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, not being able to use event |
57 |
processing or similar modules in the processes they create. |
58 |
|
59 |
This module doesn't try to replace any of them - instead it tries to |
60 |
solve the problem of creating processes with a minimum of fuss and |
61 |
overhead (and also luxury). Ideally, most of these would use |
62 |
AnyEvent::Fork internally, except they were written before AnyEvent:Fork |
63 |
was available, so obviously had to roll their own. |
64 |
|
65 |
PROBLEM STATEMENT |
66 |
There are two traditional ways to implement parallel processing on UNIX |
67 |
like operating systems - fork and process, and fork+exec and process. |
68 |
They have different advantages and disadvantages that I describe below, |
69 |
together with how this module tries to mitigate the disadvantages. |
70 |
|
71 |
Forking from a big process can be very slow. |
72 |
A 5GB process needs 0.05s to fork on my 3.6GHz amd64 GNU/Linux box. |
73 |
This overhead is often shared with exec (because you have to fork |
74 |
first), but in some circumstances (e.g. when vfork is used), |
75 |
fork+exec can be much faster. |
76 |
|
77 |
This module can help here by telling a small(er) helper process to |
78 |
fork, which is faster then forking the main process, and also uses |
79 |
vfork where possible. This gives the speed of vfork, with the |
80 |
flexibility of fork. |
81 |
|
82 |
Forking usually creates a copy-on-write copy of the parent process. |
83 |
For example, modules or data files that are loaded will not use |
84 |
additional memory after a fork. When exec'ing a new process, modules |
85 |
and data files might need to be loaded again, at extra CPU and |
86 |
memory cost. But when forking, literally all data structures are |
87 |
copied - if the program frees them and replaces them by new data, |
88 |
the child processes will retain the old version even if it isn't |
89 |
used, which can suddenly and unexpectedly increase memory usage when |
90 |
freeing memory. |
91 |
|
92 |
The trade-off is between more sharing with fork (which can be good |
93 |
or bad), and no sharing with exec. |
94 |
|
95 |
This module allows the main program to do a controlled fork, and |
96 |
allows modules to exec processes safely at any time. When creating a |
97 |
custom process pool you can take advantage of data sharing via fork |
98 |
without risking to share large dynamic data structures that will |
99 |
blow up child memory usage. |
100 |
|
101 |
In other words, this module puts you into control over what is being |
102 |
shared and what isn't, at all times. |
103 |
|
104 |
Exec'ing a new perl process might be difficult. |
105 |
For example, it is not easy to find the correct path to the perl |
106 |
interpreter - $^X might not be a perl interpreter at all. |
107 |
|
108 |
This module tries hard to identify the correct path to the perl |
109 |
interpreter. With a cooperative main program, exec'ing the |
110 |
interpreter might not even be necessary, but even without help from |
111 |
the main program, it will still work when used from a module. |
112 |
|
113 |
Exec'ing a new perl process might be slow, as all necessary modules have |
114 |
to be loaded from disk again, with no guarantees of success. |
115 |
Long running processes might run into problems when perl is upgraded |
116 |
and modules are no longer loadable because they refer to a different |
117 |
perl version, or parts of a distribution are newer than the ones |
118 |
already loaded. |
119 |
|
120 |
This module supports creating pre-initialised perl processes to be |
121 |
used as a template for new processes. |
122 |
|
123 |
Forking might be impossible when a program is running. |
124 |
For example, POSIX makes it almost impossible to fork from a |
125 |
multi-threaded program while doing anything useful in the child - in |
126 |
fact, if your perl program uses POSIX threads (even indirectly via |
127 |
e.g. IO::AIO or threads), you cannot call fork on the perl level |
128 |
anymore without risking corruption issues on a number of operating |
129 |
systems. |
130 |
|
131 |
This module can safely fork helper processes at any time, by calling |
132 |
fork+exec in C, in a POSIX-compatible way (via Proc::FastSpawn). |
133 |
|
134 |
Parallel processing with fork might be inconvenient or difficult to |
135 |
implement. Modules might not work in both parent and child. |
136 |
For example, when a program uses an event loop and creates watchers |
137 |
it becomes very hard to use the event loop from a child program, as |
138 |
the watchers already exist but are only meaningful in the parent. |
139 |
Worse, a module might want to use such a module, not knowing whether |
140 |
another module or the main program also does, leading to problems. |
141 |
|
142 |
Apart from event loops, graphical toolkits also commonly fall into |
143 |
the "unsafe module" category, or just about anything that |
144 |
communicates with the external world, such as network libraries and |
145 |
file I/O modules, which usually don't like being copied and then |
146 |
allowed to continue in two processes. |
147 |
|
148 |
With this module only the main program is allowed to create new |
149 |
processes by forking (because only the main program can know when it |
150 |
is still safe to do so) - all other processes are created via |
151 |
fork+exec, which makes it possible to use modules such as event |
152 |
loops or window interfaces safely. |
153 |
|
154 |
EXAMPLES |
155 |
Create a single new process, tell it to run your worker function. |
156 |
AnyEvent::Fork |
157 |
->new |
158 |
->require ("MyModule") |
159 |
->run ("MyModule::worker, sub { |
160 |
my ($master_filehandle) = @_; |
161 |
|
162 |
# now $master_filehandle is connected to the |
163 |
# $slave_filehandle in the new process. |
164 |
}); |
165 |
|
166 |
"MyModule" might look like this: |
167 |
|
168 |
package MyModule; |
169 |
|
170 |
sub worker { |
171 |
my ($slave_filehandle) = @_; |
172 |
|
173 |
# now $slave_filehandle is connected to the $master_filehandle |
174 |
# in the original prorcess. have fun! |
175 |
} |
176 |
|
177 |
Create a pool of server processes all accepting on the same socket. |
178 |
# create listener socket |
179 |
my $listener = ...; |
180 |
|
181 |
# create a pool template, initialise it and give it the socket |
182 |
my $pool = AnyEvent::Fork |
183 |
->new |
184 |
->require ("Some::Stuff", "My::Server") |
185 |
->send_fh ($listener); |
186 |
|
187 |
# now create 10 identical workers |
188 |
for my $id (1..10) { |
189 |
$pool |
190 |
->fork |
191 |
->send_arg ($id) |
192 |
->run ("My::Server::run"); |
193 |
} |
194 |
|
195 |
# now do other things - maybe use the filehandle provided by run |
196 |
# to wait for the processes to die. or whatever. |
197 |
|
198 |
"My::Server" might look like this: |
199 |
|
200 |
package My::Server; |
201 |
|
202 |
sub run { |
203 |
my ($slave, $listener, $id) = @_; |
204 |
|
205 |
close $slave; # we do not use the socket, so close it to save resources |
206 |
|
207 |
# we could go ballistic and use e.g. AnyEvent here, or IO::AIO, |
208 |
# or anything we usually couldn't do in a process forked normally. |
209 |
while (my $socket = $listener->accept) { |
210 |
# do sth. with new socket |
211 |
} |
212 |
} |
213 |
|
214 |
use AnyEvent::Fork as a faster fork+exec |
215 |
This runs "/bin/echo hi", with standard output redirected to /tmp/log |
216 |
and standard error redirected to the communications socket. It is |
217 |
usually faster than fork+exec, but still lets you prepare the |
218 |
environment. |
219 |
|
220 |
open my $output, ">/tmp/log" or die "$!"; |
221 |
|
222 |
AnyEvent::Fork |
223 |
->new |
224 |
->eval (' |
225 |
# compile a helper function for later use |
226 |
sub run { |
227 |
my ($fh, $output, @cmd) = @_; |
228 |
|
229 |
# perl will clear close-on-exec on STDOUT/STDERR |
230 |
open STDOUT, ">&", $output or die; |
231 |
open STDERR, ">&", $fh or die; |
232 |
|
233 |
exec @cmd; |
234 |
} |
235 |
') |
236 |
->send_fh ($output) |
237 |
->send_arg ("/bin/echo", "hi") |
238 |
->run ("run", my $cv = AE::cv); |
239 |
|
240 |
my $stderr = $cv->recv; |
241 |
|
242 |
For stingy users: put the worker code into a "DATA" section. |
243 |
When you want to be stingy with files, you cna put your code into the |
244 |
"DATA" section of your module (or program): |
245 |
|
246 |
use AnyEvent::Fork; |
247 |
|
248 |
AnyEvent::Fork |
249 |
->new |
250 |
->eval (do { local $/; <DATA> }) |
251 |
->run ("doit", sub { ... }); |
252 |
|
253 |
__DATA__ |
254 |
|
255 |
sub doit { |
256 |
... do something! |
257 |
} |
258 |
|
259 |
For stingy standalone programs: do not rely on external files at |
260 |
all. |
261 |
For single-file scripts it can be inconvenient to rely on external files |
262 |
- even when using < "DATA" section, you still need to "exec" an external |
263 |
perl interpreter, which might not be available when using |
264 |
App::Staticperl, Urlader or PAR::Packer for example. |
265 |
|
266 |
Two modules help here - AnyEvent::Fork::Early forks a template process |
267 |
for all further calls to "new_exec", and AnyEvent::Fork::Template forks |
268 |
the main program as a template process. |
269 |
|
270 |
Here is how your main program should look like: |
271 |
|
272 |
#! perl |
273 |
|
274 |
# optional, as the very first thing. |
275 |
# in case modules want to create their own processes. |
276 |
use AnyEvent::Fork::Early; |
277 |
|
278 |
# next, load all modules you need in your template process |
279 |
use Example::My::Module |
280 |
use Example::Whatever; |
281 |
|
282 |
# next, put your run function definition and anything else you |
283 |
# need, but do not use code outside of BEGIN blocks. |
284 |
sub worker_run { |
285 |
my ($fh, @args) = @_; |
286 |
... |
287 |
} |
288 |
|
289 |
# now preserve everything so far as AnyEvent::Fork object |
290 |
# in §TEMPLATE. |
291 |
use AnyEvent::Fork::Template; |
292 |
|
293 |
# do not put code outside of BEGIN blocks until here |
294 |
|
295 |
# now use the $TEMPLATE process in any way you like |
296 |
|
297 |
# for example: create 10 worker processes |
298 |
my @worker; |
299 |
my $cv = AE::cv; |
300 |
for (1..10) { |
301 |
$cv->begin; |
302 |
$TEMPLATE->fork->send_arg ($_)->run ("worker_run", sub { |
303 |
push @worker, shift; |
304 |
$cv->end; |
305 |
}); |
306 |
} |
307 |
$cv->recv; |
308 |
|
309 |
lhead1 CONCEPTS |
310 |
|
311 |
This module can create new processes either by executing a new perl |
312 |
process, or by forking from an existing "template" process. |
313 |
|
314 |
All these processes are called "child processes" (whether they are |
315 |
direct children or not), while the process that manages them is called |
316 |
the "parent process". |
317 |
|
318 |
Each such process comes with its own file handle that can be used to |
319 |
communicate with it (it's actually a socket - one end in the new |
320 |
process, one end in the main process), and among the things you can do |
321 |
in it are load modules, fork new processes, send file handles to it, and |
322 |
execute functions. |
323 |
|
324 |
There are multiple ways to create additional processes to execute some |
325 |
jobs: |
326 |
|
327 |
fork a new process from the "default" template process, load code, run |
328 |
it |
329 |
This module has a "default" template process which it executes when |
330 |
it is needed the first time. Forking from this process shares the |
331 |
memory used for the perl interpreter with the new process, but |
332 |
loading modules takes time, and the memory is not shared with |
333 |
anything else. |
334 |
|
335 |
This is ideal for when you only need one extra process of a kind, |
336 |
with the option of starting and stopping it on demand. |
337 |
|
338 |
Example: |
339 |
|
340 |
AnyEvent::Fork |
341 |
->new |
342 |
->require ("Some::Module") |
343 |
->run ("Some::Module::run", sub { |
344 |
my ($fork_fh) = @_; |
345 |
}); |
346 |
|
347 |
fork a new template process, load code, then fork processes off of it |
348 |
and run the code |
349 |
When you need to have a bunch of processes that all execute the same |
350 |
(or very similar) tasks, then a good way is to create a new template |
351 |
process for them, loading all the modules you need, and then create |
352 |
your worker processes from this new template process. |
353 |
|
354 |
This way, all code (and data structures) that can be shared (e.g. |
355 |
the modules you loaded) is shared between the processes, and each |
356 |
new process consumes relatively little memory of its own. |
357 |
|
358 |
The disadvantage of this approach is that you need to create a |
359 |
template process for the sole purpose of forking new processes from |
360 |
it, but if you only need a fixed number of processes you can create |
361 |
them, and then destroy the template process. |
362 |
|
363 |
Example: |
364 |
|
365 |
my $template = AnyEvent::Fork->new->require ("Some::Module"); |
366 |
|
367 |
for (1..10) { |
368 |
$template->fork->run ("Some::Module::run", sub { |
369 |
my ($fork_fh) = @_; |
370 |
}); |
371 |
} |
372 |
|
373 |
# at this point, you can keep $template around to fork new processes |
374 |
# later, or you can destroy it, which causes it to vanish. |
375 |
|
376 |
execute a new perl interpreter, load some code, run it |
377 |
This is relatively slow, and doesn't allow you to share memory |
378 |
between multiple processes. |
379 |
|
380 |
The only advantage is that you don't have to have a template process |
381 |
hanging around all the time to fork off some new processes, which |
382 |
might be an advantage when there are long time spans where no extra |
383 |
processes are needed. |
384 |
|
385 |
Example: |
386 |
|
387 |
AnyEvent::Fork |
388 |
->new_exec |
389 |
->require ("Some::Module") |
390 |
->run ("Some::Module::run", sub { |
391 |
my ($fork_fh) = @_; |
392 |
}); |
393 |
|
394 |
THE "AnyEvent::Fork" CLASS |
395 |
This module exports nothing, and only implements a single class - |
396 |
"AnyEvent::Fork". |
397 |
|
398 |
There are two class constructors that both create new processes - "new" |
399 |
and "new_exec". The "fork" method creates a new process by forking an |
400 |
existing one and could be considered a third constructor. |
401 |
|
402 |
Most of the remaining methods deal with preparing the new process, by |
403 |
loading code, evaluating code and sending data to the new process. They |
404 |
usually return the process object, so you can chain method calls. |
405 |
|
406 |
If a process object is destroyed before calling its "run" method, then |
407 |
the process simply exits. After "run" is called, all responsibility is |
408 |
passed to the specified function. |
409 |
|
410 |
As long as there is any outstanding work to be done, process objects |
411 |
resist being destroyed, so there is no reason to store them unless you |
412 |
need them later - configure and forget works just fine. |
413 |
|
414 |
my $proc = new AnyEvent::Fork |
415 |
Create a new "empty" perl interpreter process and returns its |
416 |
process object for further manipulation. |
417 |
|
418 |
The new process is forked from a template process that is kept |
419 |
around for this purpose. When it doesn't exist yet, it is created by |
420 |
a call to "new_exec" first and then stays around for future calls. |
421 |
|
422 |
$new_proc = $proc->fork |
423 |
Forks $proc, creating a new process, and returns the process object |
424 |
of the new process. |
425 |
|
426 |
If any of the "send_" functions have been called before fork, then |
427 |
they will be cloned in the child. For example, in a pre-forked |
428 |
server, you might "send_fh" the listening socket into the template |
429 |
process, and then keep calling "fork" and "run". |
430 |
|
431 |
my $proc = new_exec AnyEvent::Fork |
432 |
Create a new "empty" perl interpreter process and returns its |
433 |
process object for further manipulation. |
434 |
|
435 |
Unlike the "new" method, this method *always* spawns a new perl |
436 |
process (except in some cases, see AnyEvent::Fork::Early for |
437 |
details). This reduces the amount of memory sharing that is |
438 |
possible, and is also slower. |
439 |
|
440 |
You should use "new" whenever possible, except when having a |
441 |
template process around is unacceptable. |
442 |
|
443 |
The path to the perl interpreter is divined using various methods - |
444 |
first $^X is investigated to see if the path ends with something |
445 |
that sounds as if it were the perl interpreter. Failing this, the |
446 |
module falls back to using $Config::Config{perlpath}. |
447 |
|
448 |
$pid = $proc->pid |
449 |
Returns the process id of the process *iff it is a direct child of |
450 |
the process running AnyEvent::Fork*, and "undef" otherwise. |
451 |
|
452 |
Normally, only processes created via "AnyEvent::Fork->new_exec" and |
453 |
AnyEvent::Fork::Template are direct children, and you are |
454 |
responsible to clean up their zombies when they die. |
455 |
|
456 |
All other processes are not direct children, and will be cleaned up |
457 |
by AnyEvent::Fork itself. |
458 |
|
459 |
$proc = $proc->eval ($perlcode, @args) |
460 |
Evaluates the given $perlcode as ... Perl code, while setting @_ to |
461 |
the strings specified by @args, in the "main" package. |
462 |
|
463 |
This call is meant to do any custom initialisation that might be |
464 |
required (for example, the "require" method uses it). It's not |
465 |
supposed to be used to completely take over the process, use "run" |
466 |
for that. |
467 |
|
468 |
The code will usually be executed after this call returns, and there |
469 |
is no way to pass anything back to the calling process. Any |
470 |
evaluation errors will be reported to stderr and cause the process |
471 |
to exit. |
472 |
|
473 |
If you want to execute some code (that isn't in a module) to take |
474 |
over the process, you should compile a function via "eval" first, |
475 |
and then call it via "run". This also gives you access to any |
476 |
arguments passed via the "send_xxx" methods, such as file handles. |
477 |
See the "use AnyEvent::Fork as a faster fork+exec" example to see it |
478 |
in action. |
479 |
|
480 |
Returns the process object for easy chaining of method calls. |
481 |
|
482 |
$proc = $proc->require ($module, ...) |
483 |
Tries to load the given module(s) into the process |
484 |
|
485 |
Returns the process object for easy chaining of method calls. |
486 |
|
487 |
$proc = $proc->send_fh ($handle, ...) |
488 |
Send one or more file handles (*not* file descriptors) to the |
489 |
process, to prepare a call to "run". |
490 |
|
491 |
The process object keeps a reference to the handles until they have |
492 |
been passed over to the process, so you must not explicitly close |
493 |
the handles. This is most easily accomplished by simply not storing |
494 |
the file handles anywhere after passing them to this method - when |
495 |
AnyEvent::Fork is finished using them, perl will automatically close |
496 |
them. |
497 |
|
498 |
Returns the process object for easy chaining of method calls. |
499 |
|
500 |
Example: pass a file handle to a process, and release it without |
501 |
closing. It will be closed automatically when it is no longer used. |
502 |
|
503 |
$proc->send_fh ($my_fh); |
504 |
undef $my_fh; # free the reference if you want, but DO NOT CLOSE IT |
505 |
|
506 |
$proc = $proc->send_arg ($string, ...) |
507 |
Send one or more argument strings to the process, to prepare a call |
508 |
to "run". The strings can be any octet strings. |
509 |
|
510 |
The protocol is optimised to pass a moderate number of relatively |
511 |
short strings - while you can pass up to 4GB of data in one go, this |
512 |
is more meant to pass some ID information or other startup info, not |
513 |
big chunks of data. |
514 |
|
515 |
Returns the process object for easy chaining of method calls. |
516 |
|
517 |
$proc->run ($func, $cb->($fh)) |
518 |
Enter the function specified by the function name in $func in the |
519 |
process. The function is called with the communication socket as |
520 |
first argument, followed by all file handles and string arguments |
521 |
sent earlier via "send_fh" and "send_arg" methods, in the order they |
522 |
were called. |
523 |
|
524 |
The process object becomes unusable on return from this function - |
525 |
any further method calls result in undefined behaviour. |
526 |
|
527 |
The function name should be fully qualified, but if it isn't, it |
528 |
will be looked up in the "main" package. |
529 |
|
530 |
If the called function returns, doesn't exist, or any error occurs, |
531 |
the process exits. |
532 |
|
533 |
Preparing the process is done in the background - when all commands |
534 |
have been sent, the callback is invoked with the local |
535 |
communications socket as argument. At this point you can start using |
536 |
the socket in any way you like. |
537 |
|
538 |
If the communication socket isn't used, it should be closed on both |
539 |
sides, to save on kernel memory. |
540 |
|
541 |
The socket is non-blocking in the parent, and blocking in the newly |
542 |
created process. The close-on-exec flag is set in both. |
543 |
|
544 |
Even if not used otherwise, the socket can be a good indicator for |
545 |
the existence of the process - if the other process exits, you get a |
546 |
readable event on it, because exiting the process closes the socket |
547 |
(if it didn't create any children using fork). |
548 |
|
549 |
Example: create a template for a process pool, pass a few strings, |
550 |
some file handles, then fork, pass one more string, and run some |
551 |
code. |
552 |
|
553 |
my $pool = AnyEvent::Fork |
554 |
->new |
555 |
->send_arg ("str1", "str2") |
556 |
->send_fh ($fh1, $fh2); |
557 |
|
558 |
for (1..2) { |
559 |
$pool |
560 |
->fork |
561 |
->send_arg ("str3") |
562 |
->run ("Some::function", sub { |
563 |
my ($fh) = @_; |
564 |
|
565 |
# fh is nonblocking, but we trust that the OS can accept these |
566 |
# few octets anyway. |
567 |
syswrite $fh, "hi #$_\n"; |
568 |
|
569 |
# $fh is being closed here, as we don't store it anywhere |
570 |
}); |
571 |
} |
572 |
|
573 |
# Some::function might look like this - all parameters passed before fork |
574 |
# and after will be passed, in order, after the communications socket. |
575 |
sub Some::function { |
576 |
my ($fh, $str1, $str2, $fh1, $fh2, $str3) = @_; |
577 |
|
578 |
print scalar <$fh>; # prints "hi #1\n" and "hi #2\n" in any order |
579 |
} |
580 |
|
581 |
$proc->to_fh ($cb->($fh)) # EXPERIMENTAL, MIGHT BE REMOVED |
582 |
Flushes all commands out to the process and then calls the callback |
583 |
with the communications socket. |
584 |
|
585 |
The process object becomes unusable on return from this function - |
586 |
any further method calls result in undefined behaviour. |
587 |
|
588 |
The point of this method is to give you a file handle thta you cna |
589 |
pass to another process. In that other process, you can call |
590 |
"new_from_fh AnyEvent::Fork" to create a new "AnyEvent::Fork" object |
591 |
from it, thereby effectively passing a fork object to another |
592 |
process. |
593 |
|
594 |
new_from_fh AnyEvent::Fork $fh # EXPERIMENTAL, MIGHT BE REMOVED |
595 |
Takes a file handle originally rceeived by the "to_fh" method and |
596 |
creates a new "AnyEvent:Fork" object. The child process itself will |
597 |
not change in any way, i.e. it will keep all the modifications done |
598 |
to it before calling "to_fh". |
599 |
|
600 |
The new object is very much like the original object, except that |
601 |
the "pid" method will return "undef" even if the process is a direct |
602 |
child. |
603 |
|
604 |
PERFORMANCE |
605 |
Now for some unscientific benchmark numbers (all done on an amd64 |
606 |
GNU/Linux box). These are intended to give you an idea of the relative |
607 |
performance you can expect, they are not meant to be absolute |
608 |
performance numbers. |
609 |
|
610 |
OK, so, I ran a simple benchmark that creates a socket pair, forks, |
611 |
calls exit in the child and waits for the socket to close in the parent. |
612 |
I did load AnyEvent, EV and AnyEvent::Fork, for a total process size of |
613 |
5100kB. |
614 |
|
615 |
2079 new processes per second, using manual socketpair + fork |
616 |
|
617 |
Then I did the same thing, but instead of calling fork, I called |
618 |
AnyEvent::Fork->new->run ("CORE::exit") and then again waited for the |
619 |
socket from the child to close on exit. This does the same thing as |
620 |
manual socket pair + fork, except that what is forked is the template |
621 |
process (2440kB), and the socket needs to be passed to the server at the |
622 |
other end of the socket first. |
623 |
|
624 |
2307 new processes per second, using AnyEvent::Fork->new |
625 |
|
626 |
And finally, using "new_exec" instead "new", using vforks+execs to exec |
627 |
a new perl interpreter and compile the small server each time, I get: |
628 |
|
629 |
479 vfork+execs per second, using AnyEvent::Fork->new_exec |
630 |
|
631 |
So how can "AnyEvent->new" be faster than a standard fork, even though |
632 |
it uses the same operations, but adds a lot of overhead? |
633 |
|
634 |
The difference is simply the process size: forking the 5MB process takes |
635 |
so much longer than forking the 2.5MB template process that the extra |
636 |
overhead is canceled out. |
637 |
|
638 |
If the benchmark process grows, the normal fork becomes even slower: |
639 |
|
640 |
1340 new processes, manual fork of a 20MB process |
641 |
731 new processes, manual fork of a 200MB process |
642 |
235 new processes, manual fork of a 2000MB process |
643 |
|
644 |
What that means (to me) is that I can use this module without having a |
645 |
bad conscience because of the extra overhead required to start new |
646 |
processes. |
647 |
|
648 |
TYPICAL PROBLEMS |
649 |
This section lists typical problems that remain. I hope by recognising |
650 |
them, most can be avoided. |
651 |
|
652 |
leaked file descriptors for exec'ed processes |
653 |
POSIX systems inherit file descriptors by default when exec'ing a |
654 |
new process. While perl itself laudably sets the close-on-exec flags |
655 |
on new file handles, most C libraries don't care, and even if all |
656 |
cared, it's often not possible to set the flag in a race-free |
657 |
manner. |
658 |
|
659 |
That means some file descriptors can leak through. And since it |
660 |
isn't possible to know which file descriptors are "good" and |
661 |
"necessary" (or even to know which file descriptors are open), there |
662 |
is no good way to close the ones that might harm. |
663 |
|
664 |
As an example of what "harm" can be done consider a web server that |
665 |
accepts connections and afterwards some module uses AnyEvent::Fork |
666 |
for the first time, causing it to fork and exec a new process, which |
667 |
might inherit the network socket. When the server closes the socket, |
668 |
it is still open in the child (which doesn't even know that) and the |
669 |
client might conclude that the connection is still fine. |
670 |
|
671 |
For the main program, there are multiple remedies available - |
672 |
AnyEvent::Fork::Early is one, creating a process early and not using |
673 |
"new_exec" is another, as in both cases, the first process can be |
674 |
exec'ed well before many random file descriptors are open. |
675 |
|
676 |
In general, the solution for these kind of problems is to fix the |
677 |
libraries or the code that leaks those file descriptors. |
678 |
|
679 |
Fortunately, most of these leaked descriptors do no harm, other than |
680 |
sitting on some resources. |
681 |
|
682 |
leaked file descriptors for fork'ed processes |
683 |
Normally, AnyEvent::Fork does start new processes by exec'ing them, |
684 |
which closes file descriptors not marked for being inherited. |
685 |
|
686 |
However, AnyEvent::Fork::Early and AnyEvent::Fork::Template offer a |
687 |
way to create these processes by forking, and this leaks more file |
688 |
descriptors than exec'ing them, as there is no way to mark |
689 |
descriptors as "close on fork". |
690 |
|
691 |
An example would be modules like EV, IO::AIO or Gtk2. Both create |
692 |
pipes for internal uses, and Gtk2 might open a connection to the X |
693 |
server. EV and IO::AIO can deal with fork, but Gtk2 might have |
694 |
trouble with a fork. |
695 |
|
696 |
The solution is to either not load these modules before use'ing |
697 |
AnyEvent::Fork::Early or AnyEvent::Fork::Template, or to delay |
698 |
initialising them, for example, by calling "init Gtk2" manually. |
699 |
|
700 |
exiting calls object destructors |
701 |
This only applies to users of AnyEvent::Fork:Early and |
702 |
AnyEvent::Fork::Template, or when initialising code creates objects |
703 |
that reference external resources. |
704 |
|
705 |
When a process created by AnyEvent::Fork exits, it might do so by |
706 |
calling exit, or simply letting perl reach the end of the program. |
707 |
At which point Perl runs all destructors. |
708 |
|
709 |
Not all destructors are fork-safe - for example, an object that |
710 |
represents the connection to an X display might tell the X server to |
711 |
free resources, which is inconvenient when the "real" object in the |
712 |
parent still needs to use them. |
713 |
|
714 |
This is obviously not a problem for AnyEvent::Fork::Early, as you |
715 |
used it as the very first thing, right? |
716 |
|
717 |
It is a problem for AnyEvent::Fork::Template though - and the |
718 |
solution is to not create objects with nontrivial destructors that |
719 |
might have an effect outside of Perl. |
720 |
|
721 |
PORTABILITY NOTES |
722 |
Native win32 perls are somewhat supported (AnyEvent::Fork::Early is a |
723 |
nop, and ::Template is not going to work), and it cost a lot of blood |
724 |
and sweat to make it so, mostly due to the bloody broken perl that |
725 |
nobody seems to care about. The fork emulation is a bad joke - I have |
726 |
yet to see something useful that you can do with it without running into |
727 |
memory corruption issues or other braindamage. Hrrrr. |
728 |
|
729 |
Since fork is endlessly broken on win32 perls (it doesn't even remotely |
730 |
work within it's documented limits) and quite obviously it's not getting |
731 |
improved any time soon, the best way to proceed on windows would be to |
732 |
always use "new_exec" and thus never rely on perl's fork "emulation". |
733 |
|
734 |
Cygwin perl is not supported at the moment due to some hilarious |
735 |
shortcomings of its API - see IO::FDPoll for more details. If you never |
736 |
use "send_fh" and always use "new_exec" to create processes, it should |
737 |
work though. |
738 |
|
739 |
SEE ALSO |
740 |
AnyEvent::Fork::Early, to avoid executing a perl interpreter at all |
741 |
(part of this distribution). |
742 |
|
743 |
AnyEvent::Fork::Template, to create a process by forking the main |
744 |
program at a convenient time (part of this distribution). |
745 |
|
746 |
AnyEvent::Fork::RPC, for simple RPC to child processes (on CPAN). |
747 |
|
748 |
AnyEvent::Fork::Pool, for simple worker process pool (on CPAN). |
749 |
|
750 |
AUTHOR AND CONTACT INFORMATION |
751 |
Marc Lehmann <schmorp@schmorp.de> |
752 |
http://software.schmorp.de/pkg/AnyEvent-Fork |
753 |
|