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