… | |
… | |
40 | points in your program, so locking and parallel access are rarely an |
40 | points in your program, so locking and parallel access are rarely an |
41 | issue, making thread programming much safer and easier than using other |
41 | issue, making thread programming much safer and easier than using other |
42 | thread models. |
42 | thread models. |
43 | |
43 | |
44 | Unlike the so-called "Perl threads" (which are not actually real threads |
44 | Unlike the so-called "Perl threads" (which are not actually real threads |
45 | but only the windows process emulation ported to unix), Coro provides a |
45 | but only the windows process emulation (see section of same name for more |
|
|
46 | details) ported to unix, and as such act as processes), Coro provides |
46 | full shared address space, which makes communication between threads |
47 | a full shared address space, which makes communication between threads |
47 | very easy. And threads are fast, too: disabling the Windows process |
48 | very easy. And Coro's threads are fast, too: disabling the Windows |
48 | emulation code in your perl and using Coro can easily result in a two to |
49 | process emulation code in your perl and using Coro can easily result in |
49 | four times speed increase for your programs. |
50 | a two to four times speed increase for your programs. A parallel matrix |
|
|
51 | multiplication benchmark runs over 300 times faster on a single core than |
|
|
52 | perl's pseudo-threads on a quad core using all four cores. |
50 | |
53 | |
51 | Coro achieves that by supporting multiple running interpreters that share |
54 | Coro achieves that by supporting multiple running interpreters that share |
52 | data, which is especially useful to code pseudo-parallel processes and |
55 | data, which is especially useful to code pseudo-parallel processes and |
53 | for event-based programming, such as multiple HTTP-GET requests running |
56 | for event-based programming, such as multiple HTTP-GET requests running |
54 | concurrently. See L<Coro::AnyEvent> to learn more on how to integrate Coro |
57 | concurrently. See L<Coro::AnyEvent> to learn more on how to integrate Coro |
55 | into an event-based environment. |
58 | into an event-based environment. |
56 | |
59 | |
57 | In this module, a thread is defined as "callchain + lexical variables + |
60 | In this module, a thread is defined as "callchain + lexical variables + |
58 | @_ + $_ + $@ + $/ + C stack), that is, a thread has its own callchain, |
61 | some package variables + C stack), that is, a thread has its own callchain, |
59 | its own set of lexicals and its own set of perls most important global |
62 | its own set of lexicals and its own set of perls most important global |
60 | variables (see L<Coro::State> for more configuration and background info). |
63 | variables (see L<Coro::State> for more configuration and background info). |
61 | |
64 | |
62 | See also the C<SEE ALSO> section at the end of this document - the Coro |
65 | See also the C<SEE ALSO> section at the end of this document - the Coro |
63 | module family is quite large. |
66 | module family is quite large. |
… | |
… | |
77 | |
80 | |
78 | our $idle; # idle handler |
81 | our $idle; # idle handler |
79 | our $main; # main coro |
82 | our $main; # main coro |
80 | our $current; # current coro |
83 | our $current; # current coro |
81 | |
84 | |
82 | our $VERSION = 5.13; |
85 | our $VERSION = 5.17; |
83 | |
86 | |
84 | our @EXPORT = qw(async async_pool cede schedule terminate current unblock_sub); |
87 | our @EXPORT = qw(async async_pool cede schedule terminate current unblock_sub); |
85 | our %EXPORT_TAGS = ( |
88 | our %EXPORT_TAGS = ( |
86 | prio => [qw(PRIO_MAX PRIO_HIGH PRIO_NORMAL PRIO_LOW PRIO_IDLE PRIO_MIN)], |
89 | prio => [qw(PRIO_MAX PRIO_HIGH PRIO_NORMAL PRIO_LOW PRIO_IDLE PRIO_MIN)], |
87 | ); |
90 | ); |
… | |
… | |
203 | Example: Create a new coro that just prints its arguments. |
206 | Example: Create a new coro that just prints its arguments. |
204 | |
207 | |
205 | async { |
208 | async { |
206 | print "@_\n"; |
209 | print "@_\n"; |
207 | } 1,2,3,4; |
210 | } 1,2,3,4; |
208 | |
|
|
209 | =cut |
|
|
210 | |
|
|
211 | sub async(&@) { |
|
|
212 | my $coro = new Coro @_; |
|
|
213 | $coro->ready; |
|
|
214 | $coro |
|
|
215 | } |
|
|
216 | |
211 | |
217 | =item async_pool { ... } [@args...] |
212 | =item async_pool { ... } [@args...] |
218 | |
213 | |
219 | Similar to C<async>, but uses a coro pool, so you should not call |
214 | Similar to C<async>, but uses a coro pool, so you should not call |
220 | terminate or join on it (although you are allowed to), and you get a |
215 | terminate or join on it (although you are allowed to), and you get a |
… | |
… | |
335 | |
330 | |
336 | These functions implement the same concept as C<dynamic-wind> in scheme |
331 | These functions implement the same concept as C<dynamic-wind> in scheme |
337 | does, and are useful when you want to localise some resource to a specific |
332 | does, and are useful when you want to localise some resource to a specific |
338 | coro. |
333 | coro. |
339 | |
334 | |
340 | They slow down coro switching considerably for coros that use |
335 | They slow down thread switching considerably for coros that use them |
341 | them (But coro switching is still reasonably fast if the handlers are |
336 | (about 40% for a BLOCK with a single assignment, so thread switching is |
342 | fast). |
337 | still reasonably fast if the handlers are fast). |
343 | |
338 | |
344 | These functions are best understood by an example: The following function |
339 | These functions are best understood by an example: The following function |
345 | will change the current timezone to "Antarctica/South_Pole", which |
340 | will change the current timezone to "Antarctica/South_Pole", which |
346 | requires a call to C<tzset>, but by using C<on_enter> and C<on_leave>, |
341 | requires a call to C<tzset>, but by using C<on_enter> and C<on_leave>, |
347 | which remember/change the current timezone and restore the previous |
342 | which remember/change the current timezone and restore the previous |
348 | value, respectively, the timezone is only changes for the coro that |
343 | value, respectively, the timezone is only changed for the coro that |
349 | installed those handlers. |
344 | installed those handlers. |
350 | |
345 | |
351 | use POSIX qw(tzset); |
346 | use POSIX qw(tzset); |
352 | |
347 | |
353 | async { |
348 | async { |
… | |
… | |
370 | }; |
365 | }; |
371 | |
366 | |
372 | This can be used to localise about any resource (locale, uid, current |
367 | This can be used to localise about any resource (locale, uid, current |
373 | working directory etc.) to a block, despite the existance of other |
368 | working directory etc.) to a block, despite the existance of other |
374 | coros. |
369 | coros. |
|
|
370 | |
|
|
371 | Another interesting example implements time-sliced multitasking using |
|
|
372 | interval timers (this could obviously be optimised, but does the job): |
|
|
373 | |
|
|
374 | # "timeslice" the given block |
|
|
375 | sub timeslice(&) { |
|
|
376 | use Time::HiRes (); |
|
|
377 | |
|
|
378 | Coro::on_enter { |
|
|
379 | # on entering the thread, we set an VTALRM handler to cede |
|
|
380 | $SIG{VTALRM} = sub { cede }; |
|
|
381 | # and then start the interval timer |
|
|
382 | Time::HiRes::setitimer &Time::HiRes::ITIMER_VIRTUAL, 0.01, 0.01; |
|
|
383 | }; |
|
|
384 | Coro::on_leave { |
|
|
385 | # on leaving the thread, we stop the interval timer again |
|
|
386 | Time::HiRes::setitimer &Time::HiRes::ITIMER_VIRTUAL, 0, 0; |
|
|
387 | }; |
|
|
388 | |
|
|
389 | &{+shift}; |
|
|
390 | } |
|
|
391 | |
|
|
392 | # use like this: |
|
|
393 | timeslice { |
|
|
394 | # The following is an endless loop that would normally |
|
|
395 | # monopolise the process. Since it runs in a timesliced |
|
|
396 | # environment, it will regularly cede to other threads. |
|
|
397 | while () { } |
|
|
398 | }; |
|
|
399 | |
375 | |
400 | |
376 | =item killall |
401 | =item killall |
377 | |
402 | |
378 | Kills/terminates/cancels all coros except the currently running one. |
403 | Kills/terminates/cancels all coros except the currently running one. |
379 | |
404 | |
… | |
… | |
423 | the ready queue, do nothing and return false. |
448 | the ready queue, do nothing and return false. |
424 | |
449 | |
425 | This ensures that the scheduler will resume this coro automatically |
450 | This ensures that the scheduler will resume this coro automatically |
426 | once all the coro of higher priority and all coro of the same |
451 | once all the coro of higher priority and all coro of the same |
427 | priority that were put into the ready queue earlier have been resumed. |
452 | priority that were put into the ready queue earlier have been resumed. |
|
|
453 | |
|
|
454 | =item $coro->suspend |
|
|
455 | |
|
|
456 | Suspends the specified coro. A suspended coro works just like any other |
|
|
457 | coro, except that the scheduler will not select a suspended coro for |
|
|
458 | execution. |
|
|
459 | |
|
|
460 | Suspending a coro can be useful when you want to keep the coro from |
|
|
461 | running, but you don't want to destroy it, or when you want to temporarily |
|
|
462 | freeze a coro (e.g. for debugging) to resume it later. |
|
|
463 | |
|
|
464 | A scenario for the former would be to suspend all (other) coros after a |
|
|
465 | fork and keep them alive, so their destructors aren't called, but new |
|
|
466 | coros can be created. |
|
|
467 | |
|
|
468 | =item $coro->resume |
|
|
469 | |
|
|
470 | If the specified coro was suspended, it will be resumed. Note that when |
|
|
471 | the coro was in the ready queue when it was suspended, it might have been |
|
|
472 | unreadied by the scheduler, so an activation might have been lost. |
|
|
473 | |
|
|
474 | To avoid this, it is best to put a suspended coro into the ready queue |
|
|
475 | unconditionally, as every synchronisation mechanism must protect itself |
|
|
476 | against spurious wakeups, and the one in the Coro family certainly do |
|
|
477 | that. |
428 | |
478 | |
429 | =item $is_ready = $coro->is_ready |
479 | =item $is_ready = $coro->is_ready |
430 | |
480 | |
431 | Returns true iff the Coro object is in the ready queue. Unless the Coro |
481 | Returns true iff the Coro object is in the ready queue. Unless the Coro |
432 | object gets destroyed, it will eventually be scheduled by the scheduler. |
482 | object gets destroyed, it will eventually be scheduled by the scheduler. |
… | |
… | |
693 | Wait for the specified rouse callback (or the last one that was created in |
743 | Wait for the specified rouse callback (or the last one that was created in |
694 | this coro). |
744 | this coro). |
695 | |
745 | |
696 | As soon as the callback is invoked (or when the callback was invoked |
746 | As soon as the callback is invoked (or when the callback was invoked |
697 | before C<rouse_wait>), it will return the arguments originally passed to |
747 | before C<rouse_wait>), it will return the arguments originally passed to |
698 | the rouse callback. |
748 | the rouse callback. In scalar context, that means you get the I<last> |
|
|
749 | argument, just as if C<rouse_wait> had a C<return ($a1, $a2, $a3...)> |
|
|
750 | statement at the end. |
699 | |
751 | |
700 | See the section B<HOW TO WAIT FOR A CALLBACK> for an actual usage example. |
752 | See the section B<HOW TO WAIT FOR A CALLBACK> for an actual usage example. |
701 | |
753 | |
702 | =back |
754 | =back |
703 | |
755 | |
… | |
… | |
802 | works. |
854 | works. |
803 | |
855 | |
804 | =back |
856 | =back |
805 | |
857 | |
806 | |
858 | |
|
|
859 | =head1 WINDOWS PROCESS EMULATION |
|
|
860 | |
|
|
861 | A great many people seem to be confused about ithreads (for example, Chip |
|
|
862 | Salzenberg called me unintelligent, incapable, stupid and gullible, |
|
|
863 | while in the same mail making rather confused statements about perl |
|
|
864 | ithreads (for example, that memory or files would be shared), showing his |
|
|
865 | lack of understanding of this area - if it is hard to understand for Chip, |
|
|
866 | it is probably not obvious to everybody). |
|
|
867 | |
|
|
868 | What follows is an ultra-condensed version of my talk about threads in |
|
|
869 | scripting languages given onthe perl workshop 2009: |
|
|
870 | |
|
|
871 | The so-called "ithreads" were originally implemented for two reasons: |
|
|
872 | first, to (badly) emulate unix processes on native win32 perls, and |
|
|
873 | secondly, to replace the older, real thread model ("5.005-threads"). |
|
|
874 | |
|
|
875 | It does that by using threads instead of OS processes. The difference |
|
|
876 | between processes and threads is that threads share memory (and other |
|
|
877 | state, such as files) between threads within a single process, while |
|
|
878 | processes do not share anything (at least not semantically). That |
|
|
879 | means that modifications done by one thread are seen by others, while |
|
|
880 | modifications by one process are not seen by other processes. |
|
|
881 | |
|
|
882 | The "ithreads" work exactly like that: when creating a new ithreads |
|
|
883 | process, all state is copied (memory is copied physically, files and code |
|
|
884 | is copied logically). Afterwards, it isolates all modifications. On UNIX, |
|
|
885 | the same behaviour can be achieved by using operating system processes, |
|
|
886 | except that UNIX typically uses hardware built into the system to do this |
|
|
887 | efficiently, while the windows process emulation emulates this hardware in |
|
|
888 | software (rather efficiently, but of course it is still much slower than |
|
|
889 | dedicated hardware). |
|
|
890 | |
|
|
891 | As mentioned before, loading code, modifying code, modifying data |
|
|
892 | structures and so on is only visible in the ithreads process doing the |
|
|
893 | modification, not in other ithread processes within the same OS process. |
|
|
894 | |
|
|
895 | This is why "ithreads" do not implement threads for perl at all, only |
|
|
896 | processes. What makes it so bad is that on non-windows platforms, you can |
|
|
897 | actually take advantage of custom hardware for this purpose (as evidenced |
|
|
898 | by the forks module, which gives you the (i-) threads API, just much |
|
|
899 | faster). |
|
|
900 | |
|
|
901 | Sharing data is in the i-threads model is done by transfering data |
|
|
902 | structures between threads using copying semantics, which is very slow - |
|
|
903 | shared data simply does not exist. Benchmarks using i-threads which are |
|
|
904 | communication-intensive show extremely bad behaviour with i-threads (in |
|
|
905 | fact, so bad that Coro, which cannot take direct advantage of multiple |
|
|
906 | CPUs, is often orders of magnitude faster because it shares data using |
|
|
907 | real threads, refer to my talk for details). |
|
|
908 | |
|
|
909 | As summary, i-threads *use* threads to implement processes, while |
|
|
910 | the compatible forks module *uses* processes to emulate, uhm, |
|
|
911 | processes. I-threads slow down every perl program when enabled, and |
|
|
912 | outside of windows, serve no (or little) practical purpose, but |
|
|
913 | disadvantages every single-threaded Perl program. |
|
|
914 | |
|
|
915 | This is the reason that I try to avoid the name "ithreads", as it is |
|
|
916 | misleading as it implies that it implements some kind of thread model for |
|
|
917 | perl, and prefer the name "windows process emulation", which describes the |
|
|
918 | actual use and behaviour of it much better. |
|
|
919 | |
807 | =head1 SEE ALSO |
920 | =head1 SEE ALSO |
808 | |
921 | |
809 | Event-Loop integration: L<Coro::AnyEvent>, L<Coro::EV>, L<Coro::Event>. |
922 | Event-Loop integration: L<Coro::AnyEvent>, L<Coro::EV>, L<Coro::Event>. |
810 | |
923 | |
811 | Debugging: L<Coro::Debug>. |
924 | Debugging: L<Coro::Debug>. |