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 1.70 by root, Tue Oct 24 03:40:38 2006 UTC vs.
Revision 1.97 by root, Sat Dec 23 04:49:37 2006 UTC

5=head1 SYNOPSIS 5=head1 SYNOPSIS
6 6
7 use IO::AIO; 7 use IO::AIO;
8 8
9 aio_open "/etc/passwd", O_RDONLY, 0, sub { 9 aio_open "/etc/passwd", O_RDONLY, 0, sub {
10 my ($fh) = @_; 10 my $fh = shift
11 or die "/etc/passwd: $!";
11 ... 12 ...
12 }; 13 };
13 14
14 aio_unlink "/tmp/file", sub { }; 15 aio_unlink "/tmp/file", sub { };
15 16
51=head1 DESCRIPTION 52=head1 DESCRIPTION
52 53
53This module implements asynchronous I/O using whatever means your 54This module implements asynchronous I/O using whatever means your
54operating system supports. 55operating system supports.
55 56
57Asynchronous means that operations that can normally block your program
58(e.g. reading from disk) will be done asynchronously: the operation
59will still block, but you can do something else in the meantime. This
60is extremely useful for programs that need to stay interactive even
61when doing heavy I/O (GUI programs, high performance network servers
62etc.), but can also be used to easily do operations in parallel that are
63normally done sequentially, e.g. stat'ing many files, which is much faster
64on a RAID volume or over NFS when you do a number of stat operations
65concurrently.
66
67While most of this works on all types of file descriptors (for example
68sockets), using these functions on file descriptors that support
69nonblocking operation (again, sockets, pipes etc.) is very inefficient or
70might not work (aio_read fails on sockets/pipes/fifos). Use an event loop
71for that (such as the L<Event|Event> module): IO::AIO will naturally fit
72into such an event loop itself.
73
56Currently, a number of threads are started that execute your read/writes 74In this version, a number of threads are started that execute your
57and signal their completion. You don't need thread support in perl, and 75requests and signal their completion. You don't need thread support
58the threads created by this module will not be visible to perl. In the 76in perl, and the threads created by this module will not be visible
59future, this module might make use of the native aio functions available 77to perl. In the future, this module might make use of the native aio
60on many operating systems. However, they are often not well-supported 78functions available on many operating systems. However, they are often
61(Linux doesn't allow them on normal files currently, for example), 79not well-supported or restricted (GNU/Linux doesn't allow them on normal
62and they would only support aio_read and aio_write, so the remaining 80files currently, for example), and they would only support aio_read and
63functionality would have to be implemented using threads anyway. 81aio_write, so the remaining functionality would have to be implemented
82using threads anyway.
64 83
65Although the module will work with in the presence of other threads, 84Although the module will work with in the presence of other (Perl-)
66it is currently not reentrant in any way, so use appropriate locking 85threads, it is currently not reentrant in any way, so use appropriate
67yourself, always call C<poll_cb> from within the same thread, or never 86locking yourself, always call C<poll_cb> from within the same thread, or
68call C<poll_cb> (or other C<aio_> functions) recursively. 87never call C<poll_cb> (or other C<aio_> functions) recursively.
88
89=head2 EXAMPLE
90
91This is a simple example that uses the Event module and loads
92F</etc/passwd> asynchronously:
93
94 use Fcntl;
95 use Event;
96 use IO::AIO;
97
98 # register the IO::AIO callback with Event
99 Event->io (fd => IO::AIO::poll_fileno,
100 poll => 'r',
101 cb => \&IO::AIO::poll_cb);
102
103 # queue the request to open /etc/passwd
104 aio_open "/etc/passwd", O_RDONLY, 0, sub {
105 my $fh = shift
106 or die "error while opening: $!";
107
108 # stat'ing filehandles is generally non-blocking
109 my $size = -s $fh;
110
111 # queue a request to read the file
112 my $contents;
113 aio_read $fh, 0, $size, $contents, 0, sub {
114 $_[0] == $size
115 or die "short read: $!";
116
117 close $fh;
118
119 # file contents now in $contents
120 print $contents;
121
122 # exit event loop and program
123 Event::unloop;
124 };
125 };
126
127 # possibly queue up other requests, or open GUI windows,
128 # check for sockets etc. etc.
129
130 # process events as long as there are some:
131 Event::loop;
132
133=head1 REQUEST ANATOMY AND LIFETIME
134
135Every C<aio_*> function creates a request. which is a C data structure not
136directly visible to Perl.
137
138If called in non-void context, every request function returns a Perl
139object representing the request. In void context, nothing is returned,
140which saves a bit of memory.
141
142The perl object is a fairly standard ref-to-hash object. The hash contents
143are not used by IO::AIO so you are free to store anything you like in it.
144
145During their existance, aio requests travel through the following states,
146in order:
147
148=over 4
149
150=item ready
151
152Immediately after a request is created it is put into the ready state,
153waiting for a thread to execute it.
154
155=item execute
156
157A thread has accepted the request for processing and is currently
158executing it (e.g. blocking in read).
159
160=item pending
161
162The request has been executed and is waiting for result processing.
163
164While request submission and execution is fully asynchronous, result
165processing is not and relies on the perl interpreter calling C<poll_cb>
166(or another function with the same effect).
167
168=item result
169
170The request results are processed synchronously by C<poll_cb>.
171
172The C<poll_cb> function will process all outstanding aio requests by
173calling their callbacks, freeing memory associated with them and managing
174any groups they are contained in.
175
176=item done
177
178Request has reached the end of its lifetime and holds no resources anymore
179(except possibly for the Perl object, but its connection to the actual
180aio request is severed and calling its methods will either do nothing or
181result in a runtime error).
182
183=back
69 184
70=cut 185=cut
71 186
72package IO::AIO; 187package IO::AIO;
73 188
75use strict 'vars'; 190use strict 'vars';
76 191
77use base 'Exporter'; 192use base 'Exporter';
78 193
79BEGIN { 194BEGIN {
80 our $VERSION = '2.0'; 195 our $VERSION = '2.3';
81 196
82 our @AIO_REQ = qw(aio_sendfile aio_read aio_write aio_open aio_close aio_stat 197 our @AIO_REQ = qw(aio_sendfile aio_read aio_write aio_open aio_close aio_stat
83 aio_lstat aio_unlink aio_rmdir aio_readdir aio_scandir aio_symlink 198 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 199 aio_readlink aio_fsync aio_fdatasync aio_readahead aio_rename aio_link
85 aio_group aio_nop); 200 aio_move aio_copy aio_group aio_nop aio_mknod);
86 our @EXPORT = (@AIO_REQ, qw(aioreq_pri aioreq_nice)); 201 our @EXPORT = (@AIO_REQ, qw(aioreq_pri aioreq_nice aio_block));
87 our @EXPORT_OK = qw(poll_fileno poll_cb poll_wait flush 202 our @EXPORT_OK = qw(poll_fileno poll_cb poll_wait flush
88 min_parallel max_parallel max_outstanding nreqs); 203 min_parallel max_parallel max_idle
204 nreqs nready npending nthreads
205 max_poll_time max_poll_reqs);
89 206
90 @IO::AIO::GRP::ISA = 'IO::AIO::REQ'; 207 @IO::AIO::GRP::ISA = 'IO::AIO::REQ';
91 208
92 require XSLoader; 209 require XSLoader;
93 XSLoader::load ("IO::AIO", $VERSION); 210 XSLoader::load ("IO::AIO", $VERSION);
94} 211}
95 212
96=head1 FUNCTIONS 213=head1 FUNCTIONS
97 214
98=head2 AIO FUNCTIONS 215=head2 AIO REQUEST FUNCTIONS
99 216
100All the C<aio_*> calls are more or less thin wrappers around the syscall 217All the C<aio_*> calls are more or less thin wrappers around the syscall
101with the same name (sans C<aio_>). The arguments are similar or identical, 218with the same name (sans C<aio_>). The arguments are similar or identical,
102and they all accept an additional (and optional) C<$callback> argument 219and they all accept an additional (and optional) C<$callback> argument
103which must be a code reference. This code reference will get called with 220which must be a code reference. This code reference will get called with
106syscall has been executed asynchronously. 223syscall has been executed asynchronously.
107 224
108All functions expecting a filehandle keep a copy of the filehandle 225All functions expecting a filehandle keep a copy of the filehandle
109internally until the request has finished. 226internally until the request has finished.
110 227
111All requests return objects of type L<IO::AIO::REQ> that allow further 228All functions return request objects of type L<IO::AIO::REQ> that allow
112manipulation of those requests while they are in-flight. 229further manipulation of those requests while they are in-flight.
113 230
114The pathnames you pass to these routines I<must> be absolute and 231The pathnames you pass to these routines I<must> be absolute and
115encoded in byte form. The reason for the former is that at the time the 232encoded as octets. The reason for the former is that at the time the
116request is being executed, the current working directory could have 233request is being executed, the current working directory could have
117changed. Alternatively, you can make sure that you never change the 234changed. Alternatively, you can make sure that you never change the
118current working directory. 235current working directory anywhere in the program and then use relative
236paths.
119 237
120To encode pathnames to byte form, either make sure you either: a) 238To encode pathnames as octets, either make sure you either: a) always pass
121always pass in filenames you got from outside (command line, readdir 239in filenames you got from outside (command line, readdir etc.) without
122etc.), b) are ASCII or ISO 8859-1, c) use the Encode module and encode 240tinkering, b) are ASCII or ISO 8859-1, c) use the Encode module and encode
123your pathnames to the locale (or other) encoding in effect in the user 241your pathnames to the locale (or other) encoding in effect in the user
124environment, d) use Glib::filename_from_unicode on unicode filenames or e) 242environment, d) use Glib::filename_from_unicode on unicode filenames or e)
125use something else. 243use something else to ensure your scalar has the correct contents.
244
245This works, btw. independent of the internal UTF-8 bit, which IO::AIO
246handles correctly wether it is set or not.
126 247
127=over 4 248=over 4
128 249
129=item aioreq_pri $pri 250=item $prev_pri = aioreq_pri [$pri]
130 251
131Sets the priority for the next aio request. The default priority 252Returns the priority value that would be used for the next request and, if
253C<$pri> is given, sets the priority for the next aio request.
254
132is C<0>, the minimum and maximum priorities are C<-4> and C<4>, 255The default priority is C<0>, the minimum and maximum priorities are C<-4>
133respectively. Requests with higher priority will be serviced first. 256and C<4>, respectively. Requests with higher priority will be serviced
257first.
134 258
135The priority will be reset to C<0> after each call to one of the C<aio_> 259The priority will be reset to C<0> after each call to one of the C<aio_*>
136functions. 260functions.
137 261
138Example: open a file with low priority, then read something from it with 262Example: open a file with low priority, then read something from it with
139higher priority so the read request is serviced before other low priority 263higher priority so the read request is serviced before other low priority
140open requests (potentially spamming the cache): 264open requests (potentially spamming the cache):
150 }; 274 };
151 275
152=item aioreq_nice $pri_adjust 276=item aioreq_nice $pri_adjust
153 277
154Similar to C<aioreq_pri>, but subtracts the given value from the current 278Similar to C<aioreq_pri>, but subtracts the given value from the current
155priority, so effects are cumulative. 279priority, so the effect is cumulative.
156 280
157=item aio_open $pathname, $flags, $mode, $callback->($fh) 281=item aio_open $pathname, $flags, $mode, $callback->($fh)
158 282
159Asynchronously open or create a file and call the callback with a newly 283Asynchronously open or create a file and call the callback with a newly
160created filehandle for the file. 284created filehandle for the file.
211 aio_read $fh, 7, 15, $buffer, 0, sub { 335 aio_read $fh, 7, 15, $buffer, 0, sub {
212 $_[0] > 0 or die "read error: $!"; 336 $_[0] > 0 or die "read error: $!";
213 print "read $_[0] bytes: <$buffer>\n"; 337 print "read $_[0] bytes: <$buffer>\n";
214 }; 338 };
215 339
216=item aio_move $srcpath, $dstpath, $callback->($status)
217
218[EXPERIMENTAL due to internal aio_group use]
219
220Try to move the I<file> (directories not supported as either source or
221destination) from C<$srcpath> to C<$dstpath> and call the callback with
222the C<0> (error) or C<-1> ok.
223
224This is a composite request that tries to rename(2) the file first. If
225rename files with C<EXDEV>, it creates the destination file with mode 0200
226and copies the contents of the source file into it using C<aio_sendfile>,
227followed by restoring atime, mtime, access mode and uid/gid, in that
228order, and unlinking the C<$srcpath>.
229
230If an error occurs, the partial destination file will be unlinked, if
231possible, except when setting atime, mtime, access mode and uid/gid, where
232errors are being ignored.
233
234=cut
235
236sub aio_move($$$) {
237 my ($src, $dst, $cb) = @_;
238
239 my $grp = aio_group $cb;
240
241 add $grp aio_rename $src, $dst, sub {
242 if ($_[0] && $! == EXDEV) {
243 add $grp aio_open $src, O_RDONLY, 0, sub {
244 if (my $src_fh = $_[0]) {
245 my @stat = stat $src_fh;
246
247 add $grp aio_open $dst, O_WRONLY, 0200, sub {
248 if (my $dst_fh = $_[0]) {
249 add $grp aio_sendfile $dst_fh, $src_fh, 0, $stat[7], sub {
250 close $src_fh;
251
252 if ($_[0] == $stat[7]) {
253 utime $stat[8], $stat[9], $dst;
254 chmod $stat[2] & 07777, $dst_fh;
255 chown $stat[4], $stat[5], $dst_fh;
256 close $dst_fh;
257
258 add $grp aio_unlink $src, sub {
259 $grp->result ($_[0]);
260 };
261 } else {
262 my $errno = $!;
263 add $grp aio_unlink $dst, sub {
264 $! = $errno;
265 $grp->result (-1);
266 };
267 }
268 };
269 } else {
270 $grp->result (-1);
271 }
272 },
273
274 } else {
275 $grp->result (-1);
276 }
277 };
278 } else {
279 $grp->result ($_[0]);
280 }
281 };
282
283 $grp
284}
285
286=item aio_sendfile $out_fh, $in_fh, $in_offset, $length, $callback->($retval) 340=item aio_sendfile $out_fh, $in_fh, $in_offset, $length, $callback->($retval)
287 341
288Tries to copy C<$length> bytes from C<$in_fh> to C<$out_fh>. It starts 342Tries to copy C<$length> bytes from C<$in_fh> to C<$out_fh>. It starts
289reading at byte offset C<$in_offset>, and starts writing at the current 343reading at byte offset C<$in_offset>, and starts writing at the current
290file offset of C<$out_fh>. Because of that, it is not safe to issue more 344file offset of C<$out_fh>. Because of that, it is not safe to issue more
345=item aio_unlink $pathname, $callback->($status) 399=item aio_unlink $pathname, $callback->($status)
346 400
347Asynchronously unlink (delete) a file and call the callback with the 401Asynchronously unlink (delete) a file and call the callback with the
348result code. 402result code.
349 403
404=item aio_mknod $path, $mode, $dev, $callback->($status)
405
406[EXPERIMENTAL]
407
408Asynchronously create a device node (or fifo). See mknod(2).
409
410The only (POSIX-) portable way of calling this function is:
411
412 aio_mknod $path, IO::AIO::S_IFIFO | $mode, 0, sub { ...
413
350=item aio_link $srcpath, $dstpath, $callback->($status) 414=item aio_link $srcpath, $dstpath, $callback->($status)
351 415
352Asynchronously create a new link to the existing object at C<$srcpath> at 416Asynchronously create a new link to the existing object at C<$srcpath> at
353the path C<$dstpath> and call the callback with the result code. 417the path C<$dstpath> and call the callback with the result code.
354 418
355=item aio_symlink $srcpath, $dstpath, $callback->($status) 419=item aio_symlink $srcpath, $dstpath, $callback->($status)
356 420
357Asynchronously create a new symbolic link to the existing object at C<$srcpath> at 421Asynchronously create a new symbolic link to the existing object at C<$srcpath> at
358the path C<$dstpath> and call the callback with the result code. 422the path C<$dstpath> and call the callback with the result code.
423
424=item aio_readlink $path, $callback->($link)
425
426Asynchronously read the symlink specified by C<$path> and pass it to
427the callback. If an error occurs, nothing or undef gets passed to the
428callback.
359 429
360=item aio_rename $srcpath, $dstpath, $callback->($status) 430=item aio_rename $srcpath, $dstpath, $callback->($status)
361 431
362Asynchronously rename the object at C<$srcpath> to C<$dstpath>, just as 432Asynchronously rename the object at C<$srcpath> to C<$dstpath>, just as
363rename(2) and call the callback with the result code. 433rename(2) and call the callback with the result code.
374sorted, and will B<NOT> include the C<.> and C<..> entries. 444sorted, and will B<NOT> include the C<.> and C<..> entries.
375 445
376The callback a single argument which is either C<undef> or an array-ref 446The callback a single argument which is either C<undef> or an array-ref
377with the filenames. 447with the filenames.
378 448
449=item aio_copy $srcpath, $dstpath, $callback->($status)
450
451Try to copy the I<file> (directories not supported as either source or
452destination) from C<$srcpath> to C<$dstpath> and call the callback with
453the C<0> (error) or C<-1> ok.
454
455This is a composite request that it creates the destination file with
456mode 0200 and copies the contents of the source file into it using
457C<aio_sendfile>, followed by restoring atime, mtime, access mode and
458uid/gid, in that order.
459
460If an error occurs, the partial destination file will be unlinked, if
461possible, except when setting atime, mtime, access mode and uid/gid, where
462errors are being ignored.
463
464=cut
465
466sub aio_copy($$;$) {
467 aio_block {
468 my ($src, $dst, $cb) = @_;
469
470 my $pri = aioreq_pri;
471 my $grp = aio_group $cb;
472
473 aioreq_pri $pri;
474 add $grp aio_open $src, O_RDONLY, 0, sub {
475 if (my $src_fh = $_[0]) {
476 my @stat = stat $src_fh;
477
478 aioreq_pri $pri;
479 add $grp aio_open $dst, O_CREAT | O_WRONLY | O_TRUNC, 0200, sub {
480 if (my $dst_fh = $_[0]) {
481 aioreq_pri $pri;
482 add $grp aio_sendfile $dst_fh, $src_fh, 0, $stat[7], sub {
483 if ($_[0] == $stat[7]) {
484 $grp->result (0);
485 close $src_fh;
486
487 # those should not normally block. should. should.
488 utime $stat[8], $stat[9], $dst;
489 chmod $stat[2] & 07777, $dst_fh;
490 chown $stat[4], $stat[5], $dst_fh;
491 close $dst_fh;
492 } else {
493 $grp->result (-1);
494 close $src_fh;
495 close $dst_fh;
496
497 aioreq $pri;
498 add $grp aio_unlink $dst;
499 }
500 };
501 } else {
502 $grp->result (-1);
503 }
504 },
505
506 } else {
507 $grp->result (-1);
508 }
509 };
510
511 $grp
512 }
513}
514
515=item aio_move $srcpath, $dstpath, $callback->($status)
516
517Try to move the I<file> (directories not supported as either source or
518destination) from C<$srcpath> to C<$dstpath> and call the callback with
519the C<0> (error) or C<-1> ok.
520
521This is a composite request that tries to rename(2) the file first. If
522rename files with C<EXDEV>, it copies the file with C<aio_copy> and, if
523that is successful, unlinking the C<$srcpath>.
524
525=cut
526
527sub aio_move($$;$) {
528 aio_block {
529 my ($src, $dst, $cb) = @_;
530
531 my $pri = aioreq_pri;
532 my $grp = aio_group $cb;
533
534 aioreq_pri $pri;
535 add $grp aio_rename $src, $dst, sub {
536 if ($_[0] && $! == EXDEV) {
537 aioreq_pri $pri;
538 add $grp aio_copy $src, $dst, sub {
539 $grp->result ($_[0]);
540
541 if (!$_[0]) {
542 aioreq_pri $pri;
543 add $grp aio_unlink $src;
544 }
545 };
546 } else {
547 $grp->result ($_[0]);
548 }
549 };
550
551 $grp
552 }
553}
554
379=item aio_scandir $path, $maxreq, $callback->($dirs, $nondirs) 555=item aio_scandir $path, $maxreq, $callback->($dirs, $nondirs)
380 556
381[EXPERIMENTAL due to internal aio_group use]
382
383Scans a directory (similar to C<aio_readdir>) but additionally tries to 557Scans a directory (similar to C<aio_readdir>) but additionally tries to
384separate the entries of directory C<$path> into two sets of names, ones 558efficiently separate the entries of directory C<$path> into two sets of
385you can recurse into (directories or links to them), and ones you cannot 559names, directories you can recurse into (directories), and ones you cannot
386recurse into (everything else). 560recurse into (everything else, including symlinks to directories).
387 561
388C<aio_scandir> is a composite request that creates of many sub requests_ 562C<aio_scandir> is a composite request that creates of many sub requests_
389C<$maxreq> specifies the maximum number of outstanding aio requests that 563C<$maxreq> specifies the maximum number of outstanding aio requests that
390this function generates. If it is C<< <= 0 >>, then a suitable default 564this function generates. If it is C<< <= 0 >>, then a suitable default
391will be chosen (currently 6). 565will be chosen (currently 4).
392 566
393On error, the callback is called without arguments, otherwise it receives 567On error, the callback is called without arguments, otherwise it receives
394two array-refs with path-relative entry names. 568two array-refs with path-relative entry names.
395 569
396Example: 570Example:
431directory counting heuristic. 605directory counting heuristic.
432 606
433=cut 607=cut
434 608
435sub aio_scandir($$$) { 609sub aio_scandir($$$) {
610 aio_block {
436 my ($path, $maxreq, $cb) = @_; 611 my ($path, $maxreq, $cb) = @_;
437 612
613 my $pri = aioreq_pri;
614
438 my $grp = aio_group $cb; 615 my $grp = aio_group $cb;
439 616
440 $maxreq = 6 if $maxreq <= 0; 617 $maxreq = 4 if $maxreq <= 0;
441 618
442 # stat once 619 # stat once
620 aioreq_pri $pri;
443 add $grp aio_stat $path, sub { 621 add $grp aio_stat $path, sub {
444 return $grp->result () if $_[0]; 622 return $grp->result () if $_[0];
445 my $now = time; 623 my $now = time;
446 my $hash1 = join ":", (stat _)[0,1,3,7,9]; 624 my $hash1 = join ":", (stat _)[0,1,3,7,9];
447 625
448 # read the directory entries 626 # read the directory entries
627 aioreq_pri $pri;
449 add $grp aio_readdir $path, sub { 628 add $grp aio_readdir $path, sub {
450 my $entries = shift 629 my $entries = shift
451 or return $grp->result (); 630 or return $grp->result ();
452 631
453 # stat the dir another time 632 # stat the dir another time
633 aioreq_pri $pri;
454 add $grp aio_stat $path, sub { 634 add $grp aio_stat $path, sub {
455 my $hash2 = join ":", (stat _)[0,1,3,7,9]; 635 my $hash2 = join ":", (stat _)[0,1,3,7,9];
456 636
457 my $ndirs; 637 my $ndirs;
458 638
459 # take the slow route if anything looks fishy 639 # take the slow route if anything looks fishy
460 if ($hash1 ne $hash2 or (stat _)[9] == $now) { 640 if ($hash1 ne $hash2 or (stat _)[9] == $now) {
461 $ndirs = -1; 641 $ndirs = -1;
462 } else { 642 } else {
463 # if nlink == 2, we are finished 643 # if nlink == 2, we are finished
464 # on non-posix-fs's, we rely on nlink < 2 644 # on non-posix-fs's, we rely on nlink < 2
465 $ndirs = (stat _)[3] - 2 645 $ndirs = (stat _)[3] - 2
466 or return $grp->result ([], $entries); 646 or return $grp->result ([], $entries);
467 } 647 }
468 648
469 # sort into likely dirs and likely nondirs 649 # sort into likely dirs and likely nondirs
470 # dirs == files without ".", short entries first 650 # dirs == files without ".", short entries first
471 $entries = [map $_->[0], 651 $entries = [map $_->[0],
472 sort { $b->[1] cmp $a->[1] } 652 sort { $b->[1] cmp $a->[1] }
473 map [$_, sprintf "%s%04d", (/.\./ ? "1" : "0"), length], 653 map [$_, sprintf "%s%04d", (/.\./ ? "1" : "0"), length],
474 @$entries]; 654 @$entries];
475 655
476 my (@dirs, @nondirs); 656 my (@dirs, @nondirs);
477 657
478 my ($statcb, $schedcb);
479 my $nreq = 0;
480
481 my $statgrp = add $grp aio_group; 658 my $statgrp = add $grp aio_group sub {
482
483 $schedcb = sub {
484 if (@$entries) {
485 if ($nreq < $maxreq) {
486 my $ent = pop @$entries;
487 $nreq++;
488 add $statgrp aio_stat "$path/$ent/.", sub { $statcb->($_[0], $ent) };
489 }
490 } elsif (!$nreq) {
491 # finished
492 $statgrp->cancel;
493 undef $statcb;
494 undef $schedcb;
495 $grp->result (\@dirs, \@nondirs); 659 $grp->result (\@dirs, \@nondirs);
496 } 660 };
661
662 limit $statgrp $maxreq;
663 feed $statgrp sub {
664 return unless @$entries;
665 my $entry = pop @$entries;
666
667 aioreq_pri $pri;
668 add $statgrp aio_stat "$path/$entry/.", sub {
669 if ($_[0] < 0) {
670 push @nondirs, $entry;
671 } else {
672 # need to check for real directory
673 aioreq_pri $pri;
674 add $statgrp aio_lstat "$path/$entry", sub {
675 if (-d _) {
676 push @dirs, $entry;
677
678 unless (--$ndirs) {
679 push @nondirs, @$entries;
680 feed $statgrp;
681 }
682 } else {
683 push @nondirs, $entry;
684 }
685 }
686 }
687 };
688 };
497 }; 689 };
498 $statcb = sub {
499 my ($status, $entry) = @_;
500
501 if ($status < 0) {
502 $nreq--;
503 push @nondirs, $entry;
504 &$schedcb;
505 } else {
506 # need to check for real directory
507 add $grp aio_lstat "$path/$entry", sub {
508 $nreq--;
509
510 if (-d _) {
511 push @dirs, $entry;
512
513 if (!--$ndirs) {
514 push @nondirs, @$entries;
515 $entries = [];
516 }
517 } else {
518 push @nondirs, $entry;
519 }
520
521 &$schedcb;
522 }
523 }
524 };
525
526 &$schedcb while @$entries && $nreq < $maxreq;
527 }; 690 };
528 }; 691 };
692
693 $grp
529 }; 694 }
530
531 $grp
532} 695}
533 696
534=item aio_fsync $fh, $callback->($status) 697=item aio_fsync $fh, $callback->($status)
535 698
536Asynchronously call fsync on the given filehandle and call the callback 699Asynchronously call fsync on the given filehandle and call the callback
544If this call isn't available because your OS lacks it or it couldn't be 707If this call isn't available because your OS lacks it or it couldn't be
545detected, it will be emulated by calling C<fsync> instead. 708detected, it will be emulated by calling C<fsync> instead.
546 709
547=item aio_group $callback->(...) 710=item aio_group $callback->(...)
548 711
549[EXPERIMENTAL]
550
551This is a very special aio request: Instead of doing something, it is a 712This is a very special aio request: Instead of doing something, it is a
552container for other aio requests, which is useful if you want to bundle 713container for other aio requests, which is useful if you want to bundle
553many requests into a single, composite, request. 714many requests into a single, composite, request with a definite callback
715and the ability to cancel the whole request with its subrequests.
554 716
555Returns an object of class L<IO::AIO::GRP>. See its documentation below 717Returns an object of class L<IO::AIO::GRP>. See its documentation below
556for more info. 718for more info.
557 719
558Example: 720Example:
577phase and still requires a worker thread. Thus, the callback will not 739phase and still requires a worker thread. Thus, the callback will not
578be executed immediately but only after other requests in the queue have 740be executed immediately but only after other requests in the queue have
579entered their execution phase. This can be used to measure request 741entered their execution phase. This can be used to measure request
580latency. 742latency.
581 743
582=item IO::AIO::aio_sleep $fractional_seconds, $callback->() *NOT EXPORTED* 744=item IO::AIO::aio_busy $fractional_seconds, $callback->() *NOT EXPORTED*
583 745
584Mainly used for debugging and benchmarking, this aio request puts one of 746Mainly used for debugging and benchmarking, this aio request puts one of
585the request workers to sleep for the given time. 747the request workers to sleep for the given time.
586 748
587While it is theoretically handy to have simple I/O scheduling requests 749While it is theoretically handy to have simple I/O scheduling requests
588like sleep and file handle readable/writable, the overhead this creates 750like sleep and file handle readable/writable, the overhead this creates is
589is immense, so do not use this function except to put your application 751immense (it blocks a thread for a long time) so do not use this function
590under artificial I/O pressure. 752except to put your application under artificial I/O pressure.
591 753
592=back 754=back
593 755
594=head2 IO::AIO::REQ CLASS 756=head2 IO::AIO::REQ CLASS
595 757
596All non-aggregate C<aio_*> functions return an object of this class when 758All non-aggregate C<aio_*> functions return an object of this class when
597called in non-void context. 759called in non-void context.
598
599A request always moves through the following five states in its lifetime,
600in order: B<ready> (request has been created, but has not been executed
601yet), B<execute> (request is currently being executed), B<pending>
602(request has been executed but callback has not been called yet),
603B<result> (results are being processed synchronously, includes calling the
604callback) and B<done> (request has reached the end of its lifetime and
605holds no resources anymore).
606 760
607=over 4 761=over 4
608 762
609=item cancel $req 763=item cancel $req
610 764
664=item * They can also can also be added to other IO::AIO::GRP objects. 818=item * They can also can also be added to other IO::AIO::GRP objects.
665 819
666=item * You must not add requests to a group from within the group callback (or 820=item * You must not add requests to a group from within the group callback (or
667any later time). 821any later time).
668 822
669=item * This does not harmonise well with C<max_outstanding>, so best do
670not combine C<aio_group> with it. Groups and feeders are recommended for
671this kind of concurrency-limiting.
672
673=back 823=back
674 824
675Their lifetime, simplified, looks like this: when they are empty, they 825Their lifetime, simplified, looks like this: when they are empty, they
676will finish very quickly. If they contain only requests that are in the 826will finish very quickly. If they contain only requests that are in the
677C<done> state, they will also finish. Otherwise they will continue to 827C<done> state, they will also finish. Otherwise they will continue to
692be added, including other groups, as long as you do not create circular 842be added, including other groups, as long as you do not create circular
693dependencies. 843dependencies.
694 844
695Returns all its arguments. 845Returns all its arguments.
696 846
847=item $grp->cancel_subs
848
849Cancel all subrequests and clears any feeder, but not the group request
850itself. Useful when you queued a lot of events but got a result early.
851
697=item $grp->result (...) 852=item $grp->result (...)
698 853
699Set the result value(s) that will be passed to the group callback when all 854Set the result value(s) that will be passed to the group callback when all
700subrequests have finished. By default, no argument will be passed. 855subrequests have finished and set thre groups errno to the current value
856of errno (just like calling C<errno> without an error number). By default,
857no argument will be passed and errno is zero.
858
859=item $grp->errno ([$errno])
860
861Sets the group errno value to C<$errno>, or the current value of errno
862when the argument is missing.
863
864Every aio request has an associated errno value that is restored when
865the callback is invoked. This method lets you change this value from its
866default (0).
867
868Calling C<result> will also set errno, so make sure you either set C<$!>
869before the call to C<result>, or call c<errno> after it.
701 870
702=item feed $grp $callback->($grp) 871=item feed $grp $callback->($grp)
703
704[VERY EXPERIMENTAL]
705 872
706Sets a feeder/generator on this group: every group can have an attached 873Sets a feeder/generator on this group: every group can have an attached
707generator that generates requests if idle. The idea behind this is that, 874generator that generates requests if idle. The idea behind this is that,
708although you could just queue as many requests as you want in a group, 875although you could just queue as many requests as you want in a group,
709this might starve other requests for a potentially long time. For 876this might starve other requests for a potentially long time. For
746 913
747=back 914=back
748 915
749=head2 SUPPORT FUNCTIONS 916=head2 SUPPORT FUNCTIONS
750 917
918=head3 EVENT PROCESSING AND EVENT LOOP INTEGRATION
919
751=over 4 920=over 4
752 921
753=item $fileno = IO::AIO::poll_fileno 922=item $fileno = IO::AIO::poll_fileno
754 923
755Return the I<request result pipe file descriptor>. This filehandle must be 924Return the I<request result pipe file descriptor>. This filehandle must be
759 928
760See C<poll_cb> for an example. 929See C<poll_cb> for an example.
761 930
762=item IO::AIO::poll_cb 931=item IO::AIO::poll_cb
763 932
764Process all outstanding events on the result pipe. You have to call this 933Process some outstanding events on the result pipe. You have to call this
765regularly. Returns the number of events processed. Returns immediately 934regularly. Returns the number of events processed. Returns immediately
766when no events are outstanding. 935when no events are outstanding. The amount of events processed depends on
936the settings of C<IO::AIO::max_poll_req> and C<IO::AIO::max_poll_time>.
937
938If not all requests were processed for whatever reason, the filehandle
939will still be ready when C<poll_cb> returns.
767 940
768Example: Install an Event watcher that automatically calls 941Example: Install an Event watcher that automatically calls
769IO::AIO::poll_cb with high priority: 942IO::AIO::poll_cb with high priority:
770 943
771 Event->io (fd => IO::AIO::poll_fileno, 944 Event->io (fd => IO::AIO::poll_fileno,
772 poll => 'r', async => 1, 945 poll => 'r', async => 1,
773 cb => \&IO::AIO::poll_cb); 946 cb => \&IO::AIO::poll_cb);
774 947
948=item IO::AIO::max_poll_reqs $nreqs
949
950=item IO::AIO::max_poll_time $seconds
951
952These set the maximum number of requests (default C<0>, meaning infinity)
953that are being processed by C<IO::AIO::poll_cb> in one call, respectively
954the maximum amount of time (default C<0>, meaning infinity) spent in
955C<IO::AIO::poll_cb> to process requests (more correctly the mininum amount
956of time C<poll_cb> is allowed to use).
957
958Setting C<max_poll_time> to a non-zero value creates an overhead of one
959syscall per request processed, which is not normally a problem unless your
960callbacks are really really fast or your OS is really really slow (I am
961not mentioning Solaris here). Using C<max_poll_reqs> incurs no overhead.
962
963Setting these is useful if you want to ensure some level of
964interactiveness when perl is not fast enough to process all requests in
965time.
966
967For interactive programs, values such as C<0.01> to C<0.1> should be fine.
968
969Example: Install an Event watcher that automatically calls
970IO::AIO::poll_cb with low priority, to ensure that other parts of the
971program get the CPU sometimes even under high AIO load.
972
973 # try not to spend much more than 0.1s in poll_cb
974 IO::AIO::max_poll_time 0.1;
975
976 # use a low priority so other tasks have priority
977 Event->io (fd => IO::AIO::poll_fileno,
978 poll => 'r', nice => 1,
979 cb => &IO::AIO::poll_cb);
980
775=item IO::AIO::poll_wait 981=item IO::AIO::poll_wait
776 982
983If there are any outstanding requests and none of them in the result
777Wait till the result filehandle becomes ready for reading (simply does a 984phase, wait till the result filehandle becomes ready for reading (simply
778C<select> on the filehandle. This is useful if you want to synchronously wait 985does a C<select> on the filehandle. This is useful if you want to
779for some requests to finish). 986synchronously wait for some requests to finish).
780 987
781See C<nreqs> for an example. 988See C<nreqs> for an example.
782 989
990=item IO::AIO::poll
991
992Waits until some requests have been handled.
993
994Returns the number of requests processed, but is otherwise strictly
995equivalent to:
996
997 IO::AIO::poll_wait, IO::AIO::poll_cb
998
783=item IO::AIO::nreqs 999=item IO::AIO::flush
784 1000
785Returns the number of requests currently outstanding (i.e. for which their 1001Wait till all outstanding AIO requests have been handled.
786callback has not been invoked yet).
787 1002
788Example: wait till there are no outstanding requests anymore: 1003Strictly equivalent to:
789 1004
790 IO::AIO::poll_wait, IO::AIO::poll_cb 1005 IO::AIO::poll_wait, IO::AIO::poll_cb
791 while IO::AIO::nreqs; 1006 while IO::AIO::nreqs;
792 1007
793=item IO::AIO::flush 1008=head3 CONTROLLING THE NUMBER OF THREADS
794
795Wait till all outstanding AIO requests have been handled.
796
797Strictly equivalent to:
798
799 IO::AIO::poll_wait, IO::AIO::poll_cb
800 while IO::AIO::nreqs;
801
802=item IO::AIO::poll
803
804Waits until some requests have been handled.
805
806Strictly equivalent to:
807
808 IO::AIO::poll_wait, IO::AIO::poll_cb
809 if IO::AIO::nreqs;
810 1009
811=item IO::AIO::min_parallel $nthreads 1010=item IO::AIO::min_parallel $nthreads
812 1011
813Set the minimum number of AIO threads to C<$nthreads>. The current 1012Set the minimum number of AIO threads to C<$nthreads>. The current
814default is C<8>, which means eight asynchronous operations can execute 1013default is C<8>, which means eight asynchronous operations can execute
815concurrently at any one time (the number of outstanding requests, 1014concurrently at any one time (the number of outstanding requests,
816however, is unlimited). 1015however, is unlimited).
817 1016
818IO::AIO starts threads only on demand, when an AIO request is queued and 1017IO::AIO starts threads only on demand, when an AIO request is queued and
819no free thread exists. 1018no free thread exists. Please note that queueing up a hundred requests can
1019create demand for a hundred threads, even if it turns out that everything
1020is in the cache and could have been processed faster by a single thread.
820 1021
821It is recommended to keep the number of threads relatively low, as some 1022It is recommended to keep the number of threads relatively low, as some
822Linux kernel versions will scale negatively with the number of threads 1023Linux kernel versions will scale negatively with the number of threads
823(higher parallelity => MUCH higher latency). With current Linux 2.6 1024(higher parallelity => MUCH higher latency). With current Linux 2.6
824versions, 4-32 threads should be fine. 1025versions, 4-32 threads should be fine.
838This module automatically runs C<max_parallel 0> at program end, to ensure 1039This module automatically runs C<max_parallel 0> at program end, to ensure
839that all threads are killed and that there are no outstanding requests. 1040that all threads are killed and that there are no outstanding requests.
840 1041
841Under normal circumstances you don't need to call this function. 1042Under normal circumstances you don't need to call this function.
842 1043
1044=item IO::AIO::max_idle $nthreads
1045
1046Limit the number of threads (default: 4) that are allowed to idle (i.e.,
1047threads that did not get a request to process within 10 seconds). That
1048means if a thread becomes idle while C<$nthreads> other threads are also
1049idle, it will free its resources and exit.
1050
1051This is useful when you allow a large number of threads (e.g. 100 or 1000)
1052to allow for extremely high load situations, but want to free resources
1053under normal circumstances (1000 threads can easily consume 30MB of RAM).
1054
1055The default is probably ok in most situations, especially if thread
1056creation is fast. If thread creation is very slow on your system you might
1057want to use larger values.
1058
843=item $oldnreqs = IO::AIO::max_outstanding $nreqs 1059=item $oldmaxreqs = IO::AIO::max_outstanding $maxreqs
844 1060
845[DEPRECATED] 1061This is a very bad function to use in interactive programs because it
1062blocks, and a bad way to reduce concurrency because it is inexact: Better
1063use an C<aio_group> together with a feed callback.
846 1064
847Sets the maximum number of outstanding requests to C<$nreqs>. If you 1065Sets the maximum number of outstanding requests to C<$nreqs>. If you
848try to queue up more than this number of requests, the caller will block until 1066to queue up more than this number of requests, the next call to the
849some requests have been handled. 1067C<poll_cb> (and C<poll_some> and other functions calling C<poll_cb>)
1068function will block until the limit is no longer exceeded.
850 1069
851The default is very large, so normally there is no practical limit. If you 1070The default value is very large, so there is no practical limit on the
852queue up many requests in a loop it often improves speed if you set 1071number of outstanding requests.
853this to a relatively low number, such as C<100>.
854 1072
855This function does not work well together with C<aio_group>'s, and their 1073You can still queue as many requests as you want. Therefore,
856feeder interface is better suited to limiting concurrency, so do not use 1074C<max_oustsanding> is mainly useful in simple scripts (with low values) or
857this function. 1075as a stop gap to shield against fatal memory overflow (with large values).
858 1076
859Under normal circumstances you don't need to call this function. 1077=head3 STATISTICAL INFORMATION
1078
1079=item IO::AIO::nreqs
1080
1081Returns the number of requests currently in the ready, execute or pending
1082states (i.e. for which their callback has not been invoked yet).
1083
1084Example: wait till there are no outstanding requests anymore:
1085
1086 IO::AIO::poll_wait, IO::AIO::poll_cb
1087 while IO::AIO::nreqs;
1088
1089=item IO::AIO::nready
1090
1091Returns the number of requests currently in the ready state (not yet
1092executed).
1093
1094=item IO::AIO::npending
1095
1096Returns the number of requests currently in the pending state (executed,
1097but not yet processed by poll_cb).
860 1098
861=back 1099=back
862 1100
863=cut 1101=cut
864 1102
878 *$sym 1116 *$sym
879} 1117}
880 1118
881min_parallel 8; 1119min_parallel 8;
882 1120
883END { 1121END { flush }
884 max_parallel 0;
885}
886 1122
8871; 11231;
888 1124
889=head2 FORK BEHAVIOUR 1125=head2 FORK BEHAVIOUR
890 1126
891This module should do "the right thing" when the process using it forks: 1127This module should do "the right thing" when the process using it forks:
892 1128
893Before the fork, IO::AIO enters a quiescent state where no requests 1129Before the fork, IO::AIO enters a quiescent state where no requests
894can be added in other threads and no results will be processed. After 1130can be added in other threads and no results will be processed. After
895the fork the parent simply leaves the quiescent state and continues 1131the fork the parent simply leaves the quiescent state and continues
896request/result processing, while the child clears the request/result 1132request/result processing, while the child frees the request/result queue
897queue (so the requests started before the fork will only be handled in 1133(so that the requests started before the fork will only be handled in the
898the parent). Threads will be started on demand until the limit ste in the 1134parent). Threads will be started on demand until the limit set in the
899parent process has been reached again. 1135parent process has been reached again.
900 1136
901In short: the parent will, after a short pause, continue as if fork had 1137In short: the parent will, after a short pause, continue as if fork had
902not been called, while the child will act as if IO::AIO has not been used 1138not been called, while the child will act as if IO::AIO has not been used
903yet. 1139yet.
904 1140
905=head2 MEMORY USAGE 1141=head2 MEMORY USAGE
906 1142
1143Per-request usage:
1144
907Each aio request uses - depending on your architecture - around 128 bytes 1145Each aio request uses - depending on your architecture - around 100-200
908of memory. In addition, stat requests need a stat buffer (possibly a few 1146bytes of memory. In addition, stat requests need a stat buffer (possibly
909hundred bytes). Perl scalars and other data passed into aio requests will 1147a few hundred bytes), readdir requires a result buffer and so on. Perl
910also be locked. 1148scalars and other data passed into aio requests will also be locked and
1149will consume memory till the request has entered the done state.
911 1150
912This is now awfully much, so queuing lots of requests is not usually a 1151This is now awfully much, so queuing lots of requests is not usually a
913problem. 1152problem.
914 1153
915Each thread needs a stack area which is usually around 16k, sometimes much 1154Per-thread usage:
916larger, depending on the OS. 1155
1156In the execution phase, some aio requests require more memory for
1157temporary buffers, and each thread requires a stack and other data
1158structures (usually around 16k-128k, depending on the OS).
1159
1160=head1 KNOWN BUGS
1161
1162Known bugs will be fixed in the next release.
917 1163
918=head1 SEE ALSO 1164=head1 SEE ALSO
919 1165
920L<Coro::AIO>. 1166L<Coro::AIO>.
921 1167

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines