobs: |
102 |
|
103 |
=over 4 |
104 |
|
105 |
=item fork a new process from the "default" template process, load code, |
106 |
run it |
107 |
|
108 |
This module has a "default" template process which it executes when it is |
109 |
needed the first time. Forking from this process shares the memory used |
110 |
for the perl interpreter with the new process, but loading modules takes |
111 |
time, and the memory is not shared with anything else. |
112 |
|
113 |
This is ideal for when you only need one extra process of a kind, with the |
114 |
option of starting and stipping it on demand. |
115 |
|
116 |
=item fork a new template process, load code, then fork processes off of |
117 |
it and run the code |
118 |
|
119 |
When you need to have a bunch of processes that all execute the same (or |
120 |
very similar) tasks, then a good way is to create a new template process |
121 |
for them, loading all the modules you need, and then create your worker |
122 |
processes from this new template process. |
123 |
|
124 |
This way, all code (and data structures) that can be shared (e.g. the |
125 |
modules you loaded) is shared between the processes, and each new process |
126 |
consumes relatively little memory of its own. |
127 |
|
128 |
The disadvantage of this approach is that you need to create a template |
129 |
process for the sole purpose of forking new processes from it, but if you |
130 |
only need a fixed number of proceses you can create them, and then destroy |
131 |
the template process. |
132 |
|
133 |
=item execute a new perl interpreter, load some code, run it |
134 |
|
135 |
This is relatively slow, and doesn't allow you to share memory between |
136 |
multiple processes. |
137 |
|
138 |
The only advantage is that you don't have to have a template process |
139 |
hanging around all the time to fork off some new processes, which might be |
140 |
an advantage when there are long time spans where no extra processes are |
141 |
needed. |
142 |
|
143 |
=back |
144 |
|
145 |
=head1 FUNCTIONS |
146 |
|
147 |
=over 4 |
148 |
|
149 |
=cut |
150 |
|
151 |
package AnyEvent::Fork; |
152 |
|
153 |
use common::sense; |
154 |
|
155 |
use Socket (); |
156 |
|
157 |
use AnyEvent; |
158 |
use AnyEvent::Fork::Util; |
159 |
use AnyEvent::Util (); |
160 |
|
161 |
our $PERL; # the path to the perl interpreter, deduces with various forms of magic |
162 |
|
163 |
=item my $pool = new AnyEvent::Fork key => value... |
164 |
|
165 |
Create a new process pool. The following named parameters are supported: |
166 |
|
167 |
=over 4 |
168 |
|
169 |
=back |
170 |
|
171 |
=cut |
172 |
|
173 |
# the early fork template process |
174 |
our $EARLY; |
175 |
|
176 |
# the empty template process |
177 |
our $TEMPLATE; |
178 |
|
179 |
sub _cmd { |
180 |
my $self = shift; |
181 |
|
182 |
# ideally, we would want to use "a (w/a)*" as format string, but perl versions |
183 |
# from at least 5.8.9 to 5.16.3 are all buggy and can't unpack it. |
184 |
push @{ $self->[2] }, pack "N/a", pack "(w/a)*", @_; |
185 |
|
186 |
$self->[3] ||= AE::io $self->[1], 1, sub { |
187 |
if (ref $self->[2][0]) { |
188 |
AnyEvent::Fork::Util::fd_send fileno $self->[1], fileno ${ $self->[2][0] } |
189 |
and shift @{ $self->[2] }; |
190 |
|
191 |
} else { |
192 |
my $len = syswrite $self->[1], $self->[2][0] |
193 |
or do { undef $self->[3]; die "AnyEvent::Fork: command write failure: $!" }; |
194 |
|
195 |
substr $self->[2][0], 0, $len, ""; |
196 |
shift @{ $self->[2] } unless length $self->[2][0]; |
197 |
} |
198 |
|
199 |
unless (@{ $self->[2] }) { |
200 |
undef $self->[3]; |
201 |
$self->[0]->($self->[1]) if $self->[0]; |
202 |
} |
203 |
}; |
204 |
} |
205 |
|
206 |
sub _new { |
207 |
my ($self, $fh) = @_; |
208 |
|
209 |
AnyEvent::Util::fh_nonblocking $fh, 1; |
210 |
|
211 |
$self = bless [ |
212 |
undef, # run callback |
213 |
$fh, |
214 |
[], # write queue - strings or fd's |
215 |
undef, # AE watcher |
216 |
], $self; |
217 |
|
218 |
# my ($a, $b) = AnyEvent::Util::portable_socketpair; |
219 |
|
220 |
# queue_cmd $template, "Iabc"; |
221 |
# push @{ $template->[2] }, \$b; |
222 |
|
223 |
# use Coro::AnyEvent; Coro::AnyEvent::sleep 1; |
224 |
# undef $b; |
225 |
# die "x" . <$a>; |
226 |
|
227 |
$self |
228 |
} |
229 |
|
230 |
# fork template from current process, used by AnyEvent::Fork::Early/Template |
231 |
sub _new_fork { |
232 |
my ($fh, $slave) = AnyEvent::Util::portable_socketpair; |
233 |
my $pid = fork; |
234 |
|
235 |
if ($pid eq 0) { |
236 |
require AnyEvent::Fork::Serve; |
237 |
close $fh; |
238 |
AnyEvent::Fork::Serve::serve ($slave); |
239 |
AnyEvent::Fork::Util::_exit 0; |
240 |
} elsif (!$pid) { |
241 |
die "AnyEvent::Fork::Early/Template: unable to fork template process: $!"; |
242 |
} |
243 |
|
244 |
AnyEvent::Fork->_new ($fh) |
245 |
} |
246 |
|
247 |
=item my $proc = new AnyEvent::Fork |
248 |
|
249 |
Create a new "empty" perl interpreter process and returns its process |
250 |
object for further manipulation. |
251 |
|
252 |
The new process is forked from a template process that is kept around |
253 |
for this purpose. When it doesn't exist yet, it is created by a call to |
254 |
C<new_exec> and kept around for future calls. |
255 |
|
256 |
=cut |
257 |
|
258 |
sub new { |
259 |
my $class = shift; |
260 |
|
261 |
$TEMPLATE ||= $class->new_exec; |
262 |
$TEMPLATE->fork |
263 |
} |
264 |
|
265 |
=item $new_proc = $proc->fork |
266 |
|
267 |
Forks C<$proc>, creating a new process, and returns the process object |
268 |
of the new process. |
269 |
|
270 |
If any of the C<send_> functions have been called before fork, then they |
271 |
will be cloned in the child. For example, in a pre-forked server, you |
272 |
might C<send_fh> the listening socket into the template process, and then |
273 |
keep calling C<fork> and C<run>. |
274 |
|
275 |
=cut |
276 |
|
277 |
sub fork { |
278 |
my ($self) = @_; |
279 |
|
280 |
my ($fh, $slave) = AnyEvent::Util::portable_socketpair; |
281 |
|
282 |
$self->send_fh ($slave); |
283 |
$self->_cmd ("f"); |
284 |
|
285 |
AnyEvent::Fork->_new ($fh) |
286 |
} |
287 |
|
288 |
=item my $proc = new_exec AnyEvent::Fork |
289 |
|
290 |
Create a new "empty" perl interpreter process and returns its process |
291 |
object for further manipulation. |
292 |
|
293 |
Unlike the C<new> method, this method I<always> spawns a new perl process |
294 |
(except in some cases, see L<AnyEvent::Fork::Early> for details). This |
295 |
reduces the amount of memory sharing that is possible, and is also slower. |
296 |
|
297 |
You should use C<new> whenever possible, except when having a template |
298 |
process around is unacceptable. |
299 |
|
300 |
The path to the perl interpreter is divined usign various methods - first |
301 |
C<$^X> is investigated to see if the path ends with something that sounds |
302 |
as if it were the perl interpreter. Failing this, the module falls back to |
303 |
using C<$Config::Config{perlpath}>. |
304 |
|
305 |
=cut |
306 |
|
307 |
sub new_exec { |
308 |
my ($self) = @_; |
309 |
|
310 |
return $EARLY->fork |
311 |
if $EARLY; |
312 |
|
313 |
# first find path of perl |
314 |
my $perl = $; |
315 |
|
316 |
# first we try $^X, but the path must be absolute (always on win32), and end in sth. |
317 |
# that looks like perl. this obviously only works for posix and win32 |
318 |
unless ( |
319 |
(AnyEvent::Fork::Util::WIN32 || $perl =~ m%^/%) |
320 |
&& $perl =~ m%[/\\]perl(?:[0-9]+(\.[0-9]+)+)?(\.exe)?$%i |
321 |
) { |
322 |
# if it doesn't look perlish enough, try Config |
323 |
require Config; |
324 |
$perl = $Config::Config{perlpath}; |
325 |
$perl =~ s/(?:\Q$Config::Config{_exe}\E)?$/$Config::Config{_exe}/; |
326 |
} |
327 |
|
328 |
require Proc::FastSpawn; |
329 |
|
330 |
my ($fh, $slave) = AnyEvent::Util::portable_socketpair; |
331 |
Proc::FastSpawn::fd_inherit (fileno $slave); |
332 |
|
333 |
# quick. also doesn't work in win32. of course. what did you expect |
334 |
#local $ENV{PERL5LIB} = join ":", grep !ref, @INC; |
335 |
my %env = %ENV; |
336 |
$env{PERL5LIB} = join ":", grep !ref, @INC; |
337 |
|
338 |
Proc::FastSpawn::spawn ( |
339 |
$perl, |
340 |
["perl", "-MAnyEvent::Fork::Serve", "-e", "AnyEvent::Fork::Serve::me", fileno $slave], |
341 |
[map "$_=$env{$_}", keys %env], |
342 |
) or die "unable to spawn AnyEvent::Fork server: $!"; |
343 |
|
344 |
$self->_new ($fh) |
345 |
} |
346 |
|
347 |
=item $proc = $proc->require ($module, ...) |
348 |
|
349 |
Tries to load the given modules into the process |
350 |
|
351 |
Returns the process object for easy chaining of method calls. |
352 |
|
353 |
=item $proc = $proc->send_fh ($handle, ...) |
354 |
|
355 |
Send one or more file handles (I<not> file descriptors) to the process, |
356 |
to prepare a call to C<run>. |
357 |
|
358 |
The process object keeps a reference to the handles until this is done, |
359 |
so you must not explicitly close the handles. This is most easily |
360 |
accomplished by simply not storing the file handles anywhere after passing |
361 |
them to this method. |
362 |
|
363 |
Returns the process object for easy chaining of method calls. |
364 |
|
365 |
=cut |
366 |
|
367 |
sub send_fh { |
368 |
my ($self, @fh) = @_; |
369 |
|
370 |
for my $fh (@fh) { |
371 |
$self->_cmd ("h"); |
372 |
push @{ $self->[2] }, \$fh; |
373 |
} |
374 |
|
375 |
$self |
376 |
} |
377 |
|
378 |
=item $proc = $proc->send_arg ($string, ...) |
379 |
|
380 |
Send one or more argument strings to the process, to prepare a call to |
381 |
C<run>. The strings can be any octet string. |
382 |
|
383 |
Returns the process object for easy chaining of emthod calls. |
384 |
|
385 |
=cut |
386 |
|
387 |
sub send_arg { |
388 |
my ($self, @arg) = @_; |
389 |
|
390 |
$self->_cmd (a => @arg); |
391 |
|
392 |
$self |
393 |
} |
394 |
|
395 |
=item $proc->run ($func, $cb->($fh)) |
396 |
|
397 |
Enter the function specified by the fully qualified name in C<$func> in |
398 |
the process. The function is called with the communication socket as first |
399 |
argument, followed by all file handles and string arguments sent earlier |
400 |
via C<send_fh> and C<send_arg> methods, in the order they were called. |
401 |
|
402 |
If the called function returns, the process exits. |
403 |
|
404 |
Preparing the process can take time - when the process is ready, the |
405 |
callback is invoked with the local communications socket as argument. |
406 |
|
407 |
The process object becomes unusable on return from this function. |
408 |
|
409 |
If the communication socket isn't used, it should be closed on both sides, |
410 |
to save on kernel memory. |
411 |
|
412 |
The socket is non-blocking in the parent, and blocking in the newly |
413 |
created process. The close-on-exec flag is set on both. Even if not used |
414 |
otherwise, the socket can be a good indicator for the existance of the |
415 |
process - if the othe rprocess exits, you get a readable event on it, |
416 |
because exiting the process closes the socket (if it didn't create any |
417 |
children using fork). |
418 |
|
419 |
=cut |
420 |
|
421 |
sub run { |
422 |
my ($self, $func, $cb) = @_; |
423 |
|
424 |
$self->[0] = $cb; |
425 |
$self->_cmd ("r", $func); |
426 |
} |
427 |
|
428 |
=back |
429 |
|
430 |
=head1 AUTHOR |
431 |
|
432 |
Marc Lehmann <schmorp@schmorp.de> |
433 |
http://home.schmorp.de/ |
434 |
|
435 |
=cut |
436 |
|
437 |
1 |
438 |
|