ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/IO-AIO/AIO.pm
Revision: 1.117
Committed: Sat Oct 6 14:05:19 2007 UTC (17 years, 1 month ago) by root
Branch: MAIN
CVS Tags: rel-2_51
Changes since 1.116: +61 -11 lines
Log Message:
*** empty log message ***

File Contents

# Content
1 =head1 NAME
2
3 IO::AIO - Asynchronous Input/Output
4
5 =head1 SYNOPSIS
6
7 use IO::AIO;
8
9 aio_open "/etc/passwd", O_RDONLY, 0, sub {
10 my $fh = shift
11 or die "/etc/passwd: $!";
12 ...
13 };
14
15 aio_unlink "/tmp/file", sub { };
16
17 aio_read $fh, 30000, 1024, $buffer, 0, sub {
18 $_[0] > 0 or die "read error: $!";
19 };
20
21 # version 2+ has request and group objects
22 use IO::AIO 2;
23
24 aioreq_pri 4; # give next request a very high priority
25 my $req = aio_unlink "/tmp/file", sub { };
26 $req->cancel; # cancel request if still in queue
27
28 my $grp = aio_group sub { print "all stats done\n" };
29 add $grp aio_stat "..." for ...;
30
31 # AnyEvent integration
32 open my $fh, "<&=" . IO::AIO::poll_fileno or die "$!";
33 my $w = AnyEvent->io (fh => $fh, poll => 'r', cb => sub { IO::AIO::poll_cb });
34
35 # Event integration
36 Event->io (fd => IO::AIO::poll_fileno,
37 poll => 'r',
38 cb => \&IO::AIO::poll_cb);
39
40 # Glib/Gtk2 integration
41 add_watch Glib::IO IO::AIO::poll_fileno,
42 in => sub { IO::AIO::poll_cb; 1 };
43
44 # Tk integration
45 Tk::Event::IO->fileevent (IO::AIO::poll_fileno, "",
46 readable => \&IO::AIO::poll_cb);
47
48 # Danga::Socket integration
49 Danga::Socket->AddOtherFds (IO::AIO::poll_fileno =>
50 \&IO::AIO::poll_cb);
51
52 =head1 DESCRIPTION
53
54 This module implements asynchronous I/O using whatever means your
55 operating system supports.
56
57 Asynchronous means that operations that can normally block your program
58 (e.g. reading from disk) will be done asynchronously: the operation
59 will still block, but you can do something else in the meantime. This
60 is extremely useful for programs that need to stay interactive even
61 when doing heavy I/O (GUI programs, high performance network servers
62 etc.), but can also be used to easily do operations in parallel that are
63 normally done sequentially, e.g. stat'ing many files, which is much faster
64 on a RAID volume or over NFS when you do a number of stat operations
65 concurrently.
66
67 While most of this works on all types of file descriptors (for
68 example sockets), using these functions on file descriptors that
69 support nonblocking operation (again, sockets, pipes etc.) is very
70 inefficient. Use an event loop for that (such as the L<Event|Event>
71 module): IO::AIO will naturally fit into such an event loop itself.
72
73 In this version, a number of threads are started that execute your
74 requests and signal their completion. You don't need thread support
75 in perl, and the threads created by this module will not be visible
76 to perl. In the future, this module might make use of the native aio
77 functions available on many operating systems. However, they are often
78 not well-supported or restricted (GNU/Linux doesn't allow them on normal
79 files currently, for example), and they would only support aio_read and
80 aio_write, so the remaining functionality would have to be implemented
81 using threads anyway.
82
83 Although the module will work in the presence of other (Perl-) threads,
84 it is currently not reentrant in any way, so use appropriate locking
85 yourself, always call C<poll_cb> from within the same thread, or never
86 call C<poll_cb> (or other C<aio_> functions) recursively.
87
88 =head2 EXAMPLE
89
90 This is a simple example that uses the Event module and loads
91 F</etc/passwd> asynchronously:
92
93 use Fcntl;
94 use Event;
95 use IO::AIO;
96
97 # register the IO::AIO callback with Event
98 Event->io (fd => IO::AIO::poll_fileno,
99 poll => 'r',
100 cb => \&IO::AIO::poll_cb);
101
102 # queue the request to open /etc/passwd
103 aio_open "/etc/passwd", O_RDONLY, 0, sub {
104 my $fh = shift
105 or die "error while opening: $!";
106
107 # stat'ing filehandles is generally non-blocking
108 my $size = -s $fh;
109
110 # queue a request to read the file
111 my $contents;
112 aio_read $fh, 0, $size, $contents, 0, sub {
113 $_[0] == $size
114 or die "short read: $!";
115
116 close $fh;
117
118 # file contents now in $contents
119 print $contents;
120
121 # exit event loop and program
122 Event::unloop;
123 };
124 };
125
126 # possibly queue up other requests, or open GUI windows,
127 # check for sockets etc. etc.
128
129 # process events as long as there are some:
130 Event::loop;
131
132 =head1 REQUEST ANATOMY AND LIFETIME
133
134 Every C<aio_*> function creates a request. which is a C data structure not
135 directly visible to Perl.
136
137 If called in non-void context, every request function returns a Perl
138 object representing the request. In void context, nothing is returned,
139 which saves a bit of memory.
140
141 The perl object is a fairly standard ref-to-hash object. The hash contents
142 are not used by IO::AIO so you are free to store anything you like in it.
143
144 During their existance, aio requests travel through the following states,
145 in order:
146
147 =over 4
148
149 =item ready
150
151 Immediately after a request is created it is put into the ready state,
152 waiting for a thread to execute it.
153
154 =item execute
155
156 A thread has accepted the request for processing and is currently
157 executing it (e.g. blocking in read).
158
159 =item pending
160
161 The request has been executed and is waiting for result processing.
162
163 While request submission and execution is fully asynchronous, result
164 processing is not and relies on the perl interpreter calling C<poll_cb>
165 (or another function with the same effect).
166
167 =item result
168
169 The request results are processed synchronously by C<poll_cb>.
170
171 The C<poll_cb> function will process all outstanding aio requests by
172 calling their callbacks, freeing memory associated with them and managing
173 any groups they are contained in.
174
175 =item done
176
177 Request has reached the end of its lifetime and holds no resources anymore
178 (except possibly for the Perl object, but its connection to the actual
179 aio request is severed and calling its methods will either do nothing or
180 result in a runtime error).
181
182 =back
183
184 =cut
185
186 package IO::AIO;
187
188 use Carp ();
189
190 no warnings;
191 use strict 'vars';
192
193 use base 'Exporter';
194
195 BEGIN {
196 our $VERSION = '2.51';
197
198 our @AIO_REQ = qw(aio_sendfile aio_read aio_write aio_open aio_close aio_stat
199 aio_lstat aio_unlink aio_rmdir aio_readdir aio_scandir aio_symlink
200 aio_readlink aio_fsync aio_fdatasync aio_readahead aio_rename aio_link
201 aio_move aio_copy aio_group aio_nop aio_mknod aio_load aio_rmtree aio_mkdir
202 aio_chown aio_chmod aio_utime aio_truncate);
203 our @EXPORT = (@AIO_REQ, qw(aioreq_pri aioreq_nice aio_block));
204 our @EXPORT_OK = qw(poll_fileno poll_cb poll_wait flush
205 min_parallel max_parallel max_idle
206 nreqs nready npending nthreads
207 max_poll_time max_poll_reqs);
208
209 @IO::AIO::GRP::ISA = 'IO::AIO::REQ';
210
211 require XSLoader;
212 XSLoader::load ("IO::AIO", $VERSION);
213 }
214
215 =head1 FUNCTIONS
216
217 =head2 AIO REQUEST FUNCTIONS
218
219 All the C<aio_*> calls are more or less thin wrappers around the syscall
220 with the same name (sans C<aio_>). The arguments are similar or identical,
221 and they all accept an additional (and optional) C<$callback> argument
222 which must be a code reference. This code reference will get called with
223 the syscall return code (e.g. most syscalls return C<-1> on error, unlike
224 perl, which usually delivers "false") as it's sole argument when the given
225 syscall has been executed asynchronously.
226
227 All functions expecting a filehandle keep a copy of the filehandle
228 internally until the request has finished.
229
230 All functions return request objects of type L<IO::AIO::REQ> that allow
231 further manipulation of those requests while they are in-flight.
232
233 The pathnames you pass to these routines I<must> be absolute and
234 encoded as octets. The reason for the former is that at the time the
235 request is being executed, the current working directory could have
236 changed. Alternatively, you can make sure that you never change the
237 current working directory anywhere in the program and then use relative
238 paths.
239
240 To encode pathnames as octets, either make sure you either: a) always pass
241 in filenames you got from outside (command line, readdir etc.) without
242 tinkering, b) are ASCII or ISO 8859-1, c) use the Encode module and encode
243 your pathnames to the locale (or other) encoding in effect in the user
244 environment, d) use Glib::filename_from_unicode on unicode filenames or e)
245 use something else to ensure your scalar has the correct contents.
246
247 This works, btw. independent of the internal UTF-8 bit, which IO::AIO
248 handles correctly wether it is set or not.
249
250 =over 4
251
252 =item $prev_pri = aioreq_pri [$pri]
253
254 Returns the priority value that would be used for the next request and, if
255 C<$pri> is given, sets the priority for the next aio request.
256
257 The default priority is C<0>, the minimum and maximum priorities are C<-4>
258 and C<4>, respectively. Requests with higher priority will be serviced
259 first.
260
261 The priority will be reset to C<0> after each call to one of the C<aio_*>
262 functions.
263
264 Example: open a file with low priority, then read something from it with
265 higher priority so the read request is serviced before other low priority
266 open requests (potentially spamming the cache):
267
268 aioreq_pri -3;
269 aio_open ..., sub {
270 return unless $_[0];
271
272 aioreq_pri -2;
273 aio_read $_[0], ..., sub {
274 ...
275 };
276 };
277
278
279 =item aioreq_nice $pri_adjust
280
281 Similar to C<aioreq_pri>, but subtracts the given value from the current
282 priority, so the effect is cumulative.
283
284
285 =item aio_open $pathname, $flags, $mode, $callback->($fh)
286
287 Asynchronously open or create a file and call the callback with a newly
288 created filehandle for the file.
289
290 The pathname passed to C<aio_open> must be absolute. See API NOTES, above,
291 for an explanation.
292
293 The C<$flags> argument is a bitmask. See the C<Fcntl> module for a
294 list. They are the same as used by C<sysopen>.
295
296 Likewise, C<$mode> specifies the mode of the newly created file, if it
297 didn't exist and C<O_CREAT> has been given, just like perl's C<sysopen>,
298 except that it is mandatory (i.e. use C<0> if you don't create new files,
299 and C<0666> or C<0777> if you do). Note that the C<$mode> will be modified
300 by the umask in effect then the request is being executed, so better never
301 change the umask.
302
303 Example:
304
305 aio_open "/etc/passwd", O_RDONLY, 0, sub {
306 if ($_[0]) {
307 print "open successful, fh is $_[0]\n";
308 ...
309 } else {
310 die "open failed: $!\n";
311 }
312 };
313
314
315 =item aio_close $fh, $callback->($status)
316
317 Asynchronously close a file and call the callback with the result
318 code.
319
320 Unfortunately, you can't do this to perl. Perl I<insists> very strongly on
321 closing the file descriptor associated with the filehandle itself. Here is
322 what aio_close will try:
323
324 1. dup()licate the fd
325 2. asynchronously close() the duplicated fd
326 3. dup()licate the fd once more
327 4. let perl close() the filehandle
328 5. asynchronously close the duplicated fd
329
330 The idea is that the first close() flushes stuff to disk that closing an
331 fd will flush, so when perl closes the fd, nothing much will need to be
332 flushed. The second async. close() will then flush stuff to disk that
333 closing the last fd to the file will flush.
334
335 Just FYI, SuSv3 has this to say on close:
336
337 All outstanding record locks owned by the process on the file
338 associated with the file descriptor shall be removed.
339
340 If fildes refers to a socket, close() shall cause the socket to be
341 destroyed. ... close() shall block for up to the current linger
342 interval until all data is transmitted.
343 [this actually sounds like a specification bug, but who knows]
344
345 And at least Linux additionally actually flushes stuff on every close,
346 even when the file itself is still open.
347
348 Sounds enourmously inefficient and complicated? Yes... please show me how
349 to nuke perl's fd out of existence...
350
351 =cut
352
353 sub aio_close($;$) {
354 aio_block {
355 my ($fh, $cb) = @_;
356
357 my $pri = aioreq_pri;
358 my $grp = aio_group $cb;
359
360 my $fd = fileno $fh;
361
362 defined $fd or Carp::croak "aio_close called with fd-less filehandle";
363
364 # if the dups fail we will simply get EBADF
365 my $fd2 = _dup $fd;
366 aioreq_pri $pri;
367 add $grp _aio_close $fd2, sub {
368 my $fd2 = _dup $fd;
369 close $fh;
370 aioreq_pri $pri;
371 add $grp _aio_close $fd2, sub {
372 $grp->result ($_[0]);
373 };
374 };
375
376 $grp
377 }
378 }
379
380
381 =item aio_read $fh,$offset,$length, $data,$dataoffset, $callback->($retval)
382
383 =item aio_write $fh,$offset,$length, $data,$dataoffset, $callback->($retval)
384
385 Reads or writes C<$length> bytes from the specified C<$fh> and C<$offset>
386 into the scalar given by C<$data> and offset C<$dataoffset> and calls the
387 callback without the actual number of bytes read (or -1 on error, just
388 like the syscall).
389
390 If C<$offset> is undefined, then the current file descriptor offset will
391 be used (and updated), otherwise the file descriptor offset will not be
392 changed by these calls.
393
394 If C<$length> is undefined in C<aio_write>, use the remaining length of C<$data>.
395
396 If C<$dataoffset> is less than zero, it will be counted from the end of
397 C<$data>.
398
399 The C<$data> scalar I<MUST NOT> be modified in any way while the request
400 is outstanding. Modifying it can result in segfaults or World War III (if
401 the necessary/optional hardware is installed).
402
403 Example: Read 15 bytes at offset 7 into scalar C<$buffer>, starting at
404 offset C<0> within the scalar:
405
406 aio_read $fh, 7, 15, $buffer, 0, sub {
407 $_[0] > 0 or die "read error: $!";
408 print "read $_[0] bytes: <$buffer>\n";
409 };
410
411
412 =item aio_sendfile $out_fh, $in_fh, $in_offset, $length, $callback->($retval)
413
414 Tries to copy C<$length> bytes from C<$in_fh> to C<$out_fh>. It starts
415 reading at byte offset C<$in_offset>, and starts writing at the current
416 file offset of C<$out_fh>. Because of that, it is not safe to issue more
417 than one C<aio_sendfile> per C<$out_fh>, as they will interfere with each
418 other.
419
420 This call tries to make use of a native C<sendfile> syscall to provide
421 zero-copy operation. For this to work, C<$out_fh> should refer to a
422 socket, and C<$in_fh> should refer to mmap'able file.
423
424 If the native sendfile call fails or is not implemented, it will be
425 emulated, so you can call C<aio_sendfile> on any type of filehandle
426 regardless of the limitations of the operating system.
427
428 Please note, however, that C<aio_sendfile> can read more bytes from
429 C<$in_fh> than are written, and there is no way to find out how many
430 bytes have been read from C<aio_sendfile> alone, as C<aio_sendfile> only
431 provides the number of bytes written to C<$out_fh>. Only if the result
432 value equals C<$length> one can assume that C<$length> bytes have been
433 read.
434
435
436 =item aio_readahead $fh,$offset,$length, $callback->($retval)
437
438 C<aio_readahead> populates the page cache with data from a file so that
439 subsequent reads from that file will not block on disk I/O. The C<$offset>
440 argument specifies the starting point from which data is to be read and
441 C<$length> specifies the number of bytes to be read. I/O is performed in
442 whole pages, so that offset is effectively rounded down to a page boundary
443 and bytes are read up to the next page boundary greater than or equal to
444 (off-set+length). C<aio_readahead> does not read beyond the end of the
445 file. The current file offset of the file is left unchanged.
446
447 If that syscall doesn't exist (likely if your OS isn't Linux) it will be
448 emulated by simply reading the data, which would have a similar effect.
449
450
451 =item aio_stat $fh_or_path, $callback->($status)
452
453 =item aio_lstat $fh, $callback->($status)
454
455 Works like perl's C<stat> or C<lstat> in void context. The callback will
456 be called after the stat and the results will be available using C<stat _>
457 or C<-s _> etc...
458
459 The pathname passed to C<aio_stat> must be absolute. See API NOTES, above,
460 for an explanation.
461
462 Currently, the stats are always 64-bit-stats, i.e. instead of returning an
463 error when stat'ing a large file, the results will be silently truncated
464 unless perl itself is compiled with large file support.
465
466 Example: Print the length of F</etc/passwd>:
467
468 aio_stat "/etc/passwd", sub {
469 $_[0] and die "stat failed: $!";
470 print "size is ", -s _, "\n";
471 };
472
473
474 =item aio_utime $fh_or_path, $atime, $mtime, $callback->($status)
475
476 Works like perl's C<utime> function (including the special case of $atime
477 and $mtime being undef). Fractional times are supported if the underlying
478 syscalls support them.
479
480 When called with a pathname, uses utimes(2) if available, otherwise
481 utime(2). If called on a file descriptor, uses futimes(2) if available,
482 otherwise returns ENOSYS, so this is not portable.
483
484 Examples:
485
486 # set atime and mtime to current time (basically touch(1)):
487 aio_utime "path", undef, undef;
488 # set atime to current time and mtime to beginning of the epoch:
489 aio_utime "path", time, undef; # undef==0
490
491
492 =item aio_chown $fh_or_path, $uid, $gid, $callback->($status)
493
494 Works like perl's C<chown> function, except that C<undef> for either $uid
495 or $gid is being interpreted as "do not change" (but -1 can also be used).
496
497 Examples:
498
499 # same as "chown root path" in the shell:
500 aio_chown "path", 0, -1;
501 # same as above:
502 aio_chown "path", 0, undef;
503
504
505 =item aio_truncate $fh_or_path, $offset, $callback->($status)
506
507 Works like truncate(2) or ftruncate(2).
508
509
510 =item aio_chmod $fh_or_path, $mode, $callback->($status)
511
512 Works like perl's C<chmod> function.
513
514
515 =item aio_unlink $pathname, $callback->($status)
516
517 Asynchronously unlink (delete) a file and call the callback with the
518 result code.
519
520
521 =item aio_mknod $path, $mode, $dev, $callback->($status)
522
523 [EXPERIMENTAL]
524
525 Asynchronously create a device node (or fifo). See mknod(2).
526
527 The only (POSIX-) portable way of calling this function is:
528
529 aio_mknod $path, IO::AIO::S_IFIFO | $mode, 0, sub { ...
530
531
532 =item aio_link $srcpath, $dstpath, $callback->($status)
533
534 Asynchronously create a new link to the existing object at C<$srcpath> at
535 the path C<$dstpath> and call the callback with the result code.
536
537
538 =item aio_symlink $srcpath, $dstpath, $callback->($status)
539
540 Asynchronously create a new symbolic link to the existing object at C<$srcpath> at
541 the path C<$dstpath> and call the callback with the result code.
542
543
544 =item aio_readlink $path, $callback->($link)
545
546 Asynchronously read the symlink specified by C<$path> and pass it to
547 the callback. If an error occurs, nothing or undef gets passed to the
548 callback.
549
550
551 =item aio_rename $srcpath, $dstpath, $callback->($status)
552
553 Asynchronously rename the object at C<$srcpath> to C<$dstpath>, just as
554 rename(2) and call the callback with the result code.
555
556
557 =item aio_mkdir $pathname, $mode, $callback->($status)
558
559 Asynchronously mkdir (create) a directory and call the callback with
560 the result code. C<$mode> will be modified by the umask at the time the
561 request is executed, so do not change your umask.
562
563
564 =item aio_rmdir $pathname, $callback->($status)
565
566 Asynchronously rmdir (delete) a directory and call the callback with the
567 result code.
568
569
570 =item aio_readdir $pathname, $callback->($entries)
571
572 Unlike the POSIX call of the same name, C<aio_readdir> reads an entire
573 directory (i.e. opendir + readdir + closedir). The entries will not be
574 sorted, and will B<NOT> include the C<.> and C<..> entries.
575
576 The callback a single argument which is either C<undef> or an array-ref
577 with the filenames.
578
579
580 =item aio_load $path, $data, $callback->($status)
581
582 This is a composite request that tries to fully load the given file into
583 memory. Status is the same as with aio_read.
584
585 =cut
586
587 sub aio_load($$;$) {
588 aio_block {
589 my ($path, undef, $cb) = @_;
590 my $data = \$_[1];
591
592 my $pri = aioreq_pri;
593 my $grp = aio_group $cb;
594
595 aioreq_pri $pri;
596 add $grp aio_open $path, O_RDONLY, 0, sub {
597 my $fh = shift
598 or return $grp->result (-1);
599
600 aioreq_pri $pri;
601 add $grp aio_read $fh, 0, (-s $fh), $$data, 0, sub {
602 $grp->result ($_[0]);
603 };
604 };
605
606 $grp
607 }
608 }
609
610 =item aio_copy $srcpath, $dstpath, $callback->($status)
611
612 Try to copy the I<file> (directories not supported as either source or
613 destination) from C<$srcpath> to C<$dstpath> and call the callback with
614 the C<0> (error) or C<-1> ok.
615
616 This is a composite request that it creates the destination file with
617 mode 0200 and copies the contents of the source file into it using
618 C<aio_sendfile>, followed by restoring atime, mtime, access mode and
619 uid/gid, in that order.
620
621 If an error occurs, the partial destination file will be unlinked, if
622 possible, except when setting atime, mtime, access mode and uid/gid, where
623 errors are being ignored.
624
625 =cut
626
627 sub aio_copy($$;$) {
628 aio_block {
629 my ($src, $dst, $cb) = @_;
630
631 my $pri = aioreq_pri;
632 my $grp = aio_group $cb;
633
634 aioreq_pri $pri;
635 add $grp aio_open $src, O_RDONLY, 0, sub {
636 if (my $src_fh = $_[0]) {
637 my @stat = stat $src_fh;
638
639 aioreq_pri $pri;
640 add $grp aio_open $dst, O_CREAT | O_WRONLY | O_TRUNC, 0200, sub {
641 if (my $dst_fh = $_[0]) {
642 aioreq_pri $pri;
643 add $grp aio_sendfile $dst_fh, $src_fh, 0, $stat[7], sub {
644 if ($_[0] == $stat[7]) {
645 $grp->result (0);
646 close $src_fh;
647
648 # those should not normally block. should. should.
649 utime $stat[8], $stat[9], $dst;
650 chmod $stat[2] & 07777, $dst_fh;
651 chown $stat[4], $stat[5], $dst_fh;
652 close $dst_fh;
653 } else {
654 $grp->result (-1);
655 close $src_fh;
656 close $dst_fh;
657
658 aioreq $pri;
659 add $grp aio_unlink $dst;
660 }
661 };
662 } else {
663 $grp->result (-1);
664 }
665 },
666
667 } else {
668 $grp->result (-1);
669 }
670 };
671
672 $grp
673 }
674 }
675
676 =item aio_move $srcpath, $dstpath, $callback->($status)
677
678 Try to move the I<file> (directories not supported as either source or
679 destination) from C<$srcpath> to C<$dstpath> and call the callback with
680 the C<0> (error) or C<-1> ok.
681
682 This is a composite request that tries to rename(2) the file first. If
683 rename files with C<EXDEV>, it copies the file with C<aio_copy> and, if
684 that is successful, unlinking the C<$srcpath>.
685
686 =cut
687
688 sub aio_move($$;$) {
689 aio_block {
690 my ($src, $dst, $cb) = @_;
691
692 my $pri = aioreq_pri;
693 my $grp = aio_group $cb;
694
695 aioreq_pri $pri;
696 add $grp aio_rename $src, $dst, sub {
697 if ($_[0] && $! == EXDEV) {
698 aioreq_pri $pri;
699 add $grp aio_copy $src, $dst, sub {
700 $grp->result ($_[0]);
701
702 if (!$_[0]) {
703 aioreq_pri $pri;
704 add $grp aio_unlink $src;
705 }
706 };
707 } else {
708 $grp->result ($_[0]);
709 }
710 };
711
712 $grp
713 }
714 }
715
716 =item aio_scandir $path, $maxreq, $callback->($dirs, $nondirs)
717
718 Scans a directory (similar to C<aio_readdir>) but additionally tries to
719 efficiently separate the entries of directory C<$path> into two sets of
720 names, directories you can recurse into (directories), and ones you cannot
721 recurse into (everything else, including symlinks to directories).
722
723 C<aio_scandir> is a composite request that creates of many sub requests_
724 C<$maxreq> specifies the maximum number of outstanding aio requests that
725 this function generates. If it is C<< <= 0 >>, then a suitable default
726 will be chosen (currently 4).
727
728 On error, the callback is called without arguments, otherwise it receives
729 two array-refs with path-relative entry names.
730
731 Example:
732
733 aio_scandir $dir, 0, sub {
734 my ($dirs, $nondirs) = @_;
735 print "real directories: @$dirs\n";
736 print "everything else: @$nondirs\n";
737 };
738
739 Implementation notes.
740
741 The C<aio_readdir> cannot be avoided, but C<stat()>'ing every entry can.
742
743 After reading the directory, the modification time, size etc. of the
744 directory before and after the readdir is checked, and if they match (and
745 isn't the current time), the link count will be used to decide how many
746 entries are directories (if >= 2). Otherwise, no knowledge of the number
747 of subdirectories will be assumed.
748
749 Then entries will be sorted into likely directories (everything without
750 a non-initial dot currently) and likely non-directories (everything
751 else). Then every entry plus an appended C</.> will be C<stat>'ed,
752 likely directories first. If that succeeds, it assumes that the entry
753 is a directory or a symlink to directory (which will be checked
754 seperately). This is often faster than stat'ing the entry itself because
755 filesystems might detect the type of the entry without reading the inode
756 data (e.g. ext2fs filetype feature).
757
758 If the known number of directories (link count - 2) has been reached, the
759 rest of the entries is assumed to be non-directories.
760
761 This only works with certainty on POSIX (= UNIX) filesystems, which
762 fortunately are the vast majority of filesystems around.
763
764 It will also likely work on non-POSIX filesystems with reduced efficiency
765 as those tend to return 0 or 1 as link counts, which disables the
766 directory counting heuristic.
767
768 =cut
769
770 sub aio_scandir($$;$) {
771 aio_block {
772 my ($path, $maxreq, $cb) = @_;
773
774 my $pri = aioreq_pri;
775
776 my $grp = aio_group $cb;
777
778 $maxreq = 4 if $maxreq <= 0;
779
780 # stat once
781 aioreq_pri $pri;
782 add $grp aio_stat $path, sub {
783 return $grp->result () if $_[0];
784 my $now = time;
785 my $hash1 = join ":", (stat _)[0,1,3,7,9];
786
787 # read the directory entries
788 aioreq_pri $pri;
789 add $grp aio_readdir $path, sub {
790 my $entries = shift
791 or return $grp->result ();
792
793 # stat the dir another time
794 aioreq_pri $pri;
795 add $grp aio_stat $path, sub {
796 my $hash2 = join ":", (stat _)[0,1,3,7,9];
797
798 my $ndirs;
799
800 # take the slow route if anything looks fishy
801 if ($hash1 ne $hash2 or (stat _)[9] == $now) {
802 $ndirs = -1;
803 } else {
804 # if nlink == 2, we are finished
805 # on non-posix-fs's, we rely on nlink < 2
806 $ndirs = (stat _)[3] - 2
807 or return $grp->result ([], $entries);
808 }
809
810 # sort into likely dirs and likely nondirs
811 # dirs == files without ".", short entries first
812 $entries = [map $_->[0],
813 sort { $b->[1] cmp $a->[1] }
814 map [$_, sprintf "%s%04d", (/.\./ ? "1" : "0"), length],
815 @$entries];
816
817 my (@dirs, @nondirs);
818
819 my $statgrp = add $grp aio_group sub {
820 $grp->result (\@dirs, \@nondirs);
821 };
822
823 limit $statgrp $maxreq;
824 feed $statgrp sub {
825 return unless @$entries;
826 my $entry = pop @$entries;
827
828 aioreq_pri $pri;
829 add $statgrp aio_stat "$path/$entry/.", sub {
830 if ($_[0] < 0) {
831 push @nondirs, $entry;
832 } else {
833 # need to check for real directory
834 aioreq_pri $pri;
835 add $statgrp aio_lstat "$path/$entry", sub {
836 if (-d _) {
837 push @dirs, $entry;
838
839 unless (--$ndirs) {
840 push @nondirs, @$entries;
841 feed $statgrp;
842 }
843 } else {
844 push @nondirs, $entry;
845 }
846 }
847 }
848 };
849 };
850 };
851 };
852 };
853
854 $grp
855 }
856 }
857
858 =item aio_rmtree $path, $callback->($status)
859
860 Delete a directory tree starting (and including) C<$path>, return the
861 status of the final C<rmdir> only. This is a composite request that
862 uses C<aio_scandir> to recurse into and rmdir directories, and unlink
863 everything else.
864
865 =cut
866
867 sub aio_rmtree;
868 sub aio_rmtree($;$) {
869 aio_block {
870 my ($path, $cb) = @_;
871
872 my $pri = aioreq_pri;
873 my $grp = aio_group $cb;
874
875 aioreq_pri $pri;
876 add $grp aio_scandir $path, 0, sub {
877 my ($dirs, $nondirs) = @_;
878
879 my $dirgrp = aio_group sub {
880 add $grp aio_rmdir $path, sub {
881 $grp->result ($_[0]);
882 };
883 };
884
885 (aioreq_pri $pri), add $dirgrp aio_rmtree "$path/$_" for @$dirs;
886 (aioreq_pri $pri), add $dirgrp aio_unlink "$path/$_" for @$nondirs;
887
888 add $grp $dirgrp;
889 };
890
891 $grp
892 }
893 }
894
895 =item aio_fsync $fh, $callback->($status)
896
897 Asynchronously call fsync on the given filehandle and call the callback
898 with the fsync result code.
899
900 =item aio_fdatasync $fh, $callback->($status)
901
902 Asynchronously call fdatasync on the given filehandle and call the
903 callback with the fdatasync result code.
904
905 If this call isn't available because your OS lacks it or it couldn't be
906 detected, it will be emulated by calling C<fsync> instead.
907
908 =item aio_group $callback->(...)
909
910 This is a very special aio request: Instead of doing something, it is a
911 container for other aio requests, which is useful if you want to bundle
912 many requests into a single, composite, request with a definite callback
913 and the ability to cancel the whole request with its subrequests.
914
915 Returns an object of class L<IO::AIO::GRP>. See its documentation below
916 for more info.
917
918 Example:
919
920 my $grp = aio_group sub {
921 print "all stats done\n";
922 };
923
924 add $grp
925 (aio_stat ...),
926 (aio_stat ...),
927 ...;
928
929 =item aio_nop $callback->()
930
931 This is a special request - it does nothing in itself and is only used for
932 side effects, such as when you want to add a dummy request to a group so
933 that finishing the requests in the group depends on executing the given
934 code.
935
936 While this request does nothing, it still goes through the execution
937 phase and still requires a worker thread. Thus, the callback will not
938 be executed immediately but only after other requests in the queue have
939 entered their execution phase. This can be used to measure request
940 latency.
941
942 =item IO::AIO::aio_busy $fractional_seconds, $callback->() *NOT EXPORTED*
943
944 Mainly used for debugging and benchmarking, this aio request puts one of
945 the request workers to sleep for the given time.
946
947 While it is theoretically handy to have simple I/O scheduling requests
948 like sleep and file handle readable/writable, the overhead this creates is
949 immense (it blocks a thread for a long time) so do not use this function
950 except to put your application under artificial I/O pressure.
951
952 =back
953
954 =head2 IO::AIO::REQ CLASS
955
956 All non-aggregate C<aio_*> functions return an object of this class when
957 called in non-void context.
958
959 =over 4
960
961 =item cancel $req
962
963 Cancels the request, if possible. Has the effect of skipping execution
964 when entering the B<execute> state and skipping calling the callback when
965 entering the the B<result> state, but will leave the request otherwise
966 untouched. That means that requests that currently execute will not be
967 stopped and resources held by the request will not be freed prematurely.
968
969 =item cb $req $callback->(...)
970
971 Replace (or simply set) the callback registered to the request.
972
973 =back
974
975 =head2 IO::AIO::GRP CLASS
976
977 This class is a subclass of L<IO::AIO::REQ>, so all its methods apply to
978 objects of this class, too.
979
980 A IO::AIO::GRP object is a special request that can contain multiple other
981 aio requests.
982
983 You create one by calling the C<aio_group> constructing function with a
984 callback that will be called when all contained requests have entered the
985 C<done> state:
986
987 my $grp = aio_group sub {
988 print "all requests are done\n";
989 };
990
991 You add requests by calling the C<add> method with one or more
992 C<IO::AIO::REQ> objects:
993
994 $grp->add (aio_unlink "...");
995
996 add $grp aio_stat "...", sub {
997 $_[0] or return $grp->result ("error");
998
999 # add another request dynamically, if first succeeded
1000 add $grp aio_open "...", sub {
1001 $grp->result ("ok");
1002 };
1003 };
1004
1005 This makes it very easy to create composite requests (see the source of
1006 C<aio_move> for an application) that work and feel like simple requests.
1007
1008 =over 4
1009
1010 =item * The IO::AIO::GRP objects will be cleaned up during calls to
1011 C<IO::AIO::poll_cb>, just like any other request.
1012
1013 =item * They can be canceled like any other request. Canceling will cancel not
1014 only the request itself, but also all requests it contains.
1015
1016 =item * They can also can also be added to other IO::AIO::GRP objects.
1017
1018 =item * You must not add requests to a group from within the group callback (or
1019 any later time).
1020
1021 =back
1022
1023 Their lifetime, simplified, looks like this: when they are empty, they
1024 will finish very quickly. If they contain only requests that are in the
1025 C<done> state, they will also finish. Otherwise they will continue to
1026 exist.
1027
1028 That means after creating a group you have some time to add requests. And
1029 in the callbacks of those requests, you can add further requests to the
1030 group. And only when all those requests have finished will the the group
1031 itself finish.
1032
1033 =over 4
1034
1035 =item add $grp ...
1036
1037 =item $grp->add (...)
1038
1039 Add one or more requests to the group. Any type of L<IO::AIO::REQ> can
1040 be added, including other groups, as long as you do not create circular
1041 dependencies.
1042
1043 Returns all its arguments.
1044
1045 =item $grp->cancel_subs
1046
1047 Cancel all subrequests and clears any feeder, but not the group request
1048 itself. Useful when you queued a lot of events but got a result early.
1049
1050 =item $grp->result (...)
1051
1052 Set the result value(s) that will be passed to the group callback when all
1053 subrequests have finished and set thre groups errno to the current value
1054 of errno (just like calling C<errno> without an error number). By default,
1055 no argument will be passed and errno is zero.
1056
1057 =item $grp->errno ([$errno])
1058
1059 Sets the group errno value to C<$errno>, or the current value of errno
1060 when the argument is missing.
1061
1062 Every aio request has an associated errno value that is restored when
1063 the callback is invoked. This method lets you change this value from its
1064 default (0).
1065
1066 Calling C<result> will also set errno, so make sure you either set C<$!>
1067 before the call to C<result>, or call c<errno> after it.
1068
1069 =item feed $grp $callback->($grp)
1070
1071 Sets a feeder/generator on this group: every group can have an attached
1072 generator that generates requests if idle. The idea behind this is that,
1073 although you could just queue as many requests as you want in a group,
1074 this might starve other requests for a potentially long time. For
1075 example, C<aio_scandir> might generate hundreds of thousands C<aio_stat>
1076 requests, delaying any later requests for a long time.
1077
1078 To avoid this, and allow incremental generation of requests, you can
1079 instead a group and set a feeder on it that generates those requests. The
1080 feed callback will be called whenever there are few enough (see C<limit>,
1081 below) requests active in the group itself and is expected to queue more
1082 requests.
1083
1084 The feed callback can queue as many requests as it likes (i.e. C<add> does
1085 not impose any limits).
1086
1087 If the feed does not queue more requests when called, it will be
1088 automatically removed from the group.
1089
1090 If the feed limit is C<0>, it will be set to C<2> automatically.
1091
1092 Example:
1093
1094 # stat all files in @files, but only ever use four aio requests concurrently:
1095
1096 my $grp = aio_group sub { print "finished\n" };
1097 limit $grp 4;
1098 feed $grp sub {
1099 my $file = pop @files
1100 or return;
1101
1102 add $grp aio_stat $file, sub { ... };
1103 };
1104
1105 =item limit $grp $num
1106
1107 Sets the feeder limit for the group: The feeder will be called whenever
1108 the group contains less than this many requests.
1109
1110 Setting the limit to C<0> will pause the feeding process.
1111
1112 =back
1113
1114 =head2 SUPPORT FUNCTIONS
1115
1116 =head3 EVENT PROCESSING AND EVENT LOOP INTEGRATION
1117
1118 =over 4
1119
1120 =item $fileno = IO::AIO::poll_fileno
1121
1122 Return the I<request result pipe file descriptor>. This filehandle must be
1123 polled for reading by some mechanism outside this module (e.g. Event or
1124 select, see below or the SYNOPSIS). If the pipe becomes readable you have
1125 to call C<poll_cb> to check the results.
1126
1127 See C<poll_cb> for an example.
1128
1129 =item IO::AIO::poll_cb
1130
1131 Process some outstanding events on the result pipe. You have to call this
1132 regularly. Returns the number of events processed. Returns immediately
1133 when no events are outstanding. The amount of events processed depends on
1134 the settings of C<IO::AIO::max_poll_req> and C<IO::AIO::max_poll_time>.
1135
1136 If not all requests were processed for whatever reason, the filehandle
1137 will still be ready when C<poll_cb> returns.
1138
1139 Example: Install an Event watcher that automatically calls
1140 IO::AIO::poll_cb with high priority:
1141
1142 Event->io (fd => IO::AIO::poll_fileno,
1143 poll => 'r', async => 1,
1144 cb => \&IO::AIO::poll_cb);
1145
1146 =item IO::AIO::max_poll_reqs $nreqs
1147
1148 =item IO::AIO::max_poll_time $seconds
1149
1150 These set the maximum number of requests (default C<0>, meaning infinity)
1151 that are being processed by C<IO::AIO::poll_cb> in one call, respectively
1152 the maximum amount of time (default C<0>, meaning infinity) spent in
1153 C<IO::AIO::poll_cb> to process requests (more correctly the mininum amount
1154 of time C<poll_cb> is allowed to use).
1155
1156 Setting C<max_poll_time> to a non-zero value creates an overhead of one
1157 syscall per request processed, which is not normally a problem unless your
1158 callbacks are really really fast or your OS is really really slow (I am
1159 not mentioning Solaris here). Using C<max_poll_reqs> incurs no overhead.
1160
1161 Setting these is useful if you want to ensure some level of
1162 interactiveness when perl is not fast enough to process all requests in
1163 time.
1164
1165 For interactive programs, values such as C<0.01> to C<0.1> should be fine.
1166
1167 Example: Install an Event watcher that automatically calls
1168 IO::AIO::poll_cb with low priority, to ensure that other parts of the
1169 program get the CPU sometimes even under high AIO load.
1170
1171 # try not to spend much more than 0.1s in poll_cb
1172 IO::AIO::max_poll_time 0.1;
1173
1174 # use a low priority so other tasks have priority
1175 Event->io (fd => IO::AIO::poll_fileno,
1176 poll => 'r', nice => 1,
1177 cb => &IO::AIO::poll_cb);
1178
1179 =item IO::AIO::poll_wait
1180
1181 If there are any outstanding requests and none of them in the result
1182 phase, wait till the result filehandle becomes ready for reading (simply
1183 does a C<select> on the filehandle. This is useful if you want to
1184 synchronously wait for some requests to finish).
1185
1186 See C<nreqs> for an example.
1187
1188 =item IO::AIO::poll
1189
1190 Waits until some requests have been handled.
1191
1192 Returns the number of requests processed, but is otherwise strictly
1193 equivalent to:
1194
1195 IO::AIO::poll_wait, IO::AIO::poll_cb
1196
1197 =item IO::AIO::flush
1198
1199 Wait till all outstanding AIO requests have been handled.
1200
1201 Strictly equivalent to:
1202
1203 IO::AIO::poll_wait, IO::AIO::poll_cb
1204 while IO::AIO::nreqs;
1205
1206 =back
1207
1208 =head3 CONTROLLING THE NUMBER OF THREADS
1209
1210 =over
1211
1212 =item IO::AIO::min_parallel $nthreads
1213
1214 Set the minimum number of AIO threads to C<$nthreads>. The current
1215 default is C<8>, which means eight asynchronous operations can execute
1216 concurrently at any one time (the number of outstanding requests,
1217 however, is unlimited).
1218
1219 IO::AIO starts threads only on demand, when an AIO request is queued and
1220 no free thread exists. Please note that queueing up a hundred requests can
1221 create demand for a hundred threads, even if it turns out that everything
1222 is in the cache and could have been processed faster by a single thread.
1223
1224 It is recommended to keep the number of threads relatively low, as some
1225 Linux kernel versions will scale negatively with the number of threads
1226 (higher parallelity => MUCH higher latency). With current Linux 2.6
1227 versions, 4-32 threads should be fine.
1228
1229 Under most circumstances you don't need to call this function, as the
1230 module selects a default that is suitable for low to moderate load.
1231
1232 =item IO::AIO::max_parallel $nthreads
1233
1234 Sets the maximum number of AIO threads to C<$nthreads>. If more than the
1235 specified number of threads are currently running, this function kills
1236 them. This function blocks until the limit is reached.
1237
1238 While C<$nthreads> are zero, aio requests get queued but not executed
1239 until the number of threads has been increased again.
1240
1241 This module automatically runs C<max_parallel 0> at program end, to ensure
1242 that all threads are killed and that there are no outstanding requests.
1243
1244 Under normal circumstances you don't need to call this function.
1245
1246 =item IO::AIO::max_idle $nthreads
1247
1248 Limit the number of threads (default: 4) that are allowed to idle (i.e.,
1249 threads that did not get a request to process within 10 seconds). That
1250 means if a thread becomes idle while C<$nthreads> other threads are also
1251 idle, it will free its resources and exit.
1252
1253 This is useful when you allow a large number of threads (e.g. 100 or 1000)
1254 to allow for extremely high load situations, but want to free resources
1255 under normal circumstances (1000 threads can easily consume 30MB of RAM).
1256
1257 The default is probably ok in most situations, especially if thread
1258 creation is fast. If thread creation is very slow on your system you might
1259 want to use larger values.
1260
1261 =item $oldmaxreqs = IO::AIO::max_outstanding $maxreqs
1262
1263 This is a very bad function to use in interactive programs because it
1264 blocks, and a bad way to reduce concurrency because it is inexact: Better
1265 use an C<aio_group> together with a feed callback.
1266
1267 Sets the maximum number of outstanding requests to C<$nreqs>. If you
1268 do queue up more than this number of requests, the next call to the
1269 C<poll_cb> (and C<poll_some> and other functions calling C<poll_cb>)
1270 function will block until the limit is no longer exceeded.
1271
1272 The default value is very large, so there is no practical limit on the
1273 number of outstanding requests.
1274
1275 You can still queue as many requests as you want. Therefore,
1276 C<max_oustsanding> is mainly useful in simple scripts (with low values) or
1277 as a stop gap to shield against fatal memory overflow (with large values).
1278
1279 =back
1280
1281 =head3 STATISTICAL INFORMATION
1282
1283 =over
1284
1285 =item IO::AIO::nreqs
1286
1287 Returns the number of requests currently in the ready, execute or pending
1288 states (i.e. for which their callback has not been invoked yet).
1289
1290 Example: wait till there are no outstanding requests anymore:
1291
1292 IO::AIO::poll_wait, IO::AIO::poll_cb
1293 while IO::AIO::nreqs;
1294
1295 =item IO::AIO::nready
1296
1297 Returns the number of requests currently in the ready state (not yet
1298 executed).
1299
1300 =item IO::AIO::npending
1301
1302 Returns the number of requests currently in the pending state (executed,
1303 but not yet processed by poll_cb).
1304
1305 =back
1306
1307 =cut
1308
1309 min_parallel 8;
1310
1311 END { flush }
1312
1313 1;
1314
1315 =head2 FORK BEHAVIOUR
1316
1317 This module should do "the right thing" when the process using it forks:
1318
1319 Before the fork, IO::AIO enters a quiescent state where no requests
1320 can be added in other threads and no results will be processed. After
1321 the fork the parent simply leaves the quiescent state and continues
1322 request/result processing, while the child frees the request/result queue
1323 (so that the requests started before the fork will only be handled in the
1324 parent). Threads will be started on demand until the limit set in the
1325 parent process has been reached again.
1326
1327 In short: the parent will, after a short pause, continue as if fork had
1328 not been called, while the child will act as if IO::AIO has not been used
1329 yet.
1330
1331 =head2 MEMORY USAGE
1332
1333 Per-request usage:
1334
1335 Each aio request uses - depending on your architecture - around 100-200
1336 bytes of memory. In addition, stat requests need a stat buffer (possibly
1337 a few hundred bytes), readdir requires a result buffer and so on. Perl
1338 scalars and other data passed into aio requests will also be locked and
1339 will consume memory till the request has entered the done state.
1340
1341 This is not awfully much, so queuing lots of requests is not usually a
1342 problem.
1343
1344 Per-thread usage:
1345
1346 In the execution phase, some aio requests require more memory for
1347 temporary buffers, and each thread requires a stack and other data
1348 structures (usually around 16k-128k, depending on the OS).
1349
1350 =head1 KNOWN BUGS
1351
1352 Known bugs will be fixed in the next release.
1353
1354 =head1 SEE ALSO
1355
1356 L<Coro::AIO>.
1357
1358 =head1 AUTHOR
1359
1360 Marc Lehmann <schmorp@schmorp.de>
1361 http://home.schmorp.de/
1362
1363 =cut
1364