1 |
NAME |
2 |
AnyEvent::Fork - everything you wanted to use fork() for, but couldn't |
3 |
|
4 |
SYNOPSIS |
5 |
use AnyEvent::Fork; |
6 |
|
7 |
AnyEvent::Fork |
8 |
->new |
9 |
->require ("MyModule") |
10 |
->run ("MyModule::server", my $cv = AE::cv); |
11 |
|
12 |
my $fh = $cv->recv; |
13 |
|
14 |
DESCRIPTION |
15 |
This module allows you to create new processes, without actually forking |
16 |
them from your current process (avoiding the problems of forking), but |
17 |
preserving most of the advantages of fork. |
18 |
|
19 |
It can be used to create new worker processes or new independent |
20 |
subprocesses for short- and long-running jobs, process pools (e.g. for |
21 |
use in pre-forked servers) but also to spawn new external processes |
22 |
(such as CGI scripts from a web server), which can be faster (and more |
23 |
well behaved) than using fork+exec in big processes. |
24 |
|
25 |
Special care has been taken to make this module useful from other |
26 |
modules, while still supporting specialised environments such as |
27 |
App::Staticperl or PAR::Packer. |
28 |
|
29 |
WHAT THIS MODULE IS NOT |
30 |
This module only creates processes and lets you pass file handles and |
31 |
strings to it, and run perl code. It does not implement any kind of RPC |
32 |
- there is no back channel from the process back to you, and there is no |
33 |
RPC or message passing going on. |
34 |
|
35 |
If you need some form of RPC, you could use the AnyEvent::Fork::RPC |
36 |
companion module, which adds simple RPC/job queueing to a process |
37 |
created by this module. |
38 |
|
39 |
And if you need some automatic process pool management on top of |
40 |
AnyEvent::Fork::RPC, you can look at the AnyEvent::Fork::Pool companion |
41 |
module. |
42 |
|
43 |
Or you can implement it yourself in whatever way you like: use some |
44 |
message-passing module such as AnyEvent::MP, some pipe such as |
45 |
AnyEvent::ZeroMQ, use AnyEvent::Handle on both sides to send e.g. JSON |
46 |
or Storable messages, and so on. |
47 |
|
48 |
COMPARISON TO OTHER MODULES |
49 |
There is an abundance of modules on CPAN that do "something fork", such |
50 |
as Parallel::ForkManager, AnyEvent::ForkManager, AnyEvent::Worker or |
51 |
AnyEvent::Subprocess. There are modules that implement their own process |
52 |
management, such as AnyEvent::DBI. |
53 |
|
54 |
The problems that all these modules try to solve are real, however, none |
55 |
of them (from what I have seen) tackle the very real problems of |
56 |
unwanted memory sharing, efficiency or not being able to use event |
57 |
processing, GUI toolkits or similar modules in the processes they |
58 |
create. |
59 |
|
60 |
This module doesn't try to replace any of them - instead it tries to |
61 |
solve the problem of creating processes with a minimum of fuss and |
62 |
overhead (and also luxury). Ideally, most of these would use |
63 |
AnyEvent::Fork internally, except they were written before AnyEvent:Fork |
64 |
was available, so obviously had to roll their own. |
65 |
|
66 |
PROBLEM STATEMENT |
67 |
There are two traditional ways to implement parallel processing on UNIX |
68 |
like operating systems - fork and process, and fork+exec and process. |
69 |
They have different advantages and disadvantages that I describe below, |
70 |
together with how this module tries to mitigate the disadvantages. |
71 |
|
72 |
Forking from a big process can be very slow. |
73 |
A 5GB process needs 0.05s to fork on my 3.6GHz amd64 GNU/Linux box. |
74 |
This overhead is often shared with exec (because you have to fork |
75 |
first), but in some circumstances (e.g. when vfork is used), |
76 |
fork+exec can be much faster. |
77 |
|
78 |
This module can help here by telling a small(er) helper process to |
79 |
fork, which is faster then forking the main process, and also uses |
80 |
vfork where possible. This gives the speed of vfork, with the |
81 |
flexibility of fork. |
82 |
|
83 |
Forking usually creates a copy-on-write copy of the parent process. |
84 |
For example, modules or data files that are loaded will not use |
85 |
additional memory after a fork. Exec'ing a new process, in contrast, |
86 |
means modules and data files might need to be loaded again, at extra |
87 |
CPU and memory cost. |
88 |
|
89 |
But when forking, you still create a copy of your data structures - |
90 |
if the program frees them and replaces them by new data, the child |
91 |
processes will retain the old version even if it isn't used, which |
92 |
can suddenly and unexpectedly increase memory usage when freeing |
93 |
memory. |
94 |
|
95 |
For example, Gtk2::CV is an image viewer optimised for large |
96 |
directories (millions of pictures). It also forks subprocesses for |
97 |
thumbnail generation, which inherit the data structure that stores |
98 |
all file information. If the user changes the directory, it gets |
99 |
freed in the main process, leaving a copy in the thumbnailer |
100 |
processes. This can lead to many times the memory usage that would |
101 |
actually be required. The solution is to fork early (and being |
102 |
unable to dynamically generate more subprocesses or do this from a |
103 |
module)... or to use <AnyEvent:Fork>. |
104 |
|
105 |
There is a trade-off between more sharing with fork (which can be |
106 |
good or bad), and no sharing with exec. |
107 |
|
108 |
This module allows the main program to do a controlled fork, and |
109 |
allows modules to exec processes safely at any time. When creating a |
110 |
custom process pool you can take advantage of data sharing via fork |
111 |
without risking to share large dynamic data structures that will |
112 |
blow up child memory usage. |
113 |
|
114 |
In other words, this module puts you into control over what is being |
115 |
shared and what isn't, at all times. |
116 |
|
117 |
Exec'ing a new perl process might be difficult. |
118 |
For example, it is not easy to find the correct path to the perl |
119 |
interpreter - $^X might not be a perl interpreter at all. Worse, |
120 |
there might not even be a perl binary installed on the system. |
121 |
|
122 |
This module tries hard to identify the correct path to the perl |
123 |
interpreter. With a cooperative main program, exec'ing the |
124 |
interpreter might not even be necessary, but even without help from |
125 |
the main program, it will still work when used from a module. |
126 |
|
127 |
Exec'ing a new perl process might be slow, as all necessary modules have |
128 |
to be loaded from disk again, with no guarantees of success. |
129 |
Long running processes might run into problems when perl is upgraded |
130 |
and modules are no longer loadable because they refer to a different |
131 |
perl version, or parts of a distribution are newer than the ones |
132 |
already loaded. |
133 |
|
134 |
This module supports creating pre-initialised perl processes to be |
135 |
used as a template for new processes at a later time, e.g. for use |
136 |
in a process pool. |
137 |
|
138 |
Forking might be impossible when a program is running. |
139 |
For example, POSIX makes it almost impossible to fork from a |
140 |
multi-threaded program while doing anything useful in the child - in |
141 |
fact, if your perl program uses POSIX threads (even indirectly via |
142 |
e.g. IO::AIO or threads), you cannot call fork on the perl level |
143 |
anymore without risking memory corruption or worse on a number of |
144 |
operating systems. |
145 |
|
146 |
This module can safely fork helper processes at any time, by calling |
147 |
fork+exec in C, in a POSIX-compatible way (via Proc::FastSpawn). |
148 |
|
149 |
Parallel processing with fork might be inconvenient or difficult to |
150 |
implement. Modules might not work in both parent and child. |
151 |
For example, when a program uses an event loop and creates watchers |
152 |
it becomes very hard to use the event loop from a child program, as |
153 |
the watchers already exist but are only meaningful in the parent. |
154 |
Worse, a module might want to use such a module, not knowing whether |
155 |
another module or the main program also does, leading to problems. |
156 |
|
157 |
Apart from event loops, graphical toolkits also commonly fall into |
158 |
the "unsafe module" category, or just about anything that |
159 |
communicates with the external world, such as network libraries and |
160 |
file I/O modules, which usually don't like being copied and then |
161 |
allowed to continue in two processes. |
162 |
|
163 |
With this module only the main program is allowed to create new |
164 |
processes by forking (because only the main program can know when it |
165 |
is still safe to do so) - all other processes are created via |
166 |
fork+exec, which makes it possible to use modules such as event |
167 |
loops or window interfaces safely. |
168 |
|
169 |
EXAMPLES |
170 |
This is where the wall of text ends and code speaks. |
171 |
|
172 |
Create a single new process, tell it to run your worker function. |
173 |
AnyEvent::Fork |
174 |
->new |
175 |
->require ("MyModule") |
176 |
->run ("MyModule::worker, sub { |
177 |
my ($master_filehandle) = @_; |
178 |
|
179 |
# now $master_filehandle is connected to the |
180 |
# $slave_filehandle in the new process. |
181 |
}); |
182 |
|
183 |
"MyModule" might look like this: |
184 |
|
185 |
package MyModule; |
186 |
|
187 |
sub worker { |
188 |
my ($slave_filehandle) = @_; |
189 |
|
190 |
# now $slave_filehandle is connected to the $master_filehandle |
191 |
# in the original process. have fun! |
192 |
} |
193 |
|
194 |
Create a pool of server processes all accepting on the same socket. |
195 |
# create listener socket |
196 |
my $listener = ...; |
197 |
|
198 |
# create a pool template, initialise it and give it the socket |
199 |
my $pool = AnyEvent::Fork |
200 |
->new |
201 |
->require ("Some::Stuff", "My::Server") |
202 |
->send_fh ($listener); |
203 |
|
204 |
# now create 10 identical workers |
205 |
for my $id (1..10) { |
206 |
$pool |
207 |
->fork |
208 |
->send_arg ($id) |
209 |
->run ("My::Server::run"); |
210 |
} |
211 |
|
212 |
# now do other things - maybe use the filehandle provided by run |
213 |
# to wait for the processes to die. or whatever. |
214 |
|
215 |
"My::Server" might look like this: |
216 |
|
217 |
package My::Server; |
218 |
|
219 |
sub run { |
220 |
my ($slave, $listener, $id) = @_; |
221 |
|
222 |
close $slave; # we do not use the socket, so close it to save resources |
223 |
|
224 |
# we could go ballistic and use e.g. AnyEvent here, or IO::AIO, |
225 |
# or anything we usually couldn't do in a process forked normally. |
226 |
while (my $socket = $listener->accept) { |
227 |
# do sth. with new socket |
228 |
} |
229 |
} |
230 |
|
231 |
use AnyEvent::Fork as a faster fork+exec |
232 |
This runs "/bin/echo hi", with standard output redirected to /tmp/log |
233 |
and standard error redirected to the communications socket. It is |
234 |
usually faster than fork+exec, but still lets you prepare the |
235 |
environment. |
236 |
|
237 |
open my $output, ">/tmp/log" or die "$!"; |
238 |
|
239 |
AnyEvent::Fork |
240 |
->new |
241 |
->eval (' |
242 |
# compile a helper function for later use |
243 |
sub run { |
244 |
my ($fh, $output, @cmd) = @_; |
245 |
|
246 |
# perl will clear close-on-exec on STDOUT/STDERR |
247 |
open STDOUT, ">&", $output or die; |
248 |
open STDERR, ">&", $fh or die; |
249 |
|
250 |
exec @cmd; |
251 |
} |
252 |
') |
253 |
->send_fh ($output) |
254 |
->send_arg ("/bin/echo", "hi") |
255 |
->run ("run", my $cv = AE::cv); |
256 |
|
257 |
my $stderr = $cv->recv; |
258 |
|
259 |
For stingy users: put the worker code into a "DATA" section. |
260 |
When you want to be stingy with files, you can put your code into the |
261 |
"DATA" section of your module (or program): |
262 |
|
263 |
use AnyEvent::Fork; |
264 |
|
265 |
AnyEvent::Fork |
266 |
->new |
267 |
->eval (do { local $/; <DATA> }) |
268 |
->run ("doit", sub { ... }); |
269 |
|
270 |
__DATA__ |
271 |
|
272 |
sub doit { |
273 |
... do something! |
274 |
} |
275 |
|
276 |
For stingy standalone programs: do not rely on external files at |
277 |
all. |
278 |
For single-file scripts it can be inconvenient to rely on external files |
279 |
- even when using a "DATA" section, you still need to "exec" an external |
280 |
perl interpreter, which might not be available when using |
281 |
App::Staticperl, Urlader or PAR::Packer for example. |
282 |
|
283 |
Two modules help here - AnyEvent::Fork::Early forks a template process |
284 |
for all further calls to "new_exec", and AnyEvent::Fork::Template forks |
285 |
the main program as a template process. |
286 |
|
287 |
Here is how your main program should look like: |
288 |
|
289 |
#! perl |
290 |
|
291 |
# optional, as the very first thing. |
292 |
# in case modules want to create their own processes. |
293 |
use AnyEvent::Fork::Early; |
294 |
|
295 |
# next, load all modules you need in your template process |
296 |
use Example::My::Module |
297 |
use Example::Whatever; |
298 |
|
299 |
# next, put your run function definition and anything else you |
300 |
# need, but do not use code outside of BEGIN blocks. |
301 |
sub worker_run { |
302 |
my ($fh, @args) = @_; |
303 |
... |
304 |
} |
305 |
|
306 |
# now preserve everything so far as AnyEvent::Fork object |
307 |
# in $TEMPLATE. |
308 |
use AnyEvent::Fork::Template; |
309 |
|
310 |
# do not put code outside of BEGIN blocks until here |
311 |
|
312 |
# now use the $TEMPLATE process in any way you like |
313 |
|
314 |
# for example: create 10 worker processes |
315 |
my @worker; |
316 |
my $cv = AE::cv; |
317 |
for (1..10) { |
318 |
$cv->begin; |
319 |
$TEMPLATE->fork->send_arg ($_)->run ("worker_run", sub { |
320 |
push @worker, shift; |
321 |
$cv->end; |
322 |
}); |
323 |
} |
324 |
$cv->recv; |
325 |
|
326 |
CONCEPTS |
327 |
This module can create new processes either by executing a new perl |
328 |
process, or by forking from an existing "template" process. |
329 |
|
330 |
All these processes are called "child processes" (whether they are |
331 |
direct children or not), while the process that manages them is called |
332 |
the "parent process". |
333 |
|
334 |
Each such process comes with its own file handle that can be used to |
335 |
communicate with it (it's actually a socket - one end in the new |
336 |
process, one end in the main process), and among the things you can do |
337 |
in it are load modules, fork new processes, send file handles to it, and |
338 |
execute functions. |
339 |
|
340 |
There are multiple ways to create additional processes to execute some |
341 |
jobs: |
342 |
|
343 |
fork a new process from the "default" template process, load code, run |
344 |
it |
345 |
This module has a "default" template process which it executes when |
346 |
it is needed the first time. Forking from this process shares the |
347 |
memory used for the perl interpreter with the new process, but |
348 |
loading modules takes time, and the memory is not shared with |
349 |
anything else. |
350 |
|
351 |
This is ideal for when you only need one extra process of a kind, |
352 |
with the option of starting and stopping it on demand. |
353 |
|
354 |
Example: |
355 |
|
356 |
AnyEvent::Fork |
357 |
->new |
358 |
->require ("Some::Module") |
359 |
->run ("Some::Module::run", sub { |
360 |
my ($fork_fh) = @_; |
361 |
}); |
362 |
|
363 |
fork a new template process, load code, then fork processes off of it |
364 |
and run the code |
365 |
When you need to have a bunch of processes that all execute the same |
366 |
(or very similar) tasks, then a good way is to create a new template |
367 |
process for them, loading all the modules you need, and then create |
368 |
your worker processes from this new template process. |
369 |
|
370 |
This way, all code (and data structures) that can be shared (e.g. |
371 |
the modules you loaded) is shared between the processes, and each |
372 |
new process consumes relatively little memory of its own. |
373 |
|
374 |
The disadvantage of this approach is that you need to create a |
375 |
template process for the sole purpose of forking new processes from |
376 |
it, but if you only need a fixed number of processes you can create |
377 |
them, and then destroy the template process. |
378 |
|
379 |
Example: |
380 |
|
381 |
my $template = AnyEvent::Fork->new->require ("Some::Module"); |
382 |
|
383 |
for (1..10) { |
384 |
$template->fork->run ("Some::Module::run", sub { |
385 |
my ($fork_fh) = @_; |
386 |
}); |
387 |
} |
388 |
|
389 |
# at this point, you can keep $template around to fork new processes |
390 |
# later, or you can destroy it, which causes it to vanish. |
391 |
|
392 |
execute a new perl interpreter, load some code, run it |
393 |
This is relatively slow, and doesn't allow you to share memory |
394 |
between multiple processes. |
395 |
|
396 |
The only advantage is that you don't have to have a template process |
397 |
hanging around all the time to fork off some new processes, which |
398 |
might be an advantage when there are long time spans where no extra |
399 |
processes are needed. |
400 |
|
401 |
Example: |
402 |
|
403 |
AnyEvent::Fork |
404 |
->new_exec |
405 |
->require ("Some::Module") |
406 |
->run ("Some::Module::run", sub { |
407 |
my ($fork_fh) = @_; |
408 |
}); |
409 |
|
410 |
THE "AnyEvent::Fork" CLASS |
411 |
This module exports nothing, and only implements a single class - |
412 |
"AnyEvent::Fork". |
413 |
|
414 |
There are two class constructors that both create new processes - "new" |
415 |
and "new_exec". The "fork" method creates a new process by forking an |
416 |
existing one and could be considered a third constructor. |
417 |
|
418 |
Most of the remaining methods deal with preparing the new process, by |
419 |
loading code, evaluating code and sending data to the new process. They |
420 |
usually return the process object, so you can chain method calls. |
421 |
|
422 |
If a process object is destroyed before calling its "run" method, then |
423 |
the process simply exits. After "run" is called, all responsibility is |
424 |
passed to the specified function. |
425 |
|
426 |
As long as there is any outstanding work to be done, process objects |
427 |
resist being destroyed, so there is no reason to store them unless you |
428 |
need them later - configure and forget works just fine. |
429 |
|
430 |
my $proc = new AnyEvent::Fork |
431 |
Create a new "empty" perl interpreter process and returns its |
432 |
process object for further manipulation. |
433 |
|
434 |
The new process is forked from a template process that is kept |
435 |
around for this purpose. When it doesn't exist yet, it is created by |
436 |
a call to "new_exec" first and then stays around for future calls. |
437 |
|
438 |
$new_proc = $proc->fork |
439 |
Forks $proc, creating a new process, and returns the process object |
440 |
of the new process. |
441 |
|
442 |
If any of the "send_" functions have been called before fork, then |
443 |
they will be cloned in the child. For example, in a pre-forked |
444 |
server, you might "send_fh" the listening socket into the template |
445 |
process, and then keep calling "fork" and "run". |
446 |
|
447 |
my $proc = new_exec AnyEvent::Fork |
448 |
Create a new "empty" perl interpreter process and returns its |
449 |
process object for further manipulation. |
450 |
|
451 |
Unlike the "new" method, this method *always* spawns a new perl |
452 |
process (except in some cases, see AnyEvent::Fork::Early for |
453 |
details). This reduces the amount of memory sharing that is |
454 |
possible, and is also slower. |
455 |
|
456 |
You should use "new" whenever possible, except when having a |
457 |
template process around is unacceptable. |
458 |
|
459 |
The path to the perl interpreter is divined using various methods - |
460 |
first $^X is investigated to see if the path ends with something |
461 |
that looks as if it were the perl interpreter. Failing this, the |
462 |
module falls back to using $Config::Config{perlpath}. |
463 |
|
464 |
The path to perl can also be overriden by setting the global |
465 |
variable $AnyEvent::Fork::PERL - it's value will be used for all |
466 |
subsequent invocations. |
467 |
|
468 |
$pid = $proc->pid |
469 |
Returns the process id of the process *iff it is a direct child of |
470 |
the process running AnyEvent::Fork*, and "undef" otherwise. As a |
471 |
general rule (that you cannot rely upon), processes created via |
472 |
"new_exec", AnyEvent::Fork::Early or AnyEvent::Fork::Template are |
473 |
direct children, while all other processes are not. |
474 |
|
475 |
Or in other words, you do not normally have to take care of zombies |
476 |
for processes created via "new", but when in doubt, or zombies are a |
477 |
problem, you need to check whether a process is a diretc child by |
478 |
calling this method, and possibly creating a child watcher or reap |
479 |
it manually. |
480 |
|
481 |
$proc = $proc->eval ($perlcode, @args) |
482 |
Evaluates the given $perlcode as ... Perl code, while setting @_ to |
483 |
the strings specified by @args, in the "main" package. |
484 |
|
485 |
This call is meant to do any custom initialisation that might be |
486 |
required (for example, the "require" method uses it). It's not |
487 |
supposed to be used to completely take over the process, use "run" |
488 |
for that. |
489 |
|
490 |
The code will usually be executed after this call returns, and there |
491 |
is no way to pass anything back to the calling process. Any |
492 |
evaluation errors will be reported to stderr and cause the process |
493 |
to exit. |
494 |
|
495 |
If you want to execute some code (that isn't in a module) to take |
496 |
over the process, you should compile a function via "eval" first, |
497 |
and then call it via "run". This also gives you access to any |
498 |
arguments passed via the "send_xxx" methods, such as file handles. |
499 |
See the "use AnyEvent::Fork as a faster fork+exec" example to see it |
500 |
in action. |
501 |
|
502 |
Returns the process object for easy chaining of method calls. |
503 |
|
504 |
It's common to want to call an iniitalisation function with some |
505 |
arguments. Make sure you actually pass @_ to that function (for |
506 |
example by using &name syntax), and do not just specify a function |
507 |
name: |
508 |
|
509 |
$proc->eval ('&MyModule::init', $string1, $string2); |
510 |
|
511 |
$proc = $proc->require ($module, ...) |
512 |
Tries to load the given module(s) into the process |
513 |
|
514 |
Returns the process object for easy chaining of method calls. |
515 |
|
516 |
$proc = $proc->send_fh ($handle, ...) |
517 |
Send one or more file handles (*not* file descriptors) to the |
518 |
process, to prepare a call to "run". |
519 |
|
520 |
The process object keeps a reference to the handles until they have |
521 |
been passed over to the process, so you must not explicitly close |
522 |
the handles. This is most easily accomplished by simply not storing |
523 |
the file handles anywhere after passing them to this method - when |
524 |
AnyEvent::Fork is finished using them, perl will automatically close |
525 |
them. |
526 |
|
527 |
Returns the process object for easy chaining of method calls. |
528 |
|
529 |
Example: pass a file handle to a process, and release it without |
530 |
closing. It will be closed automatically when it is no longer used. |
531 |
|
532 |
$proc->send_fh ($my_fh); |
533 |
undef $my_fh; # free the reference if you want, but DO NOT CLOSE IT |
534 |
|
535 |
$proc = $proc->send_arg ($string, ...) |
536 |
Send one or more argument strings to the process, to prepare a call |
537 |
to "run". The strings can be any octet strings. |
538 |
|
539 |
The protocol is optimised to pass a moderate number of relatively |
540 |
short strings - while you can pass up to 4GB of data in one go, this |
541 |
is more meant to pass some ID information or other startup info, not |
542 |
big chunks of data. |
543 |
|
544 |
Returns the process object for easy chaining of method calls. |
545 |
|
546 |
$proc->run ($func, $cb->($fh)) |
547 |
Enter the function specified by the function name in $func in the |
548 |
process. The function is called with the communication socket as |
549 |
first argument, followed by all file handles and string arguments |
550 |
sent earlier via "send_fh" and "send_arg" methods, in the order they |
551 |
were called. |
552 |
|
553 |
The process object becomes unusable on return from this function - |
554 |
any further method calls result in undefined behaviour. |
555 |
|
556 |
The function name should be fully qualified, but if it isn't, it |
557 |
will be looked up in the "main" package. |
558 |
|
559 |
If the called function returns, doesn't exist, or any error occurs, |
560 |
the process exits. |
561 |
|
562 |
Preparing the process is done in the background - when all commands |
563 |
have been sent, the callback is invoked with the local |
564 |
communications socket as argument. At this point you can start using |
565 |
the socket in any way you like. |
566 |
|
567 |
If the communication socket isn't used, it should be closed on both |
568 |
sides, to save on kernel memory. |
569 |
|
570 |
The socket is non-blocking in the parent, and blocking in the newly |
571 |
created process. The close-on-exec flag is set in both. |
572 |
|
573 |
Even if not used otherwise, the socket can be a good indicator for |
574 |
the existence of the process - if the other process exits, you get a |
575 |
readable event on it, because exiting the process closes the socket |
576 |
(if it didn't create any children using fork). |
577 |
|
578 |
Compatibility to AnyEvent::Fork::Remote |
579 |
If you want to write code that works with both this module and |
580 |
AnyEvent::Fork::Remote, you need to write your code so that it |
581 |
assumes there are two file handles for communications, which |
582 |
might not be unix domain sockets. The "run" function should |
583 |
start like this: |
584 |
|
585 |
sub run { |
586 |
my ($rfh, @args) = @_; # @args is your normal arguments |
587 |
my $wfh = fileno $rfh ? $rfh : *STDOUT; |
588 |
|
589 |
# now use $rfh for reading and $wfh for writing |
590 |
} |
591 |
|
592 |
This checks whether the passed file handle is, in fact, the |
593 |
process "STDIN" handle. If it is, then the function was invoked |
594 |
visa AnyEvent::Fork::Remote, so STDIN should be used for reading |
595 |
and "STDOUT" should be used for writing. |
596 |
|
597 |
In all other cases, the function was called via this module, and |
598 |
there is only one file handle that should be sued for reading |
599 |
and writing. |
600 |
|
601 |
Example: create a template for a process pool, pass a few strings, |
602 |
some file handles, then fork, pass one more string, and run some |
603 |
code. |
604 |
|
605 |
my $pool = AnyEvent::Fork |
606 |
->new |
607 |
->send_arg ("str1", "str2") |
608 |
->send_fh ($fh1, $fh2); |
609 |
|
610 |
for (1..2) { |
611 |
$pool |
612 |
->fork |
613 |
->send_arg ("str3") |
614 |
->run ("Some::function", sub { |
615 |
my ($fh) = @_; |
616 |
|
617 |
# fh is nonblocking, but we trust that the OS can accept these |
618 |
# few octets anyway. |
619 |
syswrite $fh, "hi #$_\n"; |
620 |
|
621 |
# $fh is being closed here, as we don't store it anywhere |
622 |
}); |
623 |
} |
624 |
|
625 |
# Some::function might look like this - all parameters passed before fork |
626 |
# and after will be passed, in order, after the communications socket. |
627 |
sub Some::function { |
628 |
my ($fh, $str1, $str2, $fh1, $fh2, $str3) = @_; |
629 |
|
630 |
print scalar <$fh>; # prints "hi #1\n" and "hi #2\n" in any order |
631 |
} |
632 |
|
633 |
EXPERIMENTAL METHODS |
634 |
These methods might go away completely or change behaviour, at any time. |
635 |
|
636 |
$proc->to_fh ($cb->($fh)) # EXPERIMENTAL, MIGHT BE REMOVED |
637 |
Flushes all commands out to the process and then calls the callback |
638 |
with the communications socket. |
639 |
|
640 |
The process object becomes unusable on return from this function - |
641 |
any further method calls result in undefined behaviour. |
642 |
|
643 |
The point of this method is to give you a file handle that you can |
644 |
pass to another process. In that other process, you can call |
645 |
"new_from_fh AnyEvent::Fork $fh" to create a new "AnyEvent::Fork" |
646 |
object from it, thereby effectively passing a fork object to another |
647 |
process. |
648 |
|
649 |
new_from_fh AnyEvent::Fork $fh # EXPERIMENTAL, MIGHT BE REMOVED |
650 |
Takes a file handle originally rceeived by the "to_fh" method and |
651 |
creates a new "AnyEvent:Fork" object. The child process itself will |
652 |
not change in any way, i.e. it will keep all the modifications done |
653 |
to it before calling "to_fh". |
654 |
|
655 |
The new object is very much like the original object, except that |
656 |
the "pid" method will return "undef" even if the process is a direct |
657 |
child. |
658 |
|
659 |
PERFORMANCE |
660 |
Now for some unscientific benchmark numbers (all done on an amd64 |
661 |
GNU/Linux box). These are intended to give you an idea of the relative |
662 |
performance you can expect, they are not meant to be absolute |
663 |
performance numbers. |
664 |
|
665 |
OK, so, I ran a simple benchmark that creates a socket pair, forks, |
666 |
calls exit in the child and waits for the socket to close in the parent. |
667 |
I did load AnyEvent, EV and AnyEvent::Fork, for a total process size of |
668 |
5100kB. |
669 |
|
670 |
2079 new processes per second, using manual socketpair + fork |
671 |
|
672 |
Then I did the same thing, but instead of calling fork, I called |
673 |
AnyEvent::Fork->new->run ("CORE::exit") and then again waited for the |
674 |
socket from the child to close on exit. This does the same thing as |
675 |
manual socket pair + fork, except that what is forked is the template |
676 |
process (2440kB), and the socket needs to be passed to the server at the |
677 |
other end of the socket first. |
678 |
|
679 |
2307 new processes per second, using AnyEvent::Fork->new |
680 |
|
681 |
And finally, using "new_exec" instead "new", using vforks+execs to exec |
682 |
a new perl interpreter and compile the small server each time, I get: |
683 |
|
684 |
479 vfork+execs per second, using AnyEvent::Fork->new_exec |
685 |
|
686 |
So how can "AnyEvent->new" be faster than a standard fork, even though |
687 |
it uses the same operations, but adds a lot of overhead? |
688 |
|
689 |
The difference is simply the process size: forking the 5MB process takes |
690 |
so much longer than forking the 2.5MB template process that the extra |
691 |
overhead is canceled out. |
692 |
|
693 |
If the benchmark process grows, the normal fork becomes even slower: |
694 |
|
695 |
1340 new processes, manual fork of a 20MB process |
696 |
731 new processes, manual fork of a 200MB process |
697 |
235 new processes, manual fork of a 2000MB process |
698 |
|
699 |
What that means (to me) is that I can use this module without having a |
700 |
bad conscience because of the extra overhead required to start new |
701 |
processes. |
702 |
|
703 |
TYPICAL PROBLEMS |
704 |
This section lists typical problems that remain. I hope by recognising |
705 |
them, most can be avoided. |
706 |
|
707 |
leaked file descriptors for exec'ed processes |
708 |
POSIX systems inherit file descriptors by default when exec'ing a |
709 |
new process. While perl itself laudably sets the close-on-exec flags |
710 |
on new file handles, most C libraries don't care, and even if all |
711 |
cared, it's often not possible to set the flag in a race-free |
712 |
manner. |
713 |
|
714 |
That means some file descriptors can leak through. And since it |
715 |
isn't possible to know which file descriptors are "good" and |
716 |
"necessary" (or even to know which file descriptors are open), there |
717 |
is no good way to close the ones that might harm. |
718 |
|
719 |
As an example of what "harm" can be done consider a web server that |
720 |
accepts connections and afterwards some module uses AnyEvent::Fork |
721 |
for the first time, causing it to fork and exec a new process, which |
722 |
might inherit the network socket. When the server closes the socket, |
723 |
it is still open in the child (which doesn't even know that) and the |
724 |
client might conclude that the connection is still fine. |
725 |
|
726 |
For the main program, there are multiple remedies available - |
727 |
AnyEvent::Fork::Early is one, creating a process early and not using |
728 |
"new_exec" is another, as in both cases, the first process can be |
729 |
exec'ed well before many random file descriptors are open. |
730 |
|
731 |
In general, the solution for these kind of problems is to fix the |
732 |
libraries or the code that leaks those file descriptors. |
733 |
|
734 |
Fortunately, most of these leaked descriptors do no harm, other than |
735 |
sitting on some resources. |
736 |
|
737 |
leaked file descriptors for fork'ed processes |
738 |
Normally, AnyEvent::Fork does start new processes by exec'ing them, |
739 |
which closes file descriptors not marked for being inherited. |
740 |
|
741 |
However, AnyEvent::Fork::Early and AnyEvent::Fork::Template offer a |
742 |
way to create these processes by forking, and this leaks more file |
743 |
descriptors than exec'ing them, as there is no way to mark |
744 |
descriptors as "close on fork". |
745 |
|
746 |
An example would be modules like EV, IO::AIO or Gtk2. Both create |
747 |
pipes for internal uses, and Gtk2 might open a connection to the X |
748 |
server. EV and IO::AIO can deal with fork, but Gtk2 might have |
749 |
trouble with a fork. |
750 |
|
751 |
The solution is to either not load these modules before use'ing |
752 |
AnyEvent::Fork::Early or AnyEvent::Fork::Template, or to delay |
753 |
initialising them, for example, by calling "init Gtk2" manually. |
754 |
|
755 |
exiting calls object destructors |
756 |
This only applies to users of AnyEvent::Fork:Early and |
757 |
AnyEvent::Fork::Template, or when initialising code creates objects |
758 |
that reference external resources. |
759 |
|
760 |
When a process created by AnyEvent::Fork exits, it might do so by |
761 |
calling exit, or simply letting perl reach the end of the program. |
762 |
At which point Perl runs all destructors. |
763 |
|
764 |
Not all destructors are fork-safe - for example, an object that |
765 |
represents the connection to an X display might tell the X server to |
766 |
free resources, which is inconvenient when the "real" object in the |
767 |
parent still needs to use them. |
768 |
|
769 |
This is obviously not a problem for AnyEvent::Fork::Early, as you |
770 |
used it as the very first thing, right? |
771 |
|
772 |
It is a problem for AnyEvent::Fork::Template though - and the |
773 |
solution is to not create objects with nontrivial destructors that |
774 |
might have an effect outside of Perl. |
775 |
|
776 |
PORTABILITY NOTES |
777 |
Native win32 perls are somewhat supported (AnyEvent::Fork::Early is a |
778 |
nop, and ::Template is not going to work), and it cost a lot of blood |
779 |
and sweat to make it so, mostly due to the bloody broken perl that |
780 |
nobody seems to care about. The fork emulation is a bad joke - I have |
781 |
yet to see something useful that you can do with it without running into |
782 |
memory corruption issues or other braindamage. Hrrrr. |
783 |
|
784 |
Since fork is endlessly broken on win32 perls (it doesn't even remotely |
785 |
work within it's documented limits) and quite obviously it's not getting |
786 |
improved any time soon, the best way to proceed on windows would be to |
787 |
always use "new_exec" and thus never rely on perl's fork "emulation". |
788 |
|
789 |
Cygwin perl is not supported at the moment due to some hilarious |
790 |
shortcomings of its API - see IO::FDPoll for more details. If you never |
791 |
use "send_fh" and always use "new_exec" to create processes, it should |
792 |
work though. |
793 |
|
794 |
USING AnyEvent::Fork IN SUBPROCESSES |
795 |
AnyEvent::Fork itself cannot generally be used in subprocesses. As long |
796 |
as only one process ever forks new processes, sharing the template |
797 |
processes is possible (you could use a pipe as a lock by writing a byte |
798 |
into it to unlock, and reading the byte to lock for example) |
799 |
|
800 |
To make concurrent calls possible after fork, you should get rid of the |
801 |
template and early fork processes. AnyEvent::Fork will create a new |
802 |
template process as needed. |
803 |
|
804 |
undef $AnyEvent::Fork::EARLY; |
805 |
undef $AnyEvent::Fork::TEMPLATE; |
806 |
|
807 |
It doesn't matter whether you get rid of them in the parent or child |
808 |
after a fork. |
809 |
|
810 |
SEE ALSO |
811 |
AnyEvent::Fork::Early, to avoid executing a perl interpreter at all |
812 |
(part of this distribution). |
813 |
|
814 |
AnyEvent::Fork::Template, to create a process by forking the main |
815 |
program at a convenient time (part of this distribution). |
816 |
|
817 |
AnyEvent::Fork::Remote, for another way to create processes that is |
818 |
mostly compatible to this module and modules building on top of it, but |
819 |
works better with remote processes. |
820 |
|
821 |
AnyEvent::Fork::RPC, for simple RPC to child processes (on CPAN). |
822 |
|
823 |
AnyEvent::Fork::Pool, for simple worker process pool (on CPAN). |
824 |
|
825 |
AUTHOR AND CONTACT INFORMATION |
826 |
Marc Lehmann <schmorp@schmorp.de> |
827 |
http://software.schmorp.de/pkg/AnyEvent-Fork |
828 |
|