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 1.30 by root, Wed Aug 17 04:47:38 2005 UTC vs.
Revision 1.68 by root, Tue Oct 24 03:17:39 2006 UTC

15 15
16 aio_read $fh, 30000, 1024, $buffer, 0, sub { 16 aio_read $fh, 30000, 1024, $buffer, 0, sub {
17 $_[0] > 0 or die "read error: $!"; 17 $_[0] > 0 or die "read error: $!";
18 }; 18 };
19 19
20 # Event 20 # version 2+ has request and group objects
21 use IO::AIO 2;
22
23 aioreq_pri 4; # give next request a very high priority
24 my $req = aio_unlink "/tmp/file", sub { };
25 $req->cancel; # cancel request if still in queue
26
27 my $grp = aio_group sub { print "all stats done\n" };
28 add $grp aio_stat "..." for ...;
29
30 # AnyEvent integration
31 open my $fh, "<&=" . IO::AIO::poll_fileno or die "$!";
32 my $w = AnyEvent->io (fh => $fh, poll => 'r', cb => sub { IO::AIO::poll_cb });
33
34 # Event integration
21 Event->io (fd => IO::AIO::poll_fileno, 35 Event->io (fd => IO::AIO::poll_fileno,
22 poll => 'r', 36 poll => 'r',
23 cb => \&IO::AIO::poll_cb); 37 cb => \&IO::AIO::poll_cb);
24 38
25 # Glib/Gtk2 39 # Glib/Gtk2 integration
26 add_watch Glib::IO IO::AIO::poll_fileno, 40 add_watch Glib::IO IO::AIO::poll_fileno,
27 in => sub { IO::AIO::poll_cb; 1 }; 41 in => sub { IO::AIO::poll_cb; 1 };
28 42
29 # Tk 43 # Tk integration
30 Tk::Event::IO->fileevent (IO::AIO::poll_fileno, "", 44 Tk::Event::IO->fileevent (IO::AIO::poll_fileno, "",
31 readable => \&IO::AIO::poll_cb); 45 readable => \&IO::AIO::poll_cb);
32 46
33 # Danga::Socket 47 # Danga::Socket integration
34 Danga::Socket->AddOtherFds (IO::AIO::poll_fileno => 48 Danga::Socket->AddOtherFds (IO::AIO::poll_fileno =>
35 \&IO::AIO::poll_cb); 49 \&IO::AIO::poll_cb);
36 50
37
38=head1 DESCRIPTION 51=head1 DESCRIPTION
39 52
40This module implements asynchronous I/O using whatever means your 53This module implements asynchronous I/O using whatever means your
41operating system supports. 54operating system supports.
42 55
43Currently, a number of threads are started that execute your read/writes 56Currently, a number of threads are started that execute your read/writes
44and signal their completion. You don't need thread support in your libc or 57and signal their completion. You don't need thread support in perl, and
45perl, and the threads created by this module will not be visible to the 58the threads created by this module will not be visible to perl. In the
46pthreads library. In the future, this module might make use of the native 59future, this module might make use of the native aio functions available
47aio functions available on many operating systems. However, they are often 60on many operating systems. However, they are often not well-supported
48not well-supported (Linux doesn't allow them on normal files currently, 61(Linux doesn't allow them on normal files currently, for example),
49for example), and they would only support aio_read and aio_write, so the 62and they would only support aio_read and aio_write, so the remaining
50remaining functionality would have to be implemented using threads anyway. 63functionality would have to be implemented using threads anyway.
51 64
52Although the module will work with in the presence of other threads, it is 65Although the module will work with in the presence of other threads,
53currently not reentrant, so use appropriate locking yourself, always call 66it is currently not reentrant in any way, so use appropriate locking
54C<poll_cb> from within the same thread, or never call C<poll_cb> (or other 67yourself, always call C<poll_cb> from within the same thread, or never
55C<aio_> functions) recursively. 68call C<poll_cb> (or other C<aio_> functions) recursively.
56 69
57=cut 70=cut
58 71
59package IO::AIO; 72package IO::AIO;
60 73
61no warnings; 74no warnings;
75use strict 'vars';
62 76
63use base 'Exporter'; 77use base 'Exporter';
64 78
65use Fcntl ();
66
67BEGIN { 79BEGIN {
68 $VERSION = 1.2; 80 our $VERSION = '2.0';
69 81
70 @EXPORT = qw(aio_read aio_write aio_open aio_close aio_stat aio_lstat aio_unlink 82 our @AIO_REQ = qw(aio_sendfile aio_read aio_write aio_open aio_close aio_stat
71 aio_rmdir aio_symlink aio_fsync aio_fdatasync aio_readahead); 83 aio_lstat aio_unlink aio_rmdir aio_readdir aio_scandir aio_symlink
72 @EXPORT_OK = qw(poll_fileno poll_cb min_parallel max_parallel max_outstanding nreqs); 84 aio_fsync aio_fdatasync aio_readahead aio_rename aio_link aio_move
85 aio_group aio_nop);
86 our @EXPORT = (@AIO_REQ, qw(aioreq_pri));
87 our @EXPORT_OK = qw(poll_fileno poll_cb poll_wait flush
88 min_parallel max_parallel max_outstanding nreqs);
89
90 @IO::AIO::GRP::ISA = 'IO::AIO::REQ';
73 91
74 require XSLoader; 92 require XSLoader;
75 XSLoader::load IO::AIO, $VERSION; 93 XSLoader::load ("IO::AIO", $VERSION);
76} 94}
77 95
78=head1 FUNCTIONS 96=head1 FUNCTIONS
79 97
80=head2 AIO FUNCTIONS 98=head2 AIO FUNCTIONS
88syscall has been executed asynchronously. 106syscall has been executed asynchronously.
89 107
90All functions expecting a filehandle keep a copy of the filehandle 108All functions expecting a filehandle keep a copy of the filehandle
91internally until the request has finished. 109internally until the request has finished.
92 110
111All requests return objects of type L<IO::AIO::REQ> that allow further
112manipulation of those requests while they are in-flight.
113
93The pathnames you pass to these routines I<must> be absolute and 114The pathnames you pass to these routines I<must> be absolute and
94encoded in byte form. The reason for the former is that at the time the 115encoded in byte form. The reason for the former is that at the time the
95request is being executed, the current working directory could have 116request is being executed, the current working directory could have
96changed. Alternatively, you can make sure that you never change the 117changed. Alternatively, you can make sure that you never change the
97current working directory. 118current working directory.
103environment, d) use Glib::filename_from_unicode on unicode filenames or e) 124environment, d) use Glib::filename_from_unicode on unicode filenames or e)
104use something else. 125use something else.
105 126
106=over 4 127=over 4
107 128
129=item aioreq_pri $pri
130
131Sets the priority for the next aio request. The default priority
132is C<0>, the minimum and maximum priorities are C<-4> and C<4>,
133respectively. Requests with higher priority will be serviced first.
134
135The priority will be reset to C<0> after each call to one of the C<aio_>
136functions.
137
108=item aio_open $pathname, $flags, $mode, $callback 138=item aio_open $pathname, $flags, $mode, $callback->($fh)
109 139
110Asynchronously open or create a file and call the callback with a newly 140Asynchronously open or create a file and call the callback with a newly
111created filehandle for the file. 141created filehandle for the file.
112 142
113The pathname passed to C<aio_open> must be absolute. See API NOTES, above, 143The pathname passed to C<aio_open> must be absolute. See API NOTES, above,
130 } else { 160 } else {
131 die "open failed: $!\n"; 161 die "open failed: $!\n";
132 } 162 }
133 }; 163 };
134 164
135=item aio_close $fh, $callback 165=item aio_close $fh, $callback->($status)
136 166
137Asynchronously close a file and call the callback with the result 167Asynchronously close a file and call the callback with the result
138code. I<WARNING:> although accepted, you should not pass in a perl 168code. I<WARNING:> although accepted, you should not pass in a perl
139filehandle here, as perl will likely close the file descriptor another 169filehandle here, as perl will likely close the file descriptor another
140time when the filehandle is destroyed. Normally, you can safely call perls 170time when the filehandle is destroyed. Normally, you can safely call perls
141C<close> or just let filehandles go out of scope. 171C<close> or just let filehandles go out of scope.
142 172
143This is supposed to be a bug in the API, so that might change. It's 173This is supposed to be a bug in the API, so that might change. It's
144therefore best to avoid this function. 174therefore best to avoid this function.
145 175
146=item aio_read $fh,$offset,$length, $data,$dataoffset,$callback 176=item aio_read $fh,$offset,$length, $data,$dataoffset, $callback->($retval)
147 177
148=item aio_write $fh,$offset,$length, $data,$dataoffset,$callback 178=item aio_write $fh,$offset,$length, $data,$dataoffset, $callback->($retval)
149 179
150Reads or writes C<length> bytes from the specified C<fh> and C<offset> 180Reads or writes C<length> bytes from the specified C<fh> and C<offset>
151into the scalar given by C<data> and offset C<dataoffset> and calls the 181into the scalar given by C<data> and offset C<dataoffset> and calls the
152callback without the actual number of bytes read (or -1 on error, just 182callback without the actual number of bytes read (or -1 on error, just
153like the syscall). 183like the syscall).
154 184
185The C<$data> scalar I<MUST NOT> be modified in any way while the request
186is outstanding. Modifying it can result in segfaults or WW3 (if the
187necessary/optional hardware is installed).
188
155Example: Read 15 bytes at offset 7 into scalar C<$buffer>, starting at 189Example: Read 15 bytes at offset 7 into scalar C<$buffer>, starting at
156offset C<0> within the scalar: 190offset C<0> within the scalar:
157 191
158 aio_read $fh, 7, 15, $buffer, 0, sub { 192 aio_read $fh, 7, 15, $buffer, 0, sub {
159 $_[0] > 0 or die "read error: $!"; 193 $_[0] > 0 or die "read error: $!";
160 print "read $_[0] bytes: <$buffer>\n"; 194 print "read $_[0] bytes: <$buffer>\n";
161 }; 195 };
162 196
197=item aio_move $srcpath, $dstpath, $callback->($status)
198
199[EXPERIMENTAL due to internal aio_group use]
200
201Try to move the I<file> (directories not supported as either source or
202destination) from C<$srcpath> to C<$dstpath> and call the callback with
203the C<0> (error) or C<-1> ok.
204
205This is a composite request that tries to rename(2) the file first. If
206rename files with C<EXDEV>, it creates the destination file with mode 0200
207and copies the contents of the source file into it using C<aio_sendfile>,
208followed by restoring atime, mtime, access mode and uid/gid, in that
209order, and unlinking the C<$srcpath>.
210
211If an error occurs, the partial destination file will be unlinked, if
212possible, except when setting atime, mtime, access mode and uid/gid, where
213errors are being ignored.
214
215=cut
216
217sub aio_move($$$) {
218 my ($src, $dst, $cb) = @_;
219
220 my $grp = aio_group $cb;
221
222 add $grp aio_rename $src, $dst, sub {
223 if ($_[0] && $! == EXDEV) {
224 add $grp aio_open $src, O_RDONLY, 0, sub {
225 if (my $src_fh = $_[0]) {
226 my @stat = stat $src_fh;
227
228 add $grp aio_open $dst, O_WRONLY, 0200, sub {
229 if (my $dst_fh = $_[0]) {
230 add $grp aio_sendfile $dst_fh, $src_fh, 0, $stat[7], sub {
231 close $src_fh;
232
233 if ($_[0] == $stat[7]) {
234 utime $stat[8], $stat[9], $dst;
235 chmod $stat[2] & 07777, $dst_fh;
236 chown $stat[4], $stat[5], $dst_fh;
237 close $dst_fh;
238
239 add $grp aio_unlink $src, sub {
240 $grp->result ($_[0]);
241 };
242 } else {
243 my $errno = $!;
244 add $grp aio_unlink $dst, sub {
245 $! = $errno;
246 $grp->result (-1);
247 };
248 }
249 };
250 } else {
251 $grp->result (-1);
252 }
253 },
254
255 } else {
256 $grp->result (-1);
257 }
258 };
259 } else {
260 $grp->result ($_[0]);
261 }
262 };
263
264 $grp
265}
266
267=item aio_sendfile $out_fh, $in_fh, $in_offset, $length, $callback->($retval)
268
269Tries to copy C<$length> bytes from C<$in_fh> to C<$out_fh>. It starts
270reading at byte offset C<$in_offset>, and starts writing at the current
271file offset of C<$out_fh>. Because of that, it is not safe to issue more
272than one C<aio_sendfile> per C<$out_fh>, as they will interfere with each
273other.
274
275This call tries to make use of a native C<sendfile> syscall to provide
276zero-copy operation. For this to work, C<$out_fh> should refer to a
277socket, and C<$in_fh> should refer to mmap'able file.
278
279If the native sendfile call fails or is not implemented, it will be
280emulated, so you can call C<aio_sendfile> on any type of filehandle
281regardless of the limitations of the operating system.
282
283Please note, however, that C<aio_sendfile> can read more bytes from
284C<$in_fh> than are written, and there is no way to find out how many
285bytes have been read from C<aio_sendfile> alone, as C<aio_sendfile> only
286provides the number of bytes written to C<$out_fh>. Only if the result
287value equals C<$length> one can assume that C<$length> bytes have been
288read.
289
163=item aio_readahead $fh,$offset,$length, $callback 290=item aio_readahead $fh,$offset,$length, $callback->($retval)
164 291
165C<aio_readahead> populates the page cache with data from a file so that 292C<aio_readahead> populates the page cache with data from a file so that
166subsequent reads from that file will not block on disk I/O. The C<$offset> 293subsequent reads from that file will not block on disk I/O. The C<$offset>
167argument specifies the starting point from which data is to be read and 294argument specifies the starting point from which data is to be read and
168C<$length> specifies the number of bytes to be read. I/O is performed in 295C<$length> specifies the number of bytes to be read. I/O is performed in
172file. The current file offset of the file is left unchanged. 299file. The current file offset of the file is left unchanged.
173 300
174If that syscall doesn't exist (likely if your OS isn't Linux) it will be 301If that syscall doesn't exist (likely if your OS isn't Linux) it will be
175emulated by simply reading the data, which would have a similar effect. 302emulated by simply reading the data, which would have a similar effect.
176 303
177=item aio_stat $fh_or_path, $callback 304=item aio_stat $fh_or_path, $callback->($status)
178 305
179=item aio_lstat $fh, $callback 306=item aio_lstat $fh, $callback->($status)
180 307
181Works like perl's C<stat> or C<lstat> in void context. The callback will 308Works like perl's C<stat> or C<lstat> in void context. The callback will
182be called after the stat and the results will be available using C<stat _> 309be called after the stat and the results will be available using C<stat _>
183or C<-s _> etc... 310or C<-s _> etc...
184 311
194 aio_stat "/etc/passwd", sub { 321 aio_stat "/etc/passwd", sub {
195 $_[0] and die "stat failed: $!"; 322 $_[0] and die "stat failed: $!";
196 print "size is ", -s _, "\n"; 323 print "size is ", -s _, "\n";
197 }; 324 };
198 325
199=item aio_unlink $pathname, $callback 326=item aio_unlink $pathname, $callback->($status)
200 327
201Asynchronously unlink (delete) a file and call the callback with the 328Asynchronously unlink (delete) a file and call the callback with the
202result code. 329result code.
203 330
331=item aio_link $srcpath, $dstpath, $callback->($status)
332
333Asynchronously create a new link to the existing object at C<$srcpath> at
334the path C<$dstpath> and call the callback with the result code.
335
336=item aio_symlink $srcpath, $dstpath, $callback->($status)
337
338Asynchronously create a new symbolic link to the existing object at C<$srcpath> at
339the path C<$dstpath> and call the callback with the result code.
340
341=item aio_rename $srcpath, $dstpath, $callback->($status)
342
343Asynchronously rename the object at C<$srcpath> to C<$dstpath>, just as
344rename(2) and call the callback with the result code.
345
204=item aio_rmdir $pathname, $callback 346=item aio_rmdir $pathname, $callback->($status)
205 347
206Asynchronously rmdir (delete) a directory and call the callback with the 348Asynchronously rmdir (delete) a directory and call the callback with the
207result code. 349result code.
208 350
351=item aio_readdir $pathname, $callback->($entries)
352
353Unlike the POSIX call of the same name, C<aio_readdir> reads an entire
354directory (i.e. opendir + readdir + closedir). The entries will not be
355sorted, and will B<NOT> include the C<.> and C<..> entries.
356
357The callback a single argument which is either C<undef> or an array-ref
358with the filenames.
359
360=item aio_scandir $path, $maxreq, $callback->($dirs, $nondirs)
361
362[EXPERIMENTAL due to internal aio_group use]
363
364Scans a directory (similar to C<aio_readdir>) but additionally tries to
365separate the entries of directory C<$path> into two sets of names, ones
366you can recurse into (directories or links to them), and ones you cannot
367recurse into (everything else).
368
369C<aio_scandir> is a composite request that creates of many sub requests_
370C<$maxreq> specifies the maximum number of outstanding aio requests that
371this function generates. If it is C<< <= 0 >>, then a suitable default
372will be chosen (currently 6).
373
374On error, the callback is called without arguments, otherwise it receives
375two array-refs with path-relative entry names.
376
377Example:
378
379 aio_scandir $dir, 0, sub {
380 my ($dirs, $nondirs) = @_;
381 print "real directories: @$dirs\n";
382 print "everything else: @$nondirs\n";
383 };
384
385Implementation notes.
386
387The C<aio_readdir> cannot be avoided, but C<stat()>'ing every entry can.
388
389After reading the directory, the modification time, size etc. of the
390directory before and after the readdir is checked, and if they match (and
391isn't the current time), the link count will be used to decide how many
392entries are directories (if >= 2). Otherwise, no knowledge of the number
393of subdirectories will be assumed.
394
395Then entries will be sorted into likely directories (everything without
396a non-initial dot currently) and likely non-directories (everything
397else). Then every entry plus an appended C</.> will be C<stat>'ed,
398likely directories first. If that succeeds, it assumes that the entry
399is a directory or a symlink to directory (which will be checked
400seperately). This is often faster than stat'ing the entry itself because
401filesystems might detect the type of the entry without reading the inode
402data (e.g. ext2fs filetype feature).
403
404If the known number of directories (link count - 2) has been reached, the
405rest of the entries is assumed to be non-directories.
406
407This only works with certainty on POSIX (= UNIX) filesystems, which
408fortunately are the vast majority of filesystems around.
409
410It will also likely work on non-POSIX filesystems with reduced efficiency
411as those tend to return 0 or 1 as link counts, which disables the
412directory counting heuristic.
413
414=cut
415
416sub aio_scandir($$$) {
417 my ($path, $maxreq, $cb) = @_;
418
419 my $grp = aio_group $cb;
420
421 $maxreq = 6 if $maxreq <= 0;
422
423 # stat once
424 add $grp aio_stat $path, sub {
425 return $grp->result () if $_[0];
426 my $now = time;
427 my $hash1 = join ":", (stat _)[0,1,3,7,9];
428
429 # read the directory entries
430 add $grp aio_readdir $path, sub {
431 my $entries = shift
432 or return $grp->result ();
433
434 # stat the dir another time
435 add $grp aio_stat $path, sub {
436 my $hash2 = join ":", (stat _)[0,1,3,7,9];
437
438 my $ndirs;
439
440 # take the slow route if anything looks fishy
441 if ($hash1 ne $hash2 or (stat _)[9] == $now) {
442 $ndirs = -1;
443 } else {
444 # if nlink == 2, we are finished
445 # on non-posix-fs's, we rely on nlink < 2
446 $ndirs = (stat _)[3] - 2
447 or return $grp->result ([], $entries);
448 }
449
450 # sort into likely dirs and likely nondirs
451 # dirs == files without ".", short entries first
452 $entries = [map $_->[0],
453 sort { $b->[1] cmp $a->[1] }
454 map [$_, sprintf "%s%04d", (/.\./ ? "1" : "0"), length],
455 @$entries];
456
457 my (@dirs, @nondirs);
458
459 my ($statcb, $schedcb);
460 my $nreq = 0;
461
462 my $statgrp = add $grp aio_group;
463
464 $schedcb = sub {
465 if (@$entries) {
466 if ($nreq < $maxreq) {
467 my $ent = pop @$entries;
468 $nreq++;
469 add $statgrp aio_stat "$path/$ent/.", sub { $statcb->($_[0], $ent) };
470 }
471 } elsif (!$nreq) {
472 # finished
473 $statgrp->cancel;
474 undef $statcb;
475 undef $schedcb;
476 $grp->result (\@dirs, \@nondirs);
477 }
478 };
479 $statcb = sub {
480 my ($status, $entry) = @_;
481
482 if ($status < 0) {
483 $nreq--;
484 push @nondirs, $entry;
485 &$schedcb;
486 } else {
487 # need to check for real directory
488 add $grp aio_lstat "$path/$entry", sub {
489 $nreq--;
490
491 if (-d _) {
492 push @dirs, $entry;
493
494 if (!--$ndirs) {
495 push @nondirs, @$entries;
496 $entries = [];
497 }
498 } else {
499 push @nondirs, $entry;
500 }
501
502 &$schedcb;
503 }
504 }
505 };
506
507 &$schedcb while @$entries && $nreq < $maxreq;
508 };
509 };
510 };
511
512 $grp
513}
514
209=item aio_fsync $fh, $callback 515=item aio_fsync $fh, $callback->($status)
210 516
211Asynchronously call fsync on the given filehandle and call the callback 517Asynchronously call fsync on the given filehandle and call the callback
212with the fsync result code. 518with the fsync result code.
213 519
214=item aio_fdatasync $fh, $callback 520=item aio_fdatasync $fh, $callback->($status)
215 521
216Asynchronously call fdatasync on the given filehandle and call the 522Asynchronously call fdatasync on the given filehandle and call the
217callback with the fdatasync result code. 523callback with the fdatasync result code.
218 524
219If this call isn't available because your OS lacks it or it couldn't be 525If this call isn't available because your OS lacks it or it couldn't be
220detected, it will be emulated by calling C<fsync> instead. 526detected, it will be emulated by calling C<fsync> instead.
527
528=item aio_group $callback->(...)
529
530[EXPERIMENTAL]
531
532This is a very special aio request: Instead of doing something, it is a
533container for other aio requests, which is useful if you want to bundle
534many requests into a single, composite, request.
535
536Returns an object of class L<IO::AIO::GRP>. See its documentation below
537for more info.
538
539Example:
540
541 my $grp = aio_group sub {
542 print "all stats done\n";
543 };
544
545 add $grp
546 (aio_stat ...),
547 (aio_stat ...),
548 ...;
549
550=item aio_nop $callback->()
551
552This is a special request - it does nothing in itself and is only used for
553side effects, such as when you want to add a dummy request to a group so
554that finishing the requests in the group depends on executing the given
555code.
556
557While this request does nothing, it still goes through the execution
558phase and still requires a worker thread. Thus, the callback will not
559be executed immediately but only after other requests in the queue have
560entered their execution phase. This can be used to measure request
561latency.
562
563=item IO::AIO::aio_sleep $fractional_seconds, $callback->() *NOT EXPORTED*
564
565Mainly used for debugging and benchmarking, this aio request puts one of
566the request workers to sleep for the given time.
567
568While it is theoretically handy to have simple I/O scheduling requests
569like sleep and file handle readable/writable, the overhead this creates
570is immense, so do not use this function except to put your application
571under artificial I/O pressure.
572
573=back
574
575=head2 IO::AIO::REQ CLASS
576
577All non-aggregate C<aio_*> functions return an object of this class when
578called in non-void context.
579
580A request always moves through the following five states in its lifetime,
581in order: B<ready> (request has been created, but has not been executed
582yet), B<execute> (request is currently being executed), B<pending>
583(request has been executed but callback has not been called yet),
584B<result> (results are being processed synchronously, includes calling the
585callback) and B<done> (request has reached the end of its lifetime and
586holds no resources anymore).
587
588=over 4
589
590=item cancel $req
591
592Cancels the request, if possible. Has the effect of skipping execution
593when entering the B<execute> state and skipping calling the callback when
594entering the the B<result> state, but will leave the request otherwise
595untouched. That means that requests that currently execute will not be
596stopped and resources held by the request will not be freed prematurely.
597
598=item cb $req $callback->(...)
599
600Replace (or simply set) the callback registered to the request.
601
602=back
603
604=head2 IO::AIO::GRP CLASS
605
606This class is a subclass of L<IO::AIO::REQ>, so all its methods apply to
607objects of this class, too.
608
609A IO::AIO::GRP object is a special request that can contain multiple other
610aio requests.
611
612You create one by calling the C<aio_group> constructing function with a
613callback that will be called when all contained requests have entered the
614C<done> state:
615
616 my $grp = aio_group sub {
617 print "all requests are done\n";
618 };
619
620You add requests by calling the C<add> method with one or more
621C<IO::AIO::REQ> objects:
622
623 $grp->add (aio_unlink "...");
624
625 add $grp aio_stat "...", sub {
626 $_[0] or return $grp->result ("error");
627
628 # add another request dynamically, if first succeeded
629 add $grp aio_open "...", sub {
630 $grp->result ("ok");
631 };
632 };
633
634This makes it very easy to create composite requests (see the source of
635C<aio_move> for an application) that work and feel like simple requests.
636
637=over 4
638
639=item * The IO::AIO::GRP objects will be cleaned up during calls to
640C<IO::AIO::poll_cb>, just like any other request.
641
642=item * They can be canceled like any other request. Canceling will cancel not
643only the request itself, but also all requests it contains.
644
645=item * They can also can also be added to other IO::AIO::GRP objects.
646
647=item * You must not add requests to a group from within the group callback (or
648any later time).
649
650=item * This does not harmonise well with C<max_outstanding>, so best do
651not combine C<aio_group> with it. Groups and feeders are recommended for
652this kind of concurrency-limiting.
653
654=back
655
656Their lifetime, simplified, looks like this: when they are empty, they
657will finish very quickly. If they contain only requests that are in the
658C<done> state, they will also finish. Otherwise they will continue to
659exist.
660
661That means after creating a group you have some time to add requests. And
662in the callbacks of those requests, you can add further requests to the
663group. And only when all those requests have finished will the the group
664itself finish.
665
666=over 4
667
668=item add $grp ...
669
670=item $grp->add (...)
671
672Add one or more requests to the group. Any type of L<IO::AIO::REQ> can
673be added, including other groups, as long as you do not create circular
674dependencies.
675
676Returns all its arguments.
677
678=item $grp->result (...)
679
680Set the result value(s) that will be passed to the group callback when all
681subrequests have finished. By default, no argument will be passed.
682
683=item feed $grp $callback->($grp)
684
685[VERY EXPERIMENTAL]
686
687Sets a feeder/generator on this group: every group can have an attached
688generator that generates requests if idle. The idea behind this is that,
689although you could just queue as many requests as you want in a group,
690this might starve other requests for a potentially long time. For
691example, C<aio_scandir> might generate hundreds of thousands C<aio_stat>
692requests, delaying any later requests for a long time.
693
694To avoid this, and allow incremental generation of requests, you can
695instead a group and set a feeder on it that generates those requests. The
696feed callback will be called whenever there are few enough (see C<limit>,
697below) requests active in the group itself and is expected to queue more
698requests.
699
700The feed callback can queue as many requests as it likes (i.e. C<add> does
701not impose any limits).
702
703If the feed does not queue more requests when called, it will be
704automatically removed from the group.
705
706If the feed limit is C<0>, it will be set to C<2> automatically.
707
708Example:
709
710 # stat all files in @files, but only ever use four aio requests concurrently:
711
712 my $grp = aio_group sub { print "finished\n" };
713 limit $grp 4;
714 feed $grp sub {
715 my $file = pop @files
716 or return;
717
718 add $grp aio_stat $file, sub { ... };
719 };
720
721=item limit $grp $num
722
723Sets the feeder limit for the group: The feeder will be called whenever
724the group contains less than this many requests.
725
726Setting the limit to C<0> will pause the feeding process.
221 727
222=back 728=back
223 729
224=head2 SUPPORT FUNCTIONS 730=head2 SUPPORT FUNCTIONS
225 731
283 IO::AIO::poll_wait, IO::AIO::poll_cb 789 IO::AIO::poll_wait, IO::AIO::poll_cb
284 if IO::AIO::nreqs; 790 if IO::AIO::nreqs;
285 791
286=item IO::AIO::min_parallel $nthreads 792=item IO::AIO::min_parallel $nthreads
287 793
288Set the minimum number of AIO threads to C<$nthreads>. The default is 794Set the minimum number of AIO threads to C<$nthreads>. The current
289C<1>, which means a single asynchronous operation can be done at one time 795default is C<8>, which means eight asynchronous operations can execute
290(the number of outstanding operations, however, is unlimited). 796concurrently at any one time (the number of outstanding requests,
797however, is unlimited).
291 798
799IO::AIO starts threads only on demand, when an AIO request is queued and
800no free thread exists.
801
292It is recommended to keep the number of threads low, as some Linux 802It is recommended to keep the number of threads relatively low, as some
293kernel versions will scale negatively with the number of threads (higher 803Linux kernel versions will scale negatively with the number of threads
294parallelity => MUCH higher latency). With current Linux 2.6 versions, 4-32 804(higher parallelity => MUCH higher latency). With current Linux 2.6
295threads should be fine. 805versions, 4-32 threads should be fine.
296 806
297Under normal circumstances you don't need to call this function, as this 807Under most circumstances you don't need to call this function, as the
298module automatically starts some threads (the exact number might change, 808module selects a default that is suitable for low to moderate load.
299and is currently 4).
300 809
301=item IO::AIO::max_parallel $nthreads 810=item IO::AIO::max_parallel $nthreads
302 811
303Sets the maximum number of AIO threads to C<$nthreads>. If more than 812Sets the maximum number of AIO threads to C<$nthreads>. If more than the
304the specified number of threads are currently running, kill them. This 813specified number of threads are currently running, this function kills
305function blocks until the limit is reached. 814them. This function blocks until the limit is reached.
815
816While C<$nthreads> are zero, aio requests get queued but not executed
817until the number of threads has been increased again.
306 818
307This module automatically runs C<max_parallel 0> at program end, to ensure 819This module automatically runs C<max_parallel 0> at program end, to ensure
308that all threads are killed and that there are no outstanding requests. 820that all threads are killed and that there are no outstanding requests.
309 821
310Under normal circumstances you don't need to call this function. 822Under normal circumstances you don't need to call this function.
311 823
312=item $oldnreqs = IO::AIO::max_outstanding $nreqs 824=item $oldnreqs = IO::AIO::max_outstanding $nreqs
825
826[DEPRECATED]
313 827
314Sets the maximum number of outstanding requests to C<$nreqs>. If you 828Sets the maximum number of outstanding requests to C<$nreqs>. If you
315try to queue up more than this number of requests, the caller will block until 829try to queue up more than this number of requests, the caller will block until
316some requests have been handled. 830some requests have been handled.
317 831
318The default is very large, so normally there is no practical limit. If you 832The default is very large, so normally there is no practical limit. If you
319queue up many requests in a loop it it often improves speed if you set 833queue up many requests in a loop it often improves speed if you set
320this to a relatively low number, such as C<100>. 834this to a relatively low number, such as C<100>.
835
836This function does not work well together with C<aio_group>'s, and their
837feeder interface is better suited to limiting concurrency, so do not use
838this function.
321 839
322Under normal circumstances you don't need to call this function. 840Under normal circumstances you don't need to call this function.
323 841
324=back 842=back
325 843
339 or return undef; 857 or return undef;
340 858
341 *$sym 859 *$sym
342} 860}
343 861
344min_parallel 4; 862min_parallel 8;
345 863
346END { 864END {
347 max_parallel 0; 865 max_parallel 0;
348} 866}
349 867
3501; 8681;
351 869
352=head2 FORK BEHAVIOUR 870=head2 FORK BEHAVIOUR
353 871
354Before the fork IO::AIO first handles all outstanding requests - if other 872This module should do "the right thing" when the process using it forks:
355threads add requests during this period, this time is prolonged. It then 873
356enters a quiescent state where no requests can be added in other threads 874Before the fork, IO::AIO enters a quiescent state where no requests
357and no results will be processed. After the fork the parent simply leaves 875can be added in other threads and no results will be processed. After
358the quiescent state and continues request processing, while the child 876the fork the parent simply leaves the quiescent state and continues
359starts the same number of threads as were in use by the parent. 877request/result processing, while the child clears the request/result
878queue (so the requests started before the fork will only be handled in
879the parent). Threads will be started on demand until the limit ste in the
880parent process has been reached again.
881
882In short: the parent will, after a short pause, continue as if fork had
883not been called, while the child will act as if IO::AIO has not been used
884yet.
885
886=head2 MEMORY USAGE
887
888Each aio request uses - depending on your architecture - around 128 bytes
889of memory. In addition, stat requests need a stat buffer (possibly a few
890hundred bytes). Perl scalars and other data passed into aio requests will
891also be locked.
892
893This is now awfully much, so queuing lots of requests is not usually a
894problem.
895
896Each thread needs a stack area which is usually around 16k, sometimes much
897larger, depending on the OS.
360 898
361=head1 SEE ALSO 899=head1 SEE ALSO
362 900
363L<Coro>, L<Linux::AIO>. 901L<Coro::AIO>.
364 902
365=head1 AUTHOR 903=head1 AUTHOR
366 904
367 Marc Lehmann <schmorp@schmorp.de> 905 Marc Lehmann <schmorp@schmorp.de>
368 http://home.schmorp.de/ 906 http://home.schmorp.de/

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines