… | |
… | |
70 | preserving most of the advantages of fork. |
70 | preserving most of the advantages of fork. |
71 | |
71 | |
72 | It can be used to create new worker processes or new independent |
72 | It can be used to create new worker processes or new independent |
73 | subprocesses for short- and long-running jobs, process pools (e.g. for use |
73 | subprocesses for short- and long-running jobs, process pools (e.g. for use |
74 | in pre-forked servers) but also to spawn new external processes (such as |
74 | in pre-forked servers) but also to spawn new external processes (such as |
75 | CGI scripts from a webserver), which can be faster (and more well behaved) |
75 | CGI scripts from a web server), which can be faster (and more well behaved) |
76 | than using fork+exec in big processes. |
76 | than using fork+exec in big processes. |
77 | |
77 | |
78 | Special care has been taken to make this module useful from other modules, |
78 | Special care has been taken to make this module useful from other modules, |
79 | while still supporting specialised environments such as L<App::Staticperl> |
79 | while still supporting specialised environments such as L<App::Staticperl> |
80 | or L<PAR::Packer>. |
80 | or L<PAR::Packer>. |
… | |
… | |
110 | or fork+exec instead. |
110 | or fork+exec instead. |
111 | |
111 | |
112 | =item Forking usually creates a copy-on-write copy of the parent |
112 | =item Forking usually creates a copy-on-write copy of the parent |
113 | process. Memory (for example, modules or data files that have been |
113 | process. Memory (for example, modules or data files that have been |
114 | will not take additional memory). When exec'ing a new process, modules |
114 | will not take additional memory). When exec'ing a new process, modules |
115 | and data files might need to be loaded again, at extra cpu and memory |
115 | and data files might need to be loaded again, at extra CPU and memory |
116 | cost. Likewise when forking, all data structures are copied as well - if |
116 | cost. Likewise when forking, all data structures are copied as well - if |
117 | the program frees them and replaces them by new data, the child processes |
117 | the program frees them and replaces them by new data, the child processes |
118 | will retain the memory even if it isn't used. |
118 | will retain the memory even if it isn't used. |
119 | |
119 | |
120 | This module allows the main program to do a controlled fork, and allows |
120 | This module allows the main program to do a controlled fork, and allows |
… | |
… | |
132 | as template, and also tries hard to identify the correct path to the perl |
132 | as template, and also tries hard to identify the correct path to the perl |
133 | interpreter. With a cooperative main program, exec'ing the interpreter |
133 | interpreter. With a cooperative main program, exec'ing the interpreter |
134 | might not even be necessary. |
134 | might not even be necessary. |
135 | |
135 | |
136 | =item Forking might be impossible when a program is running. For example, |
136 | =item Forking might be impossible when a program is running. For example, |
137 | POSIX makes it almost impossible to fork from a multithreaded program and |
137 | POSIX makes it almost impossible to fork from a multi-threaded program and |
138 | do anything useful in the child - strictly speaking, if your perl program |
138 | do anything useful in the child - strictly speaking, if your perl program |
139 | uses posix threads (even indirectly via e.g. L<IO::AIO> or L<threads>), |
139 | uses posix threads (even indirectly via e.g. L<IO::AIO> or L<threads>), |
140 | you cannot call fork on the perl level anymore, at all. |
140 | you cannot call fork on the perl level anymore, at all. |
141 | |
141 | |
142 | This module can safely fork helper processes at any time, by caling |
142 | This module can safely fork helper processes at any time, by calling |
143 | fork+exec in C, in a POSIX-compatible way. |
143 | fork+exec in C, in a POSIX-compatible way. |
144 | |
144 | |
145 | =item Parallel processing with fork might be inconvenient or difficult |
145 | =item Parallel processing with fork might be inconvenient or difficult |
146 | to implement. For example, when a program uses an event loop and creates |
146 | to implement. For example, when a program uses an event loop and creates |
147 | watchers it becomes very hard to use the event loop from a child |
147 | watchers it becomes very hard to use the event loop from a child |
… | |
… | |
179 | needed the first time. Forking from this process shares the memory used |
179 | needed the first time. Forking from this process shares the memory used |
180 | for the perl interpreter with the new process, but loading modules takes |
180 | for the perl interpreter with the new process, but loading modules takes |
181 | time, and the memory is not shared with anything else. |
181 | time, and the memory is not shared with anything else. |
182 | |
182 | |
183 | This is ideal for when you only need one extra process of a kind, with the |
183 | This is ideal for when you only need one extra process of a kind, with the |
184 | option of starting and stipping it on demand. |
184 | option of starting and stopping it on demand. |
185 | |
185 | |
186 | Example: |
186 | Example: |
187 | |
187 | |
188 | AnyEvent::Fork |
188 | AnyEvent::Fork |
189 | ->new |
189 | ->new |
… | |
… | |
204 | modules you loaded) is shared between the processes, and each new process |
204 | modules you loaded) is shared between the processes, and each new process |
205 | consumes relatively little memory of its own. |
205 | consumes relatively little memory of its own. |
206 | |
206 | |
207 | The disadvantage of this approach is that you need to create a template |
207 | The disadvantage of this approach is that you need to create a template |
208 | process for the sole purpose of forking new processes from it, but if you |
208 | process for the sole purpose of forking new processes from it, but if you |
209 | only need a fixed number of proceses you can create them, and then destroy |
209 | only need a fixed number of processes you can create them, and then destroy |
210 | the template process. |
210 | the template process. |
211 | |
211 | |
212 | Example: |
212 | Example: |
213 | |
213 | |
214 | my $template = AnyEvent::Fork->new->require ("Some::Module"); |
214 | my $template = AnyEvent::Fork->new->require ("Some::Module"); |
… | |
… | |
251 | |
251 | |
252 | package AnyEvent::Fork; |
252 | package AnyEvent::Fork; |
253 | |
253 | |
254 | use common::sense; |
254 | use common::sense; |
255 | |
255 | |
256 | use Socket (); |
256 | use Errno (); |
257 | |
257 | |
258 | use AnyEvent; |
258 | use AnyEvent; |
259 | use AnyEvent::Util (); |
259 | use AnyEvent::Util (); |
260 | |
260 | |
261 | use IO::FDPass; |
261 | use IO::FDPass; |
… | |
… | |
281 | our $TEMPLATE; |
281 | our $TEMPLATE; |
282 | |
282 | |
283 | sub _cmd { |
283 | sub _cmd { |
284 | my $self = shift; |
284 | my $self = shift; |
285 | |
285 | |
286 | #TODO: maybe append the packet to any existing string command already in the queue |
|
|
287 | |
|
|
288 | # ideally, we would want to use "a (w/a)*" as format string, but perl versions |
286 | # ideally, we would want to use "a (w/a)*" as format string, but perl |
289 | # from at least 5.8.9 to 5.16.3 are all buggy and can't unpack it. |
287 | # versions from at least 5.8.9 to 5.16.3 are all buggy and can't unpack |
|
|
288 | # it. |
290 | push @{ $self->[2] }, pack "L/a*", pack "(w/a*)*", @_; |
289 | push @{ $self->[2] }, pack "L/a*", pack "(w/a*)*", @_; |
291 | |
290 | |
292 | $self->[3] ||= AE::io $self->[1], 1, sub { |
291 | unless ($self->[3]) { |
|
|
292 | my $wcb = sub { |
|
|
293 | do { |
293 | # send the next "thing" in the queue - either a reference to an fh, |
294 | # send the next "thing" in the queue - either a reference to an fh, |
294 | # or a plain string. |
295 | # or a plain string. |
295 | |
296 | |
296 | if (ref $self->[2][0]) { |
297 | if (ref $self->[2][0]) { |
297 | # send fh |
298 | # send fh |
298 | IO::FDPass::send fileno $self->[1], fileno ${ $self->[2][0] } |
299 | unless (IO::FDPass::send fileno $self->[1], fileno ${ $self->[2][0] }) { |
|
|
300 | return if $! == Errno::EAGAIN || $! == Errno::EWOULDBLOCK; |
|
|
301 | undef $self->[3]; |
|
|
302 | die "AnyEvent::Fork: file descriptor send failure: $!"; |
|
|
303 | } |
|
|
304 | |
299 | and shift @{ $self->[2] }; |
305 | shift @{ $self->[2] }; |
300 | |
306 | |
301 | } else { |
307 | } else { |
302 | # send string |
308 | # send string |
303 | my $len = syswrite $self->[1], $self->[2][0] |
309 | my $len = syswrite $self->[1], $self->[2][0]; |
|
|
310 | |
|
|
311 | unless ($len) { |
|
|
312 | return if $! == Errno::EAGAIN || $! == Errno::EWOULDBLOCK; |
|
|
313 | undef $self->[3]; |
304 | or do { undef $self->[3]; die "AnyEvent::Fork: command write failure: $!" }; |
314 | die "AnyEvent::Fork: command write failure: $!"; |
|
|
315 | } |
305 | |
316 | |
306 | substr $self->[2][0], 0, $len, ""; |
317 | substr $self->[2][0], 0, $len, ""; |
307 | shift @{ $self->[2] } unless length $self->[2][0]; |
318 | shift @{ $self->[2] } unless length $self->[2][0]; |
308 | } |
319 | } |
|
|
320 | } while @{ $self->[2] }; |
309 | |
321 | |
310 | unless (@{ $self->[2] }) { |
322 | # everything written |
311 | undef $self->[3]; |
323 | undef $self->[3]; |
312 | # invoke run callback |
324 | # invoke run callback |
313 | $self->[0]->($self->[1]) if $self->[0]; |
325 | $self->[0]->($self->[1]) if $self->[0]; |
314 | } |
326 | }; |
|
|
327 | |
|
|
328 | $wcb->(); |
|
|
329 | |
|
|
330 | $self->[3] ||= AE::io $self->[1], 1, $wcb |
|
|
331 | if @{ $self->[2] }; |
315 | }; |
332 | } |
316 | |
333 | |
317 | () # make sure we don't leak the watcher |
334 | () # make sure we don't leak the watcher |
318 | } |
335 | } |
319 | |
336 | |
320 | sub _new { |
337 | sub _new { |
… | |
… | |
410 | reduces the amount of memory sharing that is possible, and is also slower. |
427 | reduces the amount of memory sharing that is possible, and is also slower. |
411 | |
428 | |
412 | You should use C<new> whenever possible, except when having a template |
429 | You should use C<new> whenever possible, except when having a template |
413 | process around is unacceptable. |
430 | process around is unacceptable. |
414 | |
431 | |
415 | The path to the perl interpreter is divined usign various methods - first |
432 | The path to the perl interpreter is divined using various methods - first |
416 | C<$^X> is investigated to see if the path ends with something that sounds |
433 | C<$^X> is investigated to see if the path ends with something that sounds |
417 | as if it were the perl interpreter. Failing this, the module falls back to |
434 | as if it were the perl interpreter. Failing this, the module falls back to |
418 | using C<$Config::Config{perlpath}>. |
435 | using C<$Config::Config{perlpath}>. |
419 | |
436 | |
420 | =cut |
437 | =cut |
… | |
… | |
515 | accomplished by simply not storing the file handles anywhere after passing |
532 | accomplished by simply not storing the file handles anywhere after passing |
516 | them to this method. |
533 | them to this method. |
517 | |
534 | |
518 | Returns the process object for easy chaining of method calls. |
535 | Returns the process object for easy chaining of method calls. |
519 | |
536 | |
520 | Example: pass an fh to a process, and release it without closing. it will |
537 | Example: pass a file handle to a process, and release it without |
521 | be closed automatically when it is no longer used. |
538 | closing. It will be closed automatically when it is no longer used. |
522 | |
539 | |
523 | $proc->send_fh ($my_fh); |
540 | $proc->send_fh ($my_fh); |
524 | undef $my_fh; # free the reference if you want, but DO NOT CLOSE IT |
541 | undef $my_fh; # free the reference if you want, but DO NOT CLOSE IT |
525 | |
542 | |
526 | =cut |
543 | =cut |
… | |
… | |
539 | =item $proc = $proc->send_arg ($string, ...) |
556 | =item $proc = $proc->send_arg ($string, ...) |
540 | |
557 | |
541 | Send one or more argument strings to the process, to prepare a call to |
558 | Send one or more argument strings to the process, to prepare a call to |
542 | C<run>. The strings can be any octet string. |
559 | C<run>. The strings can be any octet string. |
543 | |
560 | |
|
|
561 | The protocol is optimised to pass a moderate number of relatively short |
|
|
562 | strings - while you can pass up to 4GB of data in one go, this is more |
|
|
563 | meant to pass some ID information or other startup info, not big chunks of |
|
|
564 | data. |
|
|
565 | |
544 | Returns the process object for easy chaining of emthod calls. |
566 | Returns the process object for easy chaining of method calls. |
545 | |
567 | |
546 | =cut |
568 | =cut |
547 | |
569 | |
548 | sub send_arg { |
570 | sub send_arg { |
549 | my ($self, @arg) = @_; |
571 | my ($self, @arg) = @_; |
… | |
… | |
570 | If the communication socket isn't used, it should be closed on both sides, |
592 | If the communication socket isn't used, it should be closed on both sides, |
571 | to save on kernel memory. |
593 | to save on kernel memory. |
572 | |
594 | |
573 | The socket is non-blocking in the parent, and blocking in the newly |
595 | The socket is non-blocking in the parent, and blocking in the newly |
574 | created process. The close-on-exec flag is set on both. Even if not used |
596 | created process. The close-on-exec flag is set on both. Even if not used |
575 | otherwise, the socket can be a good indicator for the existance of the |
597 | otherwise, the socket can be a good indicator for the existence of the |
576 | process - if the other process exits, you get a readable event on it, |
598 | process - if the other process exits, you get a readable event on it, |
577 | because exiting the process closes the socket (if it didn't create any |
599 | because exiting the process closes the socket (if it didn't create any |
578 | children using fork). |
600 | children using fork). |
579 | |
601 | |
580 | Example: create a template for a process pool, pass a few strings, some |
602 | Example: create a template for a process pool, pass a few strings, some |
… | |
… | |
621 | |
643 | |
622 | =head1 PERFORMANCE |
644 | =head1 PERFORMANCE |
623 | |
645 | |
624 | Now for some unscientific benchmark numbers (all done on an amd64 |
646 | Now for some unscientific benchmark numbers (all done on an amd64 |
625 | GNU/Linux box). These are intended to give you an idea of the relative |
647 | GNU/Linux box). These are intended to give you an idea of the relative |
626 | performance you can expect. |
648 | performance you can expect, they are not meant to be absolute performance |
|
|
649 | numbers. |
627 | |
650 | |
628 | Ok, so, I ran a simple benchmark that creates a socketpair, forks, calls |
651 | OK, so, I ran a simple benchmark that creates a socket pair, forks, calls |
629 | exit in the child and waits for the socket to close in the parent. I did |
652 | exit in the child and waits for the socket to close in the parent. I did |
630 | load AnyEvent, EV and AnyEvent::Fork, for a total process size of 6312kB. |
653 | load AnyEvent, EV and AnyEvent::Fork, for a total process size of 5100kB. |
631 | |
654 | |
632 | 2079 new processes per second, using socketpair + fork manually |
655 | 2079 new processes per second, using manual socketpair + fork |
633 | |
656 | |
634 | Then I did the same thing, but instead of calling fork, I called |
657 | Then I did the same thing, but instead of calling fork, I called |
635 | AnyEvent::Fork->new->run ("CORE::exit") and then again waited for the |
658 | AnyEvent::Fork->new->run ("CORE::exit") and then again waited for the |
636 | socket form the child to close on exit. This does the same thing as manual |
659 | socket form the child to close on exit. This does the same thing as manual |
637 | socketpair + fork, except that what is forked is the template process |
660 | socket pair + fork, except that what is forked is the template process |
638 | (2440kB), and the socket needs to be passed to the server at the other end |
661 | (2440kB), and the socket needs to be passed to the server at the other end |
639 | of the socket first. |
662 | of the socket first. |
640 | |
663 | |
641 | 2307 new processes per second, using AnyEvent::Fork->new |
664 | 2307 new processes per second, using AnyEvent::Fork->new |
642 | |
665 | |
643 | And finally, using C<new_exec> instead C<new>, using vforks+execs to exec |
666 | And finally, using C<new_exec> instead C<new>, using vforks+execs to exec |
644 | a new perl interpreter and compile the small server each time, I get: |
667 | a new perl interpreter and compile the small server each time, I get: |
645 | |
668 | |
646 | 479 vfork+execs per second, using AnyEvent::Fork->new_exec |
669 | 479 vfork+execs per second, using AnyEvent::Fork->new_exec |
647 | |
670 | |
648 | So how can C<< AnyEvent->new >> be faster than a standard fork, een though |
671 | So how can C<< AnyEvent->new >> be faster than a standard fork, even |
649 | it uses the same operations, but adds a lot of overhead? |
672 | though it uses the same operations, but adds a lot of overhead? |
650 | |
673 | |
651 | The difference is simply the process size: forking the 6MB process takes |
674 | The difference is simply the process size: forking the 6MB process takes |
652 | so much longer than forking the 2.5MB template process that the overhead |
675 | so much longer than forking the 2.5MB template process that the overhead |
653 | introduced is canceled out. |
676 | introduced is canceled out. |
654 | |
677 | |
… | |
… | |
656 | |
679 | |
657 | 1340 new processes, manual fork in a 20MB process |
680 | 1340 new processes, manual fork in a 20MB process |
658 | 731 new processes, manual fork in a 200MB process |
681 | 731 new processes, manual fork in a 200MB process |
659 | 235 new processes, manual fork in a 2000MB process |
682 | 235 new processes, manual fork in a 2000MB process |
660 | |
683 | |
661 | What that means (to me) is that I can use this module without havign a |
684 | What that means (to me) is that I can use this module without having a |
662 | very bad conscience because of the extra overhead requried to strat new |
685 | very bad conscience because of the extra overhead required to start new |
663 | processes. |
686 | processes. |
664 | |
687 | |
665 | =head1 TYPICAL PROBLEMS |
688 | =head1 TYPICAL PROBLEMS |
666 | |
689 | |
667 | This section lists typical problems that remain. I hope by recognising |
690 | This section lists typical problems that remain. I hope by recognising |
668 | them, most can be avoided. |
691 | them, most can be avoided. |
669 | |
692 | |
670 | =over 4 |
693 | =over 4 |
|
|
694 | |
|
|
695 | =item exit runs destructors |
671 | |
696 | |
672 | =item "leaked" file descriptors for exec'ed processes |
697 | =item "leaked" file descriptors for exec'ed processes |
673 | |
698 | |
674 | POSIX systems inherit file descriptors by default when exec'ing a new |
699 | POSIX systems inherit file descriptors by default when exec'ing a new |
675 | process. While perl itself laudably sets the close-on-exec flags on new |
700 | process. While perl itself laudably sets the close-on-exec flags on new |
676 | file handles, most C libraries don't care, and even if all cared, it's |
701 | file handles, most C libraries don't care, and even if all cared, it's |
677 | often not possible to set the flag in a race-free manner. |
702 | often not possible to set the flag in a race-free manner. |
678 | |
703 | |
679 | That means some file descriptors can leak through. And since it isn't |
704 | That means some file descriptors can leak through. And since it isn't |
680 | possible to know which file descriptors are "good" and "neccessary" (or |
705 | possible to know which file descriptors are "good" and "necessary" (or |
681 | even to know which file descreiptors are open), there is no good way to |
706 | even to know which file descriptors are open), there is no good way to |
682 | close the ones that might harm. |
707 | close the ones that might harm. |
683 | |
708 | |
684 | As an example of what "harm" can be done consider a web server that |
709 | As an example of what "harm" can be done consider a web server that |
685 | accepts connections and afterwards some module uses AnyEvent::Fork for the |
710 | accepts connections and afterwards some module uses AnyEvent::Fork for the |
686 | first time, causing it to fork and exec a new process, which might inherit |
711 | first time, causing it to fork and exec a new process, which might inherit |
… | |
… | |
694 | well before many random file descriptors are open. |
719 | well before many random file descriptors are open. |
695 | |
720 | |
696 | In general, the solution for these kind of problems is to fix the |
721 | In general, the solution for these kind of problems is to fix the |
697 | libraries or the code that leaks those file descriptors. |
722 | libraries or the code that leaks those file descriptors. |
698 | |
723 | |
699 | Fortunately, most of these lekaed descriptors do no harm, other than |
724 | Fortunately, most of these leaked descriptors do no harm, other than |
700 | sitting on some resources. |
725 | sitting on some resources. |
701 | |
726 | |
702 | =item "leaked" file descriptors for fork'ed processes |
727 | =item "leaked" file descriptors for fork'ed processes |
703 | |
728 | |
704 | Normally, L<AnyEvent::Fork> does start new processes by exec'ing them, |
729 | Normally, L<AnyEvent::Fork> does start new processes by exec'ing them, |
… | |
… | |
724 | |
749 | |
725 | Native win32 perls are somewhat supported (AnyEvent::Fork::Early is a nop, |
750 | Native win32 perls are somewhat supported (AnyEvent::Fork::Early is a nop, |
726 | and ::Template is not going to work), and it cost a lot of blood and sweat |
751 | and ::Template is not going to work), and it cost a lot of blood and sweat |
727 | to make it so, mostly due to the bloody broken perl that nobody seems to |
752 | to make it so, mostly due to the bloody broken perl that nobody seems to |
728 | care about. The fork emulation is a bad joke - I have yet to see something |
753 | care about. The fork emulation is a bad joke - I have yet to see something |
729 | useful that you cna do with it without running into memory corruption |
754 | useful that you can do with it without running into memory corruption |
730 | issues or other braindamage. Hrrrr. |
755 | issues or other braindamage. Hrrrr. |
731 | |
756 | |
732 | Cygwin perl is not supported at the moment, as it should implement fd |
757 | Cygwin perl is not supported at the moment, as it should implement fd |
733 | passing, but doesn't, and rolling my own is hard, as cygwin doesn't |
758 | passing, but doesn't, and rolling my own is hard, as cygwin doesn't |
734 | support enough functionality to do it. |
759 | support enough functionality to do it. |