ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/IO-AIO/AIO.pm
Revision: 1.149
Committed: Sat Jun 6 18:19:35 2009 UTC (14 years, 11 months ago) by root
Branch: MAIN
Changes since 1.148: +22 -20 lines
Log Message:
*** empty log message ***

File Contents

# User Rev Content
1 root 1.1 =head1 NAME
2    
3     IO::AIO - Asynchronous Input/Output
4    
5     =head1 SYNOPSIS
6    
7     use IO::AIO;
8    
9 root 1.6 aio_open "/etc/passwd", O_RDONLY, 0, sub {
10 root 1.94 my $fh = shift
11     or die "/etc/passwd: $!";
12 root 1.6 ...
13     };
14    
15     aio_unlink "/tmp/file", sub { };
16    
17     aio_read $fh, 30000, 1024, $buffer, 0, sub {
18 root 1.8 $_[0] > 0 or die "read error: $!";
19 root 1.6 };
20    
21 root 1.56 # version 2+ has request and group objects
22     use IO::AIO 2;
23 root 1.52
24 root 1.68 aioreq_pri 4; # give next request a very high priority
25 root 1.52 my $req = aio_unlink "/tmp/file", sub { };
26     $req->cancel; # cancel request if still in queue
27    
28 root 1.56 my $grp = aio_group sub { print "all stats done\n" };
29     add $grp aio_stat "..." for ...;
30    
31 root 1.125 # AnyEvent integration (EV, Event, Glib, Tk, POE, urxvt, pureperl...)
32     use AnyEvent::AIO;
33 root 1.42
34 root 1.118 # EV integration
35     my $w = EV::io IO::AIO::poll_fileno, EV::READ, \&IO::AIO::poll_cb;
36    
37 root 1.56 # Event integration
38 root 1.6 Event->io (fd => IO::AIO::poll_fileno,
39 root 1.7 poll => 'r',
40 root 1.6 cb => \&IO::AIO::poll_cb);
41    
42 root 1.56 # Glib/Gtk2 integration
43 root 1.6 add_watch Glib::IO IO::AIO::poll_fileno,
44 root 1.22 in => sub { IO::AIO::poll_cb; 1 };
45 root 1.6
46 root 1.56 # Tk integration
47 root 1.6 Tk::Event::IO->fileevent (IO::AIO::poll_fileno, "",
48     readable => \&IO::AIO::poll_cb);
49    
50 root 1.56 # Danga::Socket integration
51 root 1.11 Danga::Socket->AddOtherFds (IO::AIO::poll_fileno =>
52     \&IO::AIO::poll_cb);
53    
54 root 1.1 =head1 DESCRIPTION
55    
56     This module implements asynchronous I/O using whatever means your
57 root 1.2 operating system supports.
58 root 1.1
59 root 1.85 Asynchronous means that operations that can normally block your program
60     (e.g. reading from disk) will be done asynchronously: the operation
61     will still block, but you can do something else in the meantime. This
62     is extremely useful for programs that need to stay interactive even
63     when doing heavy I/O (GUI programs, high performance network servers
64     etc.), but can also be used to easily do operations in parallel that are
65     normally done sequentially, e.g. stat'ing many files, which is much faster
66     on a RAID volume or over NFS when you do a number of stat operations
67     concurrently.
68    
69 root 1.108 While most of this works on all types of file descriptors (for
70     example sockets), using these functions on file descriptors that
71     support nonblocking operation (again, sockets, pipes etc.) is very
72     inefficient. Use an event loop for that (such as the L<Event|Event>
73     module): IO::AIO will naturally fit into such an event loop itself.
74 root 1.85
75 root 1.72 In this version, a number of threads are started that execute your
76     requests and signal their completion. You don't need thread support
77     in perl, and the threads created by this module will not be visible
78     to perl. In the future, this module might make use of the native aio
79     functions available on many operating systems. However, they are often
80 root 1.85 not well-supported or restricted (GNU/Linux doesn't allow them on normal
81 root 1.72 files currently, for example), and they would only support aio_read and
82     aio_write, so the remaining functionality would have to be implemented
83     using threads anyway.
84    
85 root 1.108 Although the module will work in the presence of other (Perl-) threads,
86     it is currently not reentrant in any way, so use appropriate locking
87     yourself, always call C<poll_cb> from within the same thread, or never
88     call C<poll_cb> (or other C<aio_> functions) recursively.
89 root 1.72
90 root 1.86 =head2 EXAMPLE
91    
92     This is a simple example that uses the Event module and loads
93     F</etc/passwd> asynchronously:
94    
95     use Fcntl;
96     use Event;
97     use IO::AIO;
98    
99     # register the IO::AIO callback with Event
100     Event->io (fd => IO::AIO::poll_fileno,
101     poll => 'r',
102     cb => \&IO::AIO::poll_cb);
103    
104     # queue the request to open /etc/passwd
105     aio_open "/etc/passwd", O_RDONLY, 0, sub {
106 root 1.94 my $fh = shift
107 root 1.86 or die "error while opening: $!";
108    
109     # stat'ing filehandles is generally non-blocking
110     my $size = -s $fh;
111    
112     # queue a request to read the file
113     my $contents;
114     aio_read $fh, 0, $size, $contents, 0, sub {
115     $_[0] == $size
116     or die "short read: $!";
117    
118     close $fh;
119    
120     # file contents now in $contents
121     print $contents;
122    
123     # exit event loop and program
124     Event::unloop;
125     };
126     };
127    
128     # possibly queue up other requests, or open GUI windows,
129     # check for sockets etc. etc.
130    
131     # process events as long as there are some:
132     Event::loop;
133    
134 root 1.72 =head1 REQUEST ANATOMY AND LIFETIME
135    
136     Every C<aio_*> function creates a request. which is a C data structure not
137     directly visible to Perl.
138    
139     If called in non-void context, every request function returns a Perl
140     object representing the request. In void context, nothing is returned,
141     which saves a bit of memory.
142    
143     The perl object is a fairly standard ref-to-hash object. The hash contents
144     are not used by IO::AIO so you are free to store anything you like in it.
145    
146     During their existance, aio requests travel through the following states,
147     in order:
148    
149     =over 4
150    
151     =item ready
152    
153     Immediately after a request is created it is put into the ready state,
154     waiting for a thread to execute it.
155    
156     =item execute
157    
158     A thread has accepted the request for processing and is currently
159     executing it (e.g. blocking in read).
160    
161     =item pending
162    
163     The request has been executed and is waiting for result processing.
164    
165     While request submission and execution is fully asynchronous, result
166     processing is not and relies on the perl interpreter calling C<poll_cb>
167     (or another function with the same effect).
168    
169     =item result
170    
171     The request results are processed synchronously by C<poll_cb>.
172    
173     The C<poll_cb> function will process all outstanding aio requests by
174     calling their callbacks, freeing memory associated with them and managing
175     any groups they are contained in.
176    
177     =item done
178    
179     Request has reached the end of its lifetime and holds no resources anymore
180     (except possibly for the Perl object, but its connection to the actual
181     aio request is severed and calling its methods will either do nothing or
182     result in a runtime error).
183 root 1.1
184 root 1.88 =back
185    
186 root 1.1 =cut
187    
188     package IO::AIO;
189    
190 root 1.117 use Carp ();
191    
192 root 1.23 no warnings;
193 root 1.51 use strict 'vars';
194 root 1.23
195 root 1.1 use base 'Exporter';
196    
197     BEGIN {
198 root 1.149 our $VERSION = '3.2';
199 root 1.1
200 root 1.120 our @AIO_REQ = qw(aio_sendfile aio_read aio_write aio_open aio_close
201 root 1.148 aio_stat aio_lstat aio_unlink aio_rmdir aio_readdir aio_readdirx
202 root 1.120 aio_scandir aio_symlink aio_readlink aio_sync aio_fsync
203 root 1.142 aio_fdatasync aio_sync_file_range aio_pathsync aio_readahead
204 root 1.120 aio_rename aio_link aio_move aio_copy aio_group
205     aio_nop aio_mknod aio_load aio_rmtree aio_mkdir aio_chown
206     aio_chmod aio_utime aio_truncate);
207    
208 root 1.123 our @EXPORT = (@AIO_REQ, qw(aioreq_pri aioreq_nice));
209 root 1.67 our @EXPORT_OK = qw(poll_fileno poll_cb poll_wait flush
210 root 1.86 min_parallel max_parallel max_idle
211     nreqs nready npending nthreads
212     max_poll_time max_poll_reqs);
213 root 1.1
214 root 1.143 push @AIO_REQ, qw(aio_busy); # not exported
215    
216 root 1.54 @IO::AIO::GRP::ISA = 'IO::AIO::REQ';
217    
218 root 1.1 require XSLoader;
219 root 1.51 XSLoader::load ("IO::AIO", $VERSION);
220 root 1.1 }
221    
222 root 1.5 =head1 FUNCTIONS
223 root 1.1
224 root 1.87 =head2 AIO REQUEST FUNCTIONS
225 root 1.1
226 root 1.5 All the C<aio_*> calls are more or less thin wrappers around the syscall
227     with the same name (sans C<aio_>). The arguments are similar or identical,
228 root 1.14 and they all accept an additional (and optional) C<$callback> argument
229     which must be a code reference. This code reference will get called with
230     the syscall return code (e.g. most syscalls return C<-1> on error, unlike
231 root 1.136 perl, which usually delivers "false") as its sole argument after the given
232 root 1.14 syscall has been executed asynchronously.
233 root 1.1
234 root 1.23 All functions expecting a filehandle keep a copy of the filehandle
235     internally until the request has finished.
236 root 1.1
237 root 1.87 All functions return request objects of type L<IO::AIO::REQ> that allow
238     further manipulation of those requests while they are in-flight.
239 root 1.52
240 root 1.28 The pathnames you pass to these routines I<must> be absolute and
241 root 1.87 encoded as octets. The reason for the former is that at the time the
242 root 1.28 request is being executed, the current working directory could have
243     changed. Alternatively, you can make sure that you never change the
244 root 1.87 current working directory anywhere in the program and then use relative
245     paths.
246 root 1.28
247 root 1.87 To encode pathnames as octets, either make sure you either: a) always pass
248     in filenames you got from outside (command line, readdir etc.) without
249     tinkering, b) are ASCII or ISO 8859-1, c) use the Encode module and encode
250 root 1.28 your pathnames to the locale (or other) encoding in effect in the user
251     environment, d) use Glib::filename_from_unicode on unicode filenames or e)
252 root 1.87 use something else to ensure your scalar has the correct contents.
253    
254     This works, btw. independent of the internal UTF-8 bit, which IO::AIO
255 root 1.136 handles correctly whether it is set or not.
256 root 1.1
257 root 1.5 =over 4
258 root 1.1
259 root 1.80 =item $prev_pri = aioreq_pri [$pri]
260 root 1.68
261 root 1.80 Returns the priority value that would be used for the next request and, if
262     C<$pri> is given, sets the priority for the next aio request.
263 root 1.68
264 root 1.80 The default priority is C<0>, the minimum and maximum priorities are C<-4>
265     and C<4>, respectively. Requests with higher priority will be serviced
266     first.
267    
268     The priority will be reset to C<0> after each call to one of the C<aio_*>
269 root 1.68 functions.
270    
271 root 1.69 Example: open a file with low priority, then read something from it with
272     higher priority so the read request is serviced before other low priority
273     open requests (potentially spamming the cache):
274    
275     aioreq_pri -3;
276     aio_open ..., sub {
277     return unless $_[0];
278    
279     aioreq_pri -2;
280     aio_read $_[0], ..., sub {
281     ...
282     };
283     };
284    
285 root 1.106
286 root 1.69 =item aioreq_nice $pri_adjust
287    
288     Similar to C<aioreq_pri>, but subtracts the given value from the current
289 root 1.87 priority, so the effect is cumulative.
290 root 1.69
291 root 1.106
292 root 1.40 =item aio_open $pathname, $flags, $mode, $callback->($fh)
293 root 1.1
294 root 1.2 Asynchronously open or create a file and call the callback with a newly
295     created filehandle for the file.
296 root 1.1
297     The pathname passed to C<aio_open> must be absolute. See API NOTES, above,
298     for an explanation.
299    
300 root 1.20 The C<$flags> argument is a bitmask. See the C<Fcntl> module for a
301     list. They are the same as used by C<sysopen>.
302    
303     Likewise, C<$mode> specifies the mode of the newly created file, if it
304     didn't exist and C<O_CREAT> has been given, just like perl's C<sysopen>,
305     except that it is mandatory (i.e. use C<0> if you don't create new files,
306 root 1.101 and C<0666> or C<0777> if you do). Note that the C<$mode> will be modified
307     by the umask in effect then the request is being executed, so better never
308     change the umask.
309 root 1.1
310     Example:
311    
312     aio_open "/etc/passwd", O_RDONLY, 0, sub {
313 root 1.2 if ($_[0]) {
314     print "open successful, fh is $_[0]\n";
315 root 1.1 ...
316     } else {
317     die "open failed: $!\n";
318     }
319     };
320    
321 root 1.106
322 root 1.40 =item aio_close $fh, $callback->($status)
323 root 1.1
324 root 1.2 Asynchronously close a file and call the callback with the result
325 root 1.116 code.
326    
327 root 1.117 Unfortunately, you can't do this to perl. Perl I<insists> very strongly on
328 root 1.121 closing the file descriptor associated with the filehandle itself.
329 root 1.117
330 root 1.121 Therefore, C<aio_close> will not close the filehandle - instead it will
331     use dup2 to overwrite the file descriptor with the write-end of a pipe
332     (the pipe fd will be created on demand and will be cached).
333 root 1.117
334 root 1.121 Or in other words: the file descriptor will be closed, but it will not be
335     free for reuse until the perl filehandle is closed.
336 root 1.117
337     =cut
338    
339 root 1.40 =item aio_read $fh,$offset,$length, $data,$dataoffset, $callback->($retval)
340 root 1.1
341 root 1.40 =item aio_write $fh,$offset,$length, $data,$dataoffset, $callback->($retval)
342 root 1.1
343 root 1.145 Reads or writes C<$length> bytes from or to the specified C<$fh> and
344     C<$offset> into the scalar given by C<$data> and offset C<$dataoffset>
345     and calls the callback without the actual number of bytes read (or -1 on
346     error, just like the syscall).
347 root 1.109
348 root 1.146 C<aio_read> will, like C<sysread>, shrink or grow the C<$data> scalar to
349     offset plus the actual number of bytes read.
350    
351 root 1.112 If C<$offset> is undefined, then the current file descriptor offset will
352     be used (and updated), otherwise the file descriptor offset will not be
353     changed by these calls.
354 root 1.109
355 root 1.145 If C<$length> is undefined in C<aio_write>, use the remaining length of
356     C<$data>.
357 root 1.109
358     If C<$dataoffset> is less than zero, it will be counted from the end of
359     C<$data>.
360 root 1.1
361 root 1.31 The C<$data> scalar I<MUST NOT> be modified in any way while the request
362 root 1.108 is outstanding. Modifying it can result in segfaults or World War III (if
363     the necessary/optional hardware is installed).
364 root 1.31
365 root 1.17 Example: Read 15 bytes at offset 7 into scalar C<$buffer>, starting at
366 root 1.1 offset C<0> within the scalar:
367    
368     aio_read $fh, 7, 15, $buffer, 0, sub {
369 root 1.9 $_[0] > 0 or die "read error: $!";
370     print "read $_[0] bytes: <$buffer>\n";
371 root 1.1 };
372    
373 root 1.106
374 root 1.40 =item aio_sendfile $out_fh, $in_fh, $in_offset, $length, $callback->($retval)
375 root 1.35
376     Tries to copy C<$length> bytes from C<$in_fh> to C<$out_fh>. It starts
377     reading at byte offset C<$in_offset>, and starts writing at the current
378     file offset of C<$out_fh>. Because of that, it is not safe to issue more
379     than one C<aio_sendfile> per C<$out_fh>, as they will interfere with each
380     other.
381    
382     This call tries to make use of a native C<sendfile> syscall to provide
383     zero-copy operation. For this to work, C<$out_fh> should refer to a
384     socket, and C<$in_fh> should refer to mmap'able file.
385    
386     If the native sendfile call fails or is not implemented, it will be
387 root 1.36 emulated, so you can call C<aio_sendfile> on any type of filehandle
388     regardless of the limitations of the operating system.
389 root 1.35
390     Please note, however, that C<aio_sendfile> can read more bytes from
391     C<$in_fh> than are written, and there is no way to find out how many
392 root 1.36 bytes have been read from C<aio_sendfile> alone, as C<aio_sendfile> only
393     provides the number of bytes written to C<$out_fh>. Only if the result
394     value equals C<$length> one can assume that C<$length> bytes have been
395     read.
396 root 1.35
397 root 1.106
398 root 1.40 =item aio_readahead $fh,$offset,$length, $callback->($retval)
399 root 1.1
400 root 1.20 C<aio_readahead> populates the page cache with data from a file so that
401 root 1.1 subsequent reads from that file will not block on disk I/O. The C<$offset>
402     argument specifies the starting point from which data is to be read and
403     C<$length> specifies the number of bytes to be read. I/O is performed in
404     whole pages, so that offset is effectively rounded down to a page boundary
405     and bytes are read up to the next page boundary greater than or equal to
406 root 1.20 (off-set+length). C<aio_readahead> does not read beyond the end of the
407 root 1.1 file. The current file offset of the file is left unchanged.
408    
409 root 1.26 If that syscall doesn't exist (likely if your OS isn't Linux) it will be
410     emulated by simply reading the data, which would have a similar effect.
411    
412 root 1.106
413 root 1.40 =item aio_stat $fh_or_path, $callback->($status)
414 root 1.1
415 root 1.40 =item aio_lstat $fh, $callback->($status)
416 root 1.1
417     Works like perl's C<stat> or C<lstat> in void context. The callback will
418     be called after the stat and the results will be available using C<stat _>
419     or C<-s _> etc...
420    
421     The pathname passed to C<aio_stat> must be absolute. See API NOTES, above,
422     for an explanation.
423    
424     Currently, the stats are always 64-bit-stats, i.e. instead of returning an
425     error when stat'ing a large file, the results will be silently truncated
426     unless perl itself is compiled with large file support.
427    
428     Example: Print the length of F</etc/passwd>:
429    
430     aio_stat "/etc/passwd", sub {
431     $_[0] and die "stat failed: $!";
432     print "size is ", -s _, "\n";
433     };
434    
435 root 1.106
436     =item aio_utime $fh_or_path, $atime, $mtime, $callback->($status)
437    
438     Works like perl's C<utime> function (including the special case of $atime
439     and $mtime being undef). Fractional times are supported if the underlying
440     syscalls support them.
441    
442     When called with a pathname, uses utimes(2) if available, otherwise
443     utime(2). If called on a file descriptor, uses futimes(2) if available,
444     otherwise returns ENOSYS, so this is not portable.
445    
446     Examples:
447    
448 root 1.107 # set atime and mtime to current time (basically touch(1)):
449 root 1.106 aio_utime "path", undef, undef;
450     # set atime to current time and mtime to beginning of the epoch:
451     aio_utime "path", time, undef; # undef==0
452    
453    
454     =item aio_chown $fh_or_path, $uid, $gid, $callback->($status)
455    
456     Works like perl's C<chown> function, except that C<undef> for either $uid
457     or $gid is being interpreted as "do not change" (but -1 can also be used).
458    
459     Examples:
460    
461     # same as "chown root path" in the shell:
462     aio_chown "path", 0, -1;
463     # same as above:
464     aio_chown "path", 0, undef;
465    
466    
467 root 1.110 =item aio_truncate $fh_or_path, $offset, $callback->($status)
468    
469     Works like truncate(2) or ftruncate(2).
470    
471    
472 root 1.106 =item aio_chmod $fh_or_path, $mode, $callback->($status)
473    
474     Works like perl's C<chmod> function.
475    
476    
477 root 1.40 =item aio_unlink $pathname, $callback->($status)
478 root 1.1
479     Asynchronously unlink (delete) a file and call the callback with the
480     result code.
481    
482 root 1.106
483 root 1.82 =item aio_mknod $path, $mode, $dev, $callback->($status)
484    
485 root 1.86 [EXPERIMENTAL]
486    
487 root 1.83 Asynchronously create a device node (or fifo). See mknod(2).
488    
489 root 1.86 The only (POSIX-) portable way of calling this function is:
490 root 1.83
491     aio_mknod $path, IO::AIO::S_IFIFO | $mode, 0, sub { ...
492 root 1.82
493 root 1.106
494 root 1.50 =item aio_link $srcpath, $dstpath, $callback->($status)
495    
496     Asynchronously create a new link to the existing object at C<$srcpath> at
497     the path C<$dstpath> and call the callback with the result code.
498    
499 root 1.106
500 root 1.50 =item aio_symlink $srcpath, $dstpath, $callback->($status)
501    
502     Asynchronously create a new symbolic link to the existing object at C<$srcpath> at
503     the path C<$dstpath> and call the callback with the result code.
504    
505 root 1.106
506 root 1.90 =item aio_readlink $path, $callback->($link)
507    
508     Asynchronously read the symlink specified by C<$path> and pass it to
509     the callback. If an error occurs, nothing or undef gets passed to the
510     callback.
511    
512 root 1.106
513 root 1.50 =item aio_rename $srcpath, $dstpath, $callback->($status)
514    
515     Asynchronously rename the object at C<$srcpath> to C<$dstpath>, just as
516     rename(2) and call the callback with the result code.
517    
518 root 1.106
519 root 1.101 =item aio_mkdir $pathname, $mode, $callback->($status)
520    
521     Asynchronously mkdir (create) a directory and call the callback with
522     the result code. C<$mode> will be modified by the umask at the time the
523     request is executed, so do not change your umask.
524    
525 root 1.106
526 root 1.40 =item aio_rmdir $pathname, $callback->($status)
527 root 1.27
528     Asynchronously rmdir (delete) a directory and call the callback with the
529     result code.
530    
531 root 1.106
532 root 1.46 =item aio_readdir $pathname, $callback->($entries)
533 root 1.37
534     Unlike the POSIX call of the same name, C<aio_readdir> reads an entire
535     directory (i.e. opendir + readdir + closedir). The entries will not be
536     sorted, and will B<NOT> include the C<.> and C<..> entries.
537    
538 root 1.148 The callback is passed a single argument which is either C<undef> or an
539     array-ref with the filenames.
540    
541    
542     =item aio_readdirx $pathname, $flags, $callback->($entries, $flags)
543    
544     Quite similar to C<aio_readdir>, but the C<$flags> argument allows to tune
545     behaviour and output format. In case of an error, C<$entries> will be
546     C<undef>.
547    
548     The flags are a combination of the following constants, ORed together (the
549     flags will also be passed to the callback, possibly modified):
550    
551     =over 4
552    
553     =item AIO::READDIR_DENTS
554    
555     When this flag is off, then the callback gets an arrayref with of names
556     only (as with C<aio_readdir>), otherwise it gets an arrayref with
557     C<[$name, $inode, $type]> arrayrefs, each describing a single directory
558     entry in more detail.
559    
560     C<$name> is the name of the entry.
561    
562     C<$inode> is the inode number (which might not be exact on systems with 64
563     bit inode numbers and 32 bit perls). On systems that do not deliver the
564     inode information, this will always be zero.
565    
566     C<$type> is one of the C<AIO::DT_xxx> constants:
567    
568     C<AIO::DT_UNKNOWN>, C<AIO::DT_FIFO>, C<AIO::DT_CHR>, C<AIO::DT_DIR>,
569     C<AIO::DT_BLK>, C<AIO::DT_REG>, C<AIO::DT_LNK>, C<AIO::DT_SOCK>,
570     C<AIO::DT_WHT>.
571    
572     C<AIO::DT_UNKNOWN> means just that: readdir does not know. If you need to
573     know, you have to run stat yourself. Also, for speed reasons, the C<$type>
574     scalars are read-only: you can not modify them.
575    
576     =item AIO::READDIR_DIRS_FIRST
577    
578     When this flag is set, then the names will be returned in an order where
579     likely directories come first. This is useful when you need to quickly
580     find directories, or you want to find all directories while avoiding to
581     stat() each entry.
582    
583 root 1.149 If the system returns type information in readdir, then this is used
584     to find directories directly. Otherwise, likely directories are files
585     beginning with ".", or otherwise files with no dots, of which files with
586     short names are tried first.
587    
588 root 1.148 =item AIO::READDIR_STAT_ORDER
589    
590     When this flag is set, then the names will be returned in an order
591     suitable for stat()'ing each one. That is, when you plan to stat()
592     all files in the given directory, then the returned order will likely
593     be fastest.
594    
595     If both this flag and IO::READDIR_DIRS_FIRST are specified, then the
596     likely dirs come first, resulting in a less optimal stat order.
597    
598     =item AIO::READDIR_FOUND_UNKNOWN
599    
600     This flag should not be set when calling C<aio_readdirx>. Instead, it
601     is being set by C<aio_readdirx>, when any of the C<$type>'s found were
602     C<AIO::DT_UNKNOWN>. The absense of this flag therefore indicates that all
603     C<$type>'s are known, which can be used to speed up some algorithms.
604    
605     =back
606 root 1.37
607 root 1.106
608 root 1.98 =item aio_load $path, $data, $callback->($status)
609    
610     This is a composite request that tries to fully load the given file into
611     memory. Status is the same as with aio_read.
612    
613     =cut
614    
615     sub aio_load($$;$) {
616 root 1.123 my ($path, undef, $cb) = @_;
617     my $data = \$_[1];
618 root 1.98
619 root 1.123 my $pri = aioreq_pri;
620     my $grp = aio_group $cb;
621    
622     aioreq_pri $pri;
623     add $grp aio_open $path, O_RDONLY, 0, sub {
624     my $fh = shift
625     or return $grp->result (-1);
626 root 1.98
627     aioreq_pri $pri;
628 root 1.123 add $grp aio_read $fh, 0, (-s $fh), $$data, 0, sub {
629     $grp->result ($_[0]);
630 root 1.98 };
631 root 1.123 };
632 root 1.98
633 root 1.123 $grp
634 root 1.98 }
635    
636 root 1.82 =item aio_copy $srcpath, $dstpath, $callback->($status)
637    
638     Try to copy the I<file> (directories not supported as either source or
639     destination) from C<$srcpath> to C<$dstpath> and call the callback with
640     the C<0> (error) or C<-1> ok.
641    
642 root 1.134 This is a composite request that creates the destination file with
643 root 1.82 mode 0200 and copies the contents of the source file into it using
644     C<aio_sendfile>, followed by restoring atime, mtime, access mode and
645     uid/gid, in that order.
646    
647     If an error occurs, the partial destination file will be unlinked, if
648     possible, except when setting atime, mtime, access mode and uid/gid, where
649     errors are being ignored.
650    
651     =cut
652    
653     sub aio_copy($$;$) {
654 root 1.123 my ($src, $dst, $cb) = @_;
655 root 1.82
656 root 1.123 my $pri = aioreq_pri;
657     my $grp = aio_group $cb;
658 root 1.82
659 root 1.123 aioreq_pri $pri;
660     add $grp aio_open $src, O_RDONLY, 0, sub {
661     if (my $src_fh = $_[0]) {
662 root 1.147 my @stat = stat $src_fh; # hmm, might bock over nfs?
663 root 1.95
664 root 1.123 aioreq_pri $pri;
665     add $grp aio_open $dst, O_CREAT | O_WRONLY | O_TRUNC, 0200, sub {
666     if (my $dst_fh = $_[0]) {
667     aioreq_pri $pri;
668     add $grp aio_sendfile $dst_fh, $src_fh, 0, $stat[7], sub {
669     if ($_[0] == $stat[7]) {
670     $grp->result (0);
671     close $src_fh;
672    
673 root 1.147 my $ch = sub {
674     aioreq_pri $pri;
675     add $grp aio_chmod $dst_fh, $stat[2] & 07777, sub {
676     aioreq_pri $pri;
677     add $grp aio_chown $dst_fh, $stat[4], $stat[5], sub {
678     aioreq_pri $pri;
679     add $grp aio_close $dst_fh;
680     }
681     };
682     };
683 root 1.123
684     aioreq_pri $pri;
685 root 1.147 add $grp aio_utime $dst_fh, $stat[8], $stat[9], sub {
686     if ($_[0] < 0 && $! == ENOSYS) {
687     aioreq_pri $pri;
688     add $grp aio_utime $dst, $stat[8], $stat[9], $ch;
689     } else {
690     $ch->();
691     }
692     };
693 root 1.123 } else {
694     $grp->result (-1);
695     close $src_fh;
696     close $dst_fh;
697    
698     aioreq $pri;
699     add $grp aio_unlink $dst;
700     }
701     };
702     } else {
703     $grp->result (-1);
704     }
705     },
706 root 1.82
707 root 1.123 } else {
708     $grp->result (-1);
709     }
710     };
711 root 1.82
712 root 1.123 $grp
713 root 1.82 }
714    
715     =item aio_move $srcpath, $dstpath, $callback->($status)
716    
717     Try to move the I<file> (directories not supported as either source or
718     destination) from C<$srcpath> to C<$dstpath> and call the callback with
719     the C<0> (error) or C<-1> ok.
720    
721 root 1.137 This is a composite request that tries to rename(2) the file first; if
722     rename fails with C<EXDEV>, it copies the file with C<aio_copy> and, if
723     that is successful, unlinks the C<$srcpath>.
724 root 1.82
725     =cut
726    
727     sub aio_move($$;$) {
728 root 1.123 my ($src, $dst, $cb) = @_;
729 root 1.82
730 root 1.123 my $pri = aioreq_pri;
731     my $grp = aio_group $cb;
732 root 1.82
733 root 1.123 aioreq_pri $pri;
734     add $grp aio_rename $src, $dst, sub {
735     if ($_[0] && $! == EXDEV) {
736     aioreq_pri $pri;
737     add $grp aio_copy $src, $dst, sub {
738     $grp->result ($_[0]);
739 root 1.95
740 root 1.123 if (!$_[0]) {
741     aioreq_pri $pri;
742     add $grp aio_unlink $src;
743     }
744     };
745     } else {
746     $grp->result ($_[0]);
747     }
748     };
749 root 1.82
750 root 1.123 $grp
751 root 1.82 }
752    
753 root 1.40 =item aio_scandir $path, $maxreq, $callback->($dirs, $nondirs)
754    
755 root 1.52 Scans a directory (similar to C<aio_readdir>) but additionally tries to
756 root 1.76 efficiently separate the entries of directory C<$path> into two sets of
757     names, directories you can recurse into (directories), and ones you cannot
758     recurse into (everything else, including symlinks to directories).
759 root 1.52
760 root 1.61 C<aio_scandir> is a composite request that creates of many sub requests_
761     C<$maxreq> specifies the maximum number of outstanding aio requests that
762     this function generates. If it is C<< <= 0 >>, then a suitable default
763 root 1.81 will be chosen (currently 4).
764 root 1.40
765     On error, the callback is called without arguments, otherwise it receives
766     two array-refs with path-relative entry names.
767    
768     Example:
769    
770     aio_scandir $dir, 0, sub {
771     my ($dirs, $nondirs) = @_;
772     print "real directories: @$dirs\n";
773     print "everything else: @$nondirs\n";
774     };
775    
776     Implementation notes.
777    
778     The C<aio_readdir> cannot be avoided, but C<stat()>'ing every entry can.
779    
780 root 1.149 If readdir returns file type information, then this is used directly to
781     find directories.
782    
783     Otherwise, after reading the directory, the modification time, size etc.
784     of the directory before and after the readdir is checked, and if they
785     match (and isn't the current time), the link count will be used to decide
786     how many entries are directories (if >= 2). Otherwise, no knowledge of the
787     number of subdirectories will be assumed.
788    
789     Then entries will be sorted into likely directories a non-initial dot
790     currently) and likely non-directories (see C<aio_readdirx>). Then every
791     entry plus an appended C</.> will be C<stat>'ed, likely directories first,
792     in order of their inode numbers. If that succeeds, it assumes that the
793     entry is a directory or a symlink to directory (which will be checked
794 root 1.52 seperately). This is often faster than stat'ing the entry itself because
795     filesystems might detect the type of the entry without reading the inode
796 root 1.149 data (e.g. ext2fs filetype feature), even on systems that cannot return
797     the filetype information on readdir.
798 root 1.52
799     If the known number of directories (link count - 2) has been reached, the
800     rest of the entries is assumed to be non-directories.
801    
802     This only works with certainty on POSIX (= UNIX) filesystems, which
803     fortunately are the vast majority of filesystems around.
804    
805     It will also likely work on non-POSIX filesystems with reduced efficiency
806     as those tend to return 0 or 1 as link counts, which disables the
807     directory counting heuristic.
808 root 1.40
809     =cut
810    
811 root 1.100 sub aio_scandir($$;$) {
812 root 1.123 my ($path, $maxreq, $cb) = @_;
813    
814     my $pri = aioreq_pri;
815 root 1.40
816 root 1.123 my $grp = aio_group $cb;
817 root 1.80
818 root 1.123 $maxreq = 4 if $maxreq <= 0;
819 root 1.55
820 root 1.123 # stat once
821     aioreq_pri $pri;
822     add $grp aio_stat $path, sub {
823     return $grp->result () if $_[0];
824     my $now = time;
825     my $hash1 = join ":", (stat _)[0,1,3,7,9];
826 root 1.40
827 root 1.123 # read the directory entries
828 root 1.80 aioreq_pri $pri;
829 root 1.148 add $grp aio_readdirx $path, READDIR_DIRS_FIRST, sub {
830 root 1.123 my $entries = shift
831     or return $grp->result ();
832 root 1.40
833 root 1.123 # stat the dir another time
834 root 1.80 aioreq_pri $pri;
835 root 1.123 add $grp aio_stat $path, sub {
836     my $hash2 = join ":", (stat _)[0,1,3,7,9];
837 root 1.95
838 root 1.123 my $ndirs;
839 root 1.95
840 root 1.123 # take the slow route if anything looks fishy
841     if ($hash1 ne $hash2 or (stat _)[9] == $now) {
842     $ndirs = -1;
843     } else {
844     # if nlink == 2, we are finished
845     # on non-posix-fs's, we rely on nlink < 2
846     $ndirs = (stat _)[3] - 2
847     or return $grp->result ([], $entries);
848     }
849    
850     my (@dirs, @nondirs);
851 root 1.40
852 root 1.123 my $statgrp = add $grp aio_group sub {
853     $grp->result (\@dirs, \@nondirs);
854     };
855 root 1.40
856 root 1.123 limit $statgrp $maxreq;
857     feed $statgrp sub {
858     return unless @$entries;
859     my $entry = pop @$entries;
860 root 1.40
861 root 1.123 aioreq_pri $pri;
862     add $statgrp aio_stat "$path/$entry/.", sub {
863     if ($_[0] < 0) {
864     push @nondirs, $entry;
865     } else {
866     # need to check for real directory
867     aioreq_pri $pri;
868     add $statgrp aio_lstat "$path/$entry", sub {
869     if (-d _) {
870     push @dirs, $entry;
871    
872     unless (--$ndirs) {
873     push @nondirs, @$entries;
874     feed $statgrp;
875 root 1.74 }
876 root 1.123 } else {
877     push @nondirs, $entry;
878 root 1.40 }
879     }
880 root 1.123 }
881 root 1.74 };
882 root 1.40 };
883     };
884     };
885 root 1.123 };
886 root 1.55
887 root 1.123 $grp
888 root 1.40 }
889    
890 root 1.99 =item aio_rmtree $path, $callback->($status)
891    
892 root 1.100 Delete a directory tree starting (and including) C<$path>, return the
893     status of the final C<rmdir> only. This is a composite request that
894     uses C<aio_scandir> to recurse into and rmdir directories, and unlink
895     everything else.
896 root 1.99
897     =cut
898    
899     sub aio_rmtree;
900 root 1.100 sub aio_rmtree($;$) {
901 root 1.123 my ($path, $cb) = @_;
902 root 1.99
903 root 1.123 my $pri = aioreq_pri;
904     my $grp = aio_group $cb;
905 root 1.99
906 root 1.123 aioreq_pri $pri;
907     add $grp aio_scandir $path, 0, sub {
908     my ($dirs, $nondirs) = @_;
909 root 1.99
910 root 1.123 my $dirgrp = aio_group sub {
911     add $grp aio_rmdir $path, sub {
912     $grp->result ($_[0]);
913 root 1.99 };
914 root 1.123 };
915 root 1.99
916 root 1.123 (aioreq_pri $pri), add $dirgrp aio_rmtree "$path/$_" for @$dirs;
917     (aioreq_pri $pri), add $dirgrp aio_unlink "$path/$_" for @$nondirs;
918 root 1.99
919 root 1.123 add $grp $dirgrp;
920     };
921 root 1.99
922 root 1.123 $grp
923 root 1.99 }
924    
925 root 1.119 =item aio_sync $callback->($status)
926    
927     Asynchronously call sync and call the callback when finished.
928    
929 root 1.40 =item aio_fsync $fh, $callback->($status)
930 root 1.1
931     Asynchronously call fsync on the given filehandle and call the callback
932     with the fsync result code.
933    
934 root 1.40 =item aio_fdatasync $fh, $callback->($status)
935 root 1.1
936     Asynchronously call fdatasync on the given filehandle and call the
937 root 1.26 callback with the fdatasync result code.
938    
939     If this call isn't available because your OS lacks it or it couldn't be
940     detected, it will be emulated by calling C<fsync> instead.
941 root 1.1
942 root 1.142 =item aio_sync_file_range $fh, $offset, $nbytes, $flags, $callback->($status)
943    
944     Sync the data portion of the file specified by C<$offset> and C<$length>
945     to disk (but NOT the metadata), by calling the Linux-specific
946     sync_file_range call. If sync_file_range is not available or it returns
947     ENOSYS, then fdatasync or fsync is being substituted.
948    
949     C<$flags> can be a combination of C<IO::AIO::SYNC_FILE_RANGE_WAIT_BEFORE>,
950     C<IO::AIO::SYNC_FILE_RANGE_WRITE> and
951     C<IO::AIO::SYNC_FILE_RANGE_WAIT_AFTER>: refer to the sync_file_range
952     manpage for details.
953    
954 root 1.120 =item aio_pathsync $path, $callback->($status)
955    
956     This request tries to open, fsync and close the given path. This is a
957 root 1.135 composite request intended to sync directories after directory operations
958 root 1.120 (E.g. rename). This might not work on all operating systems or have any
959     specific effect, but usually it makes sure that directory changes get
960     written to disc. It works for anything that can be opened for read-only,
961     not just directories.
962    
963     Passes C<0> when everything went ok, and C<-1> on error.
964    
965     =cut
966    
967     sub aio_pathsync($;$) {
968 root 1.123 my ($path, $cb) = @_;
969    
970     my $pri = aioreq_pri;
971     my $grp = aio_group $cb;
972 root 1.120
973 root 1.123 aioreq_pri $pri;
974     add $grp aio_open $path, O_RDONLY, 0, sub {
975     my ($fh) = @_;
976     if ($fh) {
977     aioreq_pri $pri;
978     add $grp aio_fsync $fh, sub {
979     $grp->result ($_[0]);
980 root 1.120
981     aioreq_pri $pri;
982 root 1.123 add $grp aio_close $fh;
983     };
984     } else {
985     $grp->result (-1);
986     }
987     };
988 root 1.120
989 root 1.123 $grp
990 root 1.120 }
991    
992 root 1.58 =item aio_group $callback->(...)
993 root 1.54
994 root 1.55 This is a very special aio request: Instead of doing something, it is a
995     container for other aio requests, which is useful if you want to bundle
996 root 1.71 many requests into a single, composite, request with a definite callback
997     and the ability to cancel the whole request with its subrequests.
998 root 1.55
999     Returns an object of class L<IO::AIO::GRP>. See its documentation below
1000     for more info.
1001    
1002     Example:
1003    
1004     my $grp = aio_group sub {
1005     print "all stats done\n";
1006     };
1007    
1008     add $grp
1009     (aio_stat ...),
1010     (aio_stat ...),
1011     ...;
1012    
1013 root 1.63 =item aio_nop $callback->()
1014    
1015     This is a special request - it does nothing in itself and is only used for
1016     side effects, such as when you want to add a dummy request to a group so
1017     that finishing the requests in the group depends on executing the given
1018     code.
1019    
1020 root 1.64 While this request does nothing, it still goes through the execution
1021     phase and still requires a worker thread. Thus, the callback will not
1022     be executed immediately but only after other requests in the queue have
1023     entered their execution phase. This can be used to measure request
1024     latency.
1025    
1026 root 1.71 =item IO::AIO::aio_busy $fractional_seconds, $callback->() *NOT EXPORTED*
1027 root 1.54
1028     Mainly used for debugging and benchmarking, this aio request puts one of
1029     the request workers to sleep for the given time.
1030    
1031 root 1.56 While it is theoretically handy to have simple I/O scheduling requests
1032 root 1.71 like sleep and file handle readable/writable, the overhead this creates is
1033     immense (it blocks a thread for a long time) so do not use this function
1034     except to put your application under artificial I/O pressure.
1035 root 1.56
1036 root 1.5 =back
1037    
1038 root 1.53 =head2 IO::AIO::REQ CLASS
1039 root 1.52
1040     All non-aggregate C<aio_*> functions return an object of this class when
1041     called in non-void context.
1042    
1043     =over 4
1044    
1045 root 1.65 =item cancel $req
1046 root 1.52
1047     Cancels the request, if possible. Has the effect of skipping execution
1048     when entering the B<execute> state and skipping calling the callback when
1049     entering the the B<result> state, but will leave the request otherwise
1050     untouched. That means that requests that currently execute will not be
1051     stopped and resources held by the request will not be freed prematurely.
1052    
1053 root 1.65 =item cb $req $callback->(...)
1054    
1055     Replace (or simply set) the callback registered to the request.
1056    
1057 root 1.52 =back
1058    
1059 root 1.55 =head2 IO::AIO::GRP CLASS
1060    
1061     This class is a subclass of L<IO::AIO::REQ>, so all its methods apply to
1062     objects of this class, too.
1063    
1064     A IO::AIO::GRP object is a special request that can contain multiple other
1065     aio requests.
1066    
1067     You create one by calling the C<aio_group> constructing function with a
1068     callback that will be called when all contained requests have entered the
1069     C<done> state:
1070    
1071     my $grp = aio_group sub {
1072     print "all requests are done\n";
1073     };
1074    
1075     You add requests by calling the C<add> method with one or more
1076     C<IO::AIO::REQ> objects:
1077    
1078     $grp->add (aio_unlink "...");
1079    
1080 root 1.58 add $grp aio_stat "...", sub {
1081     $_[0] or return $grp->result ("error");
1082    
1083     # add another request dynamically, if first succeeded
1084     add $grp aio_open "...", sub {
1085     $grp->result ("ok");
1086     };
1087     };
1088 root 1.55
1089     This makes it very easy to create composite requests (see the source of
1090     C<aio_move> for an application) that work and feel like simple requests.
1091    
1092 root 1.62 =over 4
1093    
1094     =item * The IO::AIO::GRP objects will be cleaned up during calls to
1095 root 1.55 C<IO::AIO::poll_cb>, just like any other request.
1096    
1097 root 1.62 =item * They can be canceled like any other request. Canceling will cancel not
1098 root 1.59 only the request itself, but also all requests it contains.
1099 root 1.55
1100 root 1.62 =item * They can also can also be added to other IO::AIO::GRP objects.
1101 root 1.55
1102 root 1.62 =item * You must not add requests to a group from within the group callback (or
1103 root 1.60 any later time).
1104    
1105 root 1.62 =back
1106    
1107 root 1.55 Their lifetime, simplified, looks like this: when they are empty, they
1108     will finish very quickly. If they contain only requests that are in the
1109     C<done> state, they will also finish. Otherwise they will continue to
1110     exist.
1111    
1112 root 1.133 That means after creating a group you have some time to add requests
1113     (precisely before the callback has been invoked, which is only done within
1114     the C<poll_cb>). And in the callbacks of those requests, you can add
1115     further requests to the group. And only when all those requests have
1116     finished will the the group itself finish.
1117 root 1.57
1118 root 1.55 =over 4
1119    
1120 root 1.65 =item add $grp ...
1121    
1122 root 1.55 =item $grp->add (...)
1123    
1124 root 1.57 Add one or more requests to the group. Any type of L<IO::AIO::REQ> can
1125     be added, including other groups, as long as you do not create circular
1126     dependencies.
1127    
1128     Returns all its arguments.
1129 root 1.55
1130 root 1.74 =item $grp->cancel_subs
1131    
1132     Cancel all subrequests and clears any feeder, but not the group request
1133     itself. Useful when you queued a lot of events but got a result early.
1134    
1135 root 1.58 =item $grp->result (...)
1136    
1137     Set the result value(s) that will be passed to the group callback when all
1138 root 1.120 subrequests have finished and set the groups errno to the current value
1139 root 1.80 of errno (just like calling C<errno> without an error number). By default,
1140     no argument will be passed and errno is zero.
1141    
1142     =item $grp->errno ([$errno])
1143    
1144     Sets the group errno value to C<$errno>, or the current value of errno
1145     when the argument is missing.
1146    
1147     Every aio request has an associated errno value that is restored when
1148     the callback is invoked. This method lets you change this value from its
1149     default (0).
1150    
1151     Calling C<result> will also set errno, so make sure you either set C<$!>
1152     before the call to C<result>, or call c<errno> after it.
1153 root 1.58
1154 root 1.65 =item feed $grp $callback->($grp)
1155 root 1.60
1156     Sets a feeder/generator on this group: every group can have an attached
1157     generator that generates requests if idle. The idea behind this is that,
1158     although you could just queue as many requests as you want in a group,
1159 root 1.139 this might starve other requests for a potentially long time. For example,
1160     C<aio_scandir> might generate hundreds of thousands C<aio_stat> requests,
1161     delaying any later requests for a long time.
1162 root 1.60
1163     To avoid this, and allow incremental generation of requests, you can
1164     instead a group and set a feeder on it that generates those requests. The
1165 root 1.68 feed callback will be called whenever there are few enough (see C<limit>,
1166 root 1.60 below) requests active in the group itself and is expected to queue more
1167     requests.
1168    
1169 root 1.68 The feed callback can queue as many requests as it likes (i.e. C<add> does
1170     not impose any limits).
1171 root 1.60
1172 root 1.65 If the feed does not queue more requests when called, it will be
1173 root 1.60 automatically removed from the group.
1174    
1175 root 1.138 If the feed limit is C<0> when this method is called, it will be set to
1176     C<2> automatically.
1177 root 1.60
1178     Example:
1179    
1180     # stat all files in @files, but only ever use four aio requests concurrently:
1181    
1182     my $grp = aio_group sub { print "finished\n" };
1183 root 1.68 limit $grp 4;
1184 root 1.65 feed $grp sub {
1185 root 1.60 my $file = pop @files
1186     or return;
1187    
1188     add $grp aio_stat $file, sub { ... };
1189 root 1.65 };
1190 root 1.60
1191 root 1.68 =item limit $grp $num
1192 root 1.60
1193     Sets the feeder limit for the group: The feeder will be called whenever
1194     the group contains less than this many requests.
1195    
1196     Setting the limit to C<0> will pause the feeding process.
1197    
1198 root 1.138 The default value for the limit is C<0>, but note that setting a feeder
1199     automatically bumps it up to C<2>.
1200    
1201 root 1.55 =back
1202    
1203 root 1.5 =head2 SUPPORT FUNCTIONS
1204    
1205 root 1.86 =head3 EVENT PROCESSING AND EVENT LOOP INTEGRATION
1206    
1207 root 1.5 =over 4
1208    
1209     =item $fileno = IO::AIO::poll_fileno
1210    
1211 root 1.20 Return the I<request result pipe file descriptor>. This filehandle must be
1212     polled for reading by some mechanism outside this module (e.g. Event or
1213     select, see below or the SYNOPSIS). If the pipe becomes readable you have
1214     to call C<poll_cb> to check the results.
1215 root 1.5
1216     See C<poll_cb> for an example.
1217    
1218     =item IO::AIO::poll_cb
1219    
1220 root 1.86 Process some outstanding events on the result pipe. You have to call this
1221 root 1.128 regularly. Returns C<0> if all events could be processed, or C<-1> if it
1222     returned earlier for whatever reason. Returns immediately when no events
1223     are outstanding. The amount of events processed depends on the settings of
1224     C<IO::AIO::max_poll_req> and C<IO::AIO::max_poll_time>.
1225 root 1.5
1226 root 1.78 If not all requests were processed for whatever reason, the filehandle
1227 root 1.128 will still be ready when C<poll_cb> returns, so normally you don't have to
1228     do anything special to have it called later.
1229 root 1.78
1230 root 1.20 Example: Install an Event watcher that automatically calls
1231     IO::AIO::poll_cb with high priority:
1232 root 1.5
1233     Event->io (fd => IO::AIO::poll_fileno,
1234     poll => 'r', async => 1,
1235     cb => \&IO::AIO::poll_cb);
1236    
1237 root 1.86 =item IO::AIO::max_poll_reqs $nreqs
1238    
1239     =item IO::AIO::max_poll_time $seconds
1240    
1241     These set the maximum number of requests (default C<0>, meaning infinity)
1242     that are being processed by C<IO::AIO::poll_cb> in one call, respectively
1243     the maximum amount of time (default C<0>, meaning infinity) spent in
1244     C<IO::AIO::poll_cb> to process requests (more correctly the mininum amount
1245     of time C<poll_cb> is allowed to use).
1246 root 1.78
1247 root 1.89 Setting C<max_poll_time> to a non-zero value creates an overhead of one
1248     syscall per request processed, which is not normally a problem unless your
1249     callbacks are really really fast or your OS is really really slow (I am
1250     not mentioning Solaris here). Using C<max_poll_reqs> incurs no overhead.
1251    
1252 root 1.86 Setting these is useful if you want to ensure some level of
1253     interactiveness when perl is not fast enough to process all requests in
1254     time.
1255 root 1.78
1256 root 1.86 For interactive programs, values such as C<0.01> to C<0.1> should be fine.
1257 root 1.78
1258     Example: Install an Event watcher that automatically calls
1259 root 1.89 IO::AIO::poll_cb with low priority, to ensure that other parts of the
1260 root 1.78 program get the CPU sometimes even under high AIO load.
1261    
1262 root 1.86 # try not to spend much more than 0.1s in poll_cb
1263     IO::AIO::max_poll_time 0.1;
1264    
1265     # use a low priority so other tasks have priority
1266 root 1.78 Event->io (fd => IO::AIO::poll_fileno,
1267     poll => 'r', nice => 1,
1268 root 1.86 cb => &IO::AIO::poll_cb);
1269 root 1.78
1270 root 1.5 =item IO::AIO::poll_wait
1271    
1272 root 1.93 If there are any outstanding requests and none of them in the result
1273     phase, wait till the result filehandle becomes ready for reading (simply
1274     does a C<select> on the filehandle. This is useful if you want to
1275     synchronously wait for some requests to finish).
1276 root 1.5
1277     See C<nreqs> for an example.
1278    
1279 root 1.86 =item IO::AIO::poll
1280 root 1.5
1281 root 1.86 Waits until some requests have been handled.
1282 root 1.5
1283 root 1.92 Returns the number of requests processed, but is otherwise strictly
1284     equivalent to:
1285 root 1.5
1286     IO::AIO::poll_wait, IO::AIO::poll_cb
1287 root 1.80
1288 root 1.12 =item IO::AIO::flush
1289    
1290     Wait till all outstanding AIO requests have been handled.
1291    
1292 root 1.13 Strictly equivalent to:
1293    
1294     IO::AIO::poll_wait, IO::AIO::poll_cb
1295     while IO::AIO::nreqs;
1296    
1297 root 1.104 =back
1298    
1299 root 1.86 =head3 CONTROLLING THE NUMBER OF THREADS
1300 root 1.13
1301 root 1.105 =over
1302    
1303 root 1.5 =item IO::AIO::min_parallel $nthreads
1304    
1305 root 1.61 Set the minimum number of AIO threads to C<$nthreads>. The current
1306     default is C<8>, which means eight asynchronous operations can execute
1307     concurrently at any one time (the number of outstanding requests,
1308     however, is unlimited).
1309 root 1.5
1310 root 1.34 IO::AIO starts threads only on demand, when an AIO request is queued and
1311 root 1.86 no free thread exists. Please note that queueing up a hundred requests can
1312     create demand for a hundred threads, even if it turns out that everything
1313     is in the cache and could have been processed faster by a single thread.
1314 root 1.34
1315 root 1.61 It is recommended to keep the number of threads relatively low, as some
1316     Linux kernel versions will scale negatively with the number of threads
1317     (higher parallelity => MUCH higher latency). With current Linux 2.6
1318     versions, 4-32 threads should be fine.
1319 root 1.5
1320 root 1.34 Under most circumstances you don't need to call this function, as the
1321     module selects a default that is suitable for low to moderate load.
1322 root 1.5
1323     =item IO::AIO::max_parallel $nthreads
1324    
1325 root 1.34 Sets the maximum number of AIO threads to C<$nthreads>. If more than the
1326     specified number of threads are currently running, this function kills
1327     them. This function blocks until the limit is reached.
1328    
1329     While C<$nthreads> are zero, aio requests get queued but not executed
1330     until the number of threads has been increased again.
1331 root 1.5
1332     This module automatically runs C<max_parallel 0> at program end, to ensure
1333     that all threads are killed and that there are no outstanding requests.
1334    
1335     Under normal circumstances you don't need to call this function.
1336    
1337 root 1.86 =item IO::AIO::max_idle $nthreads
1338    
1339     Limit the number of threads (default: 4) that are allowed to idle (i.e.,
1340     threads that did not get a request to process within 10 seconds). That
1341     means if a thread becomes idle while C<$nthreads> other threads are also
1342     idle, it will free its resources and exit.
1343    
1344     This is useful when you allow a large number of threads (e.g. 100 or 1000)
1345     to allow for extremely high load situations, but want to free resources
1346     under normal circumstances (1000 threads can easily consume 30MB of RAM).
1347    
1348     The default is probably ok in most situations, especially if thread
1349     creation is fast. If thread creation is very slow on your system you might
1350     want to use larger values.
1351    
1352 root 1.123 =item IO::AIO::max_outstanding $maxreqs
1353 root 1.5
1354 root 1.79 This is a very bad function to use in interactive programs because it
1355     blocks, and a bad way to reduce concurrency because it is inexact: Better
1356     use an C<aio_group> together with a feed callback.
1357    
1358     Sets the maximum number of outstanding requests to C<$nreqs>. If you
1359 root 1.113 do queue up more than this number of requests, the next call to the
1360 root 1.79 C<poll_cb> (and C<poll_some> and other functions calling C<poll_cb>)
1361     function will block until the limit is no longer exceeded.
1362    
1363     The default value is very large, so there is no practical limit on the
1364     number of outstanding requests.
1365    
1366     You can still queue as many requests as you want. Therefore,
1367 root 1.123 C<max_outstanding> is mainly useful in simple scripts (with low values) or
1368 root 1.79 as a stop gap to shield against fatal memory overflow (with large values).
1369 root 1.5
1370 root 1.104 =back
1371    
1372 root 1.86 =head3 STATISTICAL INFORMATION
1373    
1374 root 1.104 =over
1375    
1376 root 1.86 =item IO::AIO::nreqs
1377    
1378     Returns the number of requests currently in the ready, execute or pending
1379     states (i.e. for which their callback has not been invoked yet).
1380    
1381     Example: wait till there are no outstanding requests anymore:
1382    
1383     IO::AIO::poll_wait, IO::AIO::poll_cb
1384     while IO::AIO::nreqs;
1385    
1386     =item IO::AIO::nready
1387    
1388     Returns the number of requests currently in the ready state (not yet
1389     executed).
1390    
1391     =item IO::AIO::npending
1392    
1393     Returns the number of requests currently in the pending state (executed,
1394     but not yet processed by poll_cb).
1395    
1396 root 1.5 =back
1397    
1398 root 1.1 =cut
1399    
1400 root 1.61 min_parallel 8;
1401 root 1.1
1402 root 1.95 END { flush }
1403 root 1.82
1404 root 1.1 1;
1405    
1406 root 1.27 =head2 FORK BEHAVIOUR
1407    
1408 root 1.52 This module should do "the right thing" when the process using it forks:
1409    
1410 root 1.34 Before the fork, IO::AIO enters a quiescent state where no requests
1411     can be added in other threads and no results will be processed. After
1412     the fork the parent simply leaves the quiescent state and continues
1413 root 1.72 request/result processing, while the child frees the request/result queue
1414     (so that the requests started before the fork will only be handled in the
1415     parent). Threads will be started on demand until the limit set in the
1416 root 1.34 parent process has been reached again.
1417 root 1.27
1418 root 1.52 In short: the parent will, after a short pause, continue as if fork had
1419     not been called, while the child will act as if IO::AIO has not been used
1420     yet.
1421    
1422 root 1.60 =head2 MEMORY USAGE
1423    
1424 root 1.72 Per-request usage:
1425    
1426     Each aio request uses - depending on your architecture - around 100-200
1427     bytes of memory. In addition, stat requests need a stat buffer (possibly
1428     a few hundred bytes), readdir requires a result buffer and so on. Perl
1429     scalars and other data passed into aio requests will also be locked and
1430     will consume memory till the request has entered the done state.
1431 root 1.60
1432 root 1.111 This is not awfully much, so queuing lots of requests is not usually a
1433 root 1.60 problem.
1434    
1435 root 1.72 Per-thread usage:
1436    
1437     In the execution phase, some aio requests require more memory for
1438     temporary buffers, and each thread requires a stack and other data
1439     structures (usually around 16k-128k, depending on the OS).
1440    
1441     =head1 KNOWN BUGS
1442    
1443 root 1.73 Known bugs will be fixed in the next release.
1444 root 1.60
1445 root 1.1 =head1 SEE ALSO
1446    
1447 root 1.125 L<AnyEvent::AIO> for easy integration into event loops, L<Coro::AIO> for a
1448     more natural syntax.
1449 root 1.1
1450     =head1 AUTHOR
1451    
1452     Marc Lehmann <schmorp@schmorp.de>
1453     http://home.schmorp.de/
1454    
1455     =cut
1456