ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/IO-AIO/AIO.pm
(Generate patch)

Comparing IO-AIO/AIO.pm (file contents):
Revision 1.71 by root, Tue Oct 24 11:57:30 2006 UTC vs.
Revision 1.77 by root, Wed Oct 25 17:57:30 2006 UTC

51=head1 DESCRIPTION 51=head1 DESCRIPTION
52 52
53This module implements asynchronous I/O using whatever means your 53This module implements asynchronous I/O using whatever means your
54operating system supports. 54operating system supports.
55 55
56Currently, a number of threads are started that execute your read/writes 56In this version, a number of threads are started that execute your
57and signal their completion. You don't need thread support in perl, and 57requests and signal their completion. You don't need thread support
58the threads created by this module will not be visible to perl. In the 58in perl, and the threads created by this module will not be visible
59future, this module might make use of the native aio functions available 59to perl. In the future, this module might make use of the native aio
60on many operating systems. However, they are often not well-supported 60functions available on many operating systems. However, they are often
61(Linux doesn't allow them on normal files currently, for example), 61not well-supported or restricted (Linux doesn't allow them on normal
62and they would only support aio_read and aio_write, so the remaining 62files currently, for example), and they would only support aio_read and
63functionality would have to be implemented using threads anyway. 63aio_write, so the remaining functionality would have to be implemented
64using threads anyway.
64 65
65Although the module will work with in the presence of other threads, 66Although the module will work with in the presence of other (Perl-)
66it is currently not reentrant in any way, so use appropriate locking 67threads, it is currently not reentrant in any way, so use appropriate
67yourself, always call C<poll_cb> from within the same thread, or never 68locking yourself, always call C<poll_cb> from within the same thread, or
68call C<poll_cb> (or other C<aio_> functions) recursively. 69never call C<poll_cb> (or other C<aio_> functions) recursively.
70
71=head1 REQUEST ANATOMY AND LIFETIME
72
73Every C<aio_*> function creates a request. which is a C data structure not
74directly visible to Perl.
75
76If called in non-void context, every request function returns a Perl
77object representing the request. In void context, nothing is returned,
78which saves a bit of memory.
79
80The perl object is a fairly standard ref-to-hash object. The hash contents
81are not used by IO::AIO so you are free to store anything you like in it.
82
83During their existance, aio requests travel through the following states,
84in order:
85
86=over 4
87
88=item ready
89
90Immediately after a request is created it is put into the ready state,
91waiting for a thread to execute it.
92
93=item execute
94
95A thread has accepted the request for processing and is currently
96executing it (e.g. blocking in read).
97
98=item pending
99
100The request has been executed and is waiting for result processing.
101
102While request submission and execution is fully asynchronous, result
103processing is not and relies on the perl interpreter calling C<poll_cb>
104(or another function with the same effect).
105
106=item result
107
108The request results are processed synchronously by C<poll_cb>.
109
110The C<poll_cb> function will process all outstanding aio requests by
111calling their callbacks, freeing memory associated with them and managing
112any groups they are contained in.
113
114=item done
115
116Request has reached the end of its lifetime and holds no resources anymore
117(except possibly for the Perl object, but its connection to the actual
118aio request is severed and calling its methods will either do nothing or
119result in a runtime error).
69 120
70=cut 121=cut
71 122
72package IO::AIO; 123package IO::AIO;
73 124
83 aio_lstat aio_unlink aio_rmdir aio_readdir aio_scandir aio_symlink 134 aio_lstat aio_unlink aio_rmdir aio_readdir aio_scandir aio_symlink
84 aio_fsync aio_fdatasync aio_readahead aio_rename aio_link aio_move 135 aio_fsync aio_fdatasync aio_readahead aio_rename aio_link aio_move
85 aio_group aio_nop); 136 aio_group aio_nop);
86 our @EXPORT = (@AIO_REQ, qw(aioreq_pri aioreq_nice)); 137 our @EXPORT = (@AIO_REQ, qw(aioreq_pri aioreq_nice));
87 our @EXPORT_OK = qw(poll_fileno poll_cb poll_wait flush 138 our @EXPORT_OK = qw(poll_fileno poll_cb poll_wait flush
88 min_parallel max_parallel max_outstanding nreqs); 139 min_parallel max_parallel nreqs);
89 140
90 @IO::AIO::GRP::ISA = 'IO::AIO::REQ'; 141 @IO::AIO::GRP::ISA = 'IO::AIO::REQ';
91 142
92 require XSLoader; 143 require XSLoader;
93 XSLoader::load ("IO::AIO", $VERSION); 144 XSLoader::load ("IO::AIO", $VERSION);
375with the filenames. 426with the filenames.
376 427
377=item aio_scandir $path, $maxreq, $callback->($dirs, $nondirs) 428=item aio_scandir $path, $maxreq, $callback->($dirs, $nondirs)
378 429
379Scans a directory (similar to C<aio_readdir>) but additionally tries to 430Scans a directory (similar to C<aio_readdir>) but additionally tries to
380separate the entries of directory C<$path> into two sets of names, ones 431efficiently separate the entries of directory C<$path> into two sets of
381you can recurse into (directories or links to them), and ones you cannot 432names, directories you can recurse into (directories), and ones you cannot
382recurse into (everything else). 433recurse into (everything else, including symlinks to directories).
383 434
384C<aio_scandir> is a composite request that creates of many sub requests_ 435C<aio_scandir> is a composite request that creates of many sub requests_
385C<$maxreq> specifies the maximum number of outstanding aio requests that 436C<$maxreq> specifies the maximum number of outstanding aio requests that
386this function generates. If it is C<< <= 0 >>, then a suitable default 437this function generates. If it is C<< <= 0 >>, then a suitable default
387will be chosen (currently 6). 438will be chosen (currently 6).
469 map [$_, sprintf "%s%04d", (/.\./ ? "1" : "0"), length], 520 map [$_, sprintf "%s%04d", (/.\./ ? "1" : "0"), length],
470 @$entries]; 521 @$entries];
471 522
472 my (@dirs, @nondirs); 523 my (@dirs, @nondirs);
473 524
474 my ($statcb, $schedcb);
475 my $nreq = 0;
476
477 my $statgrp = add $grp aio_group; 525 my $statgrp = add $grp aio_group sub {
526 $grp->result (\@dirs, \@nondirs);
527 };
478 528
479 $schedcb = sub { 529 limit $statgrp $maxreq;
480 if (@$entries) { 530 feed $statgrp sub {
481 if ($nreq < $maxreq) { 531 return unless @$entries;
482 my $ent = pop @$entries; 532 my $entry = pop @$entries;
533
534 add $statgrp aio_stat "$path/$entry/.", sub {
535 if ($_[0] < 0) {
536 push @nondirs, $entry;
537 } else {
538 # need to check for real directory
539 add $statgrp aio_lstat "$path/$entry", sub {
540 if (-d _) {
541 push @dirs, $entry;
542
543 unless (--$ndirs) {
544 push @nondirs, @$entries;
545 feed $statgrp;
546 }
547 } else {
548 push @nondirs, $entry;
549 }
483 $nreq++; 550 }
484 add $statgrp aio_stat "$path/$ent/.", sub { $statcb->($_[0], $ent) };
485 } 551 }
486 } elsif (!$nreq) {
487 # finished
488 $statgrp->cancel;
489 undef $statcb;
490 undef $schedcb;
491 $grp->result (\@dirs, \@nondirs);
492 } 552 };
493 }; 553 };
494 $statcb = sub {
495 my ($status, $entry) = @_;
496
497 if ($status < 0) {
498 $nreq--;
499 push @nondirs, $entry;
500 &$schedcb;
501 } else {
502 # need to check for real directory
503 add $grp aio_lstat "$path/$entry", sub {
504 $nreq--;
505
506 if (-d _) {
507 push @dirs, $entry;
508
509 if (!--$ndirs) {
510 push @nondirs, @$entries;
511 $entries = [];
512 }
513 } else {
514 push @nondirs, $entry;
515 }
516
517 &$schedcb;
518 }
519 }
520 };
521
522 &$schedcb while @$entries && $nreq < $maxreq;
523 }; 554 };
524 }; 555 };
525 }; 556 };
526 557
527 $grp 558 $grp
539 570
540If this call isn't available because your OS lacks it or it couldn't be 571If this call isn't available because your OS lacks it or it couldn't be
541detected, it will be emulated by calling C<fsync> instead. 572detected, it will be emulated by calling C<fsync> instead.
542 573
543=item aio_group $callback->(...) 574=item aio_group $callback->(...)
544
545[EXPERIMENTAL]
546 575
547This is a very special aio request: Instead of doing something, it is a 576This is a very special aio request: Instead of doing something, it is a
548container for other aio requests, which is useful if you want to bundle 577container for other aio requests, which is useful if you want to bundle
549many requests into a single, composite, request with a definite callback 578many requests into a single, composite, request with a definite callback
550and the ability to cancel the whole request with its subrequests. 579and the ability to cancel the whole request with its subrequests.
591=head2 IO::AIO::REQ CLASS 620=head2 IO::AIO::REQ CLASS
592 621
593All non-aggregate C<aio_*> functions return an object of this class when 622All non-aggregate C<aio_*> functions return an object of this class when
594called in non-void context. 623called in non-void context.
595 624
596A request always moves through the following five states in its lifetime,
597in order: B<ready> (request has been created, but has not been executed
598yet), B<execute> (request is currently being executed), B<pending>
599(request has been executed but callback has not been called yet),
600B<result> (results are being processed synchronously, includes calling the
601callback) and B<done> (request has reached the end of its lifetime and
602holds no resources anymore).
603
604=over 4 625=over 4
605 626
606=item cancel $req 627=item cancel $req
607 628
608Cancels the request, if possible. Has the effect of skipping execution 629Cancels the request, if possible. Has the effect of skipping execution
661=item * They can also can also be added to other IO::AIO::GRP objects. 682=item * They can also can also be added to other IO::AIO::GRP objects.
662 683
663=item * You must not add requests to a group from within the group callback (or 684=item * You must not add requests to a group from within the group callback (or
664any later time). 685any later time).
665 686
666=item * This does not harmonise well with C<max_outstanding>, so best do
667not combine C<aio_group> with it. Groups and feeders are recommended for
668this kind of concurrency-limiting.
669
670=back 687=back
671 688
672Their lifetime, simplified, looks like this: when they are empty, they 689Their lifetime, simplified, looks like this: when they are empty, they
673will finish very quickly. If they contain only requests that are in the 690will finish very quickly. If they contain only requests that are in the
674C<done> state, they will also finish. Otherwise they will continue to 691C<done> state, they will also finish. Otherwise they will continue to
689be added, including other groups, as long as you do not create circular 706be added, including other groups, as long as you do not create circular
690dependencies. 707dependencies.
691 708
692Returns all its arguments. 709Returns all its arguments.
693 710
711=item $grp->cancel_subs
712
713Cancel all subrequests and clears any feeder, but not the group request
714itself. Useful when you queued a lot of events but got a result early.
715
694=item $grp->result (...) 716=item $grp->result (...)
695 717
696Set the result value(s) that will be passed to the group callback when all 718Set the result value(s) that will be passed to the group callback when all
697subrequests have finished. By default, no argument will be passed. 719subrequests have finished. By default, no argument will be passed.
698 720
699=item feed $grp $callback->($grp) 721=item feed $grp $callback->($grp)
700
701[VERY EXPERIMENTAL]
702 722
703Sets a feeder/generator on this group: every group can have an attached 723Sets a feeder/generator on this group: every group can have an attached
704generator that generates requests if idle. The idea behind this is that, 724generator that generates requests if idle. The idea behind this is that,
705although you could just queue as many requests as you want in a group, 725although you could just queue as many requests as you want in a group,
706this might starve other requests for a potentially long time. For 726this might starve other requests for a potentially long time. For
837 857
838Under normal circumstances you don't need to call this function. 858Under normal circumstances you don't need to call this function.
839 859
840=item $oldnreqs = IO::AIO::max_outstanding $nreqs 860=item $oldnreqs = IO::AIO::max_outstanding $nreqs
841 861
842[DEPRECATED] 862[REMOVED]
843 863
864Pre-2.x versions used max_outstanding for a crude request queue length limit.
865
866In 2.x+ you are advised to use a group and a feeder to limit
867concurrency. The max_outstanding feature ran very unstable (endless
868recursions causing segfaults, bad interaction with groups etc.) and was
869removed.
870
871I am deeply sorry, but I am still on the hunt for a good limiting interface.
872
873Original description was as follows:
874
844Sets the maximum number of outstanding requests to C<$nreqs>. If you 875Sets the maximum number of outstanding requests to C<$nreqs>. If you try
845try to queue up more than this number of requests, the caller will block until 876to queue up more than this number of requests, the caller will block until
846some requests have been handled. 877some requests have been handled.
847
848The default is very large, so normally there is no practical limit. If you
849queue up many requests in a loop it often improves speed if you set
850this to a relatively low number, such as C<100>.
851
852This function does not work well together with C<aio_group>'s, and their
853feeder interface is better suited to limiting concurrency, so do not use
854this function.
855
856Under normal circumstances you don't need to call this function.
857 878
858=back 879=back
859 880
860=cut 881=cut
861 882
888This module should do "the right thing" when the process using it forks: 909This module should do "the right thing" when the process using it forks:
889 910
890Before the fork, IO::AIO enters a quiescent state where no requests 911Before the fork, IO::AIO enters a quiescent state where no requests
891can be added in other threads and no results will be processed. After 912can be added in other threads and no results will be processed. After
892the fork the parent simply leaves the quiescent state and continues 913the fork the parent simply leaves the quiescent state and continues
893request/result processing, while the child clears the request/result 914request/result processing, while the child frees the request/result queue
894queue (so the requests started before the fork will only be handled in 915(so that the requests started before the fork will only be handled in the
895the parent). Threads will be started on demand until the limit ste in the 916parent). Threads will be started on demand until the limit set in the
896parent process has been reached again. 917parent process has been reached again.
897 918
898In short: the parent will, after a short pause, continue as if fork had 919In short: the parent will, after a short pause, continue as if fork had
899not been called, while the child will act as if IO::AIO has not been used 920not been called, while the child will act as if IO::AIO has not been used
900yet. 921yet.
901 922
902=head2 MEMORY USAGE 923=head2 MEMORY USAGE
903 924
925Per-request usage:
926
904Each aio request uses - depending on your architecture - around 128 bytes 927Each aio request uses - depending on your architecture - around 100-200
905of memory. In addition, stat requests need a stat buffer (possibly a few 928bytes of memory. In addition, stat requests need a stat buffer (possibly
906hundred bytes). Perl scalars and other data passed into aio requests will 929a few hundred bytes), readdir requires a result buffer and so on. Perl
907also be locked. 930scalars and other data passed into aio requests will also be locked and
931will consume memory till the request has entered the done state.
908 932
909This is now awfully much, so queuing lots of requests is not usually a 933This is now awfully much, so queuing lots of requests is not usually a
910problem. 934problem.
911 935
912Each thread needs a stack area which is usually around 16k, sometimes much 936Per-thread usage:
913larger, depending on the OS. 937
938In the execution phase, some aio requests require more memory for
939temporary buffers, and each thread requires a stack and other data
940structures (usually around 16k-128k, depending on the OS).
941
942=head1 KNOWN BUGS
943
944Known bugs will be fixed in the next release.
914 945
915=head1 SEE ALSO 946=head1 SEE ALSO
916 947
917L<Coro::AIO>. 948L<Coro::AIO>.
918 949

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines