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