| 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) = @_; |
| 9 |
... |
| 10 |
}; |
| 11 |
|
| 12 |
aio_unlink "/tmp/file", sub { }; |
| 13 |
|
| 14 |
aio_read $fh, 30000, 1024, $buffer, 0, sub { |
| 15 |
$_[0] > 0 or die "read error: $!"; |
| 16 |
}; |
| 17 |
|
| 18 |
# Event |
| 19 |
Event->io (fd => IO::AIO::poll_fileno, |
| 20 |
poll => 'r', |
| 21 |
cb => \&IO::AIO::poll_cb); |
| 22 |
|
| 23 |
# Glib/Gtk2 |
| 24 |
add_watch Glib::IO IO::AIO::poll_fileno, |
| 25 |
in => sub { IO::AIO::poll_cb; 1 }; |
| 26 |
|
| 27 |
# Tk |
| 28 |
Tk::Event::IO->fileevent (IO::AIO::poll_fileno, "", |
| 29 |
readable => \&IO::AIO::poll_cb); |
| 30 |
|
| 31 |
# Danga::Socket |
| 32 |
Danga::Socket->AddOtherFds (IO::AIO::poll_fileno => |
| 33 |
\&IO::AIO::poll_cb); |
| 34 |
|
| 35 |
DESCRIPTION |
| 36 |
This module implements asynchronous I/O using whatever means your |
| 37 |
operating system supports. |
| 38 |
|
| 39 |
Currently, a number of threads are started that execute your read/writes |
| 40 |
and signal their completion. You don't need thread support in your libc |
| 41 |
or perl, and the threads created by this module will not be visible to |
| 42 |
the pthreads library. In the future, this module might make use of the |
| 43 |
native aio functions available on many operating systems. However, they |
| 44 |
are often not well-supported (Linux doesn't allow them on normal files |
| 45 |
currently, for example), and they would only support aio_read and |
| 46 |
aio_write, so the remaining functionality would have to be implemented |
| 47 |
using threads anyway. |
| 48 |
|
| 49 |
Although the module will work with in the presence of other threads, it |
| 50 |
is currently not reentrant, so use appropriate locking yourself, always |
| 51 |
call "poll_cb" from within the same thread, or never call "poll_cb" (or |
| 52 |
other "aio_" functions) recursively. |
| 53 |
|
| 54 |
FUNCTIONS |
| 55 |
AIO FUNCTIONS |
| 56 |
All the "aio_*" calls are more or less thin wrappers around the syscall |
| 57 |
with the same name (sans "aio_"). The arguments are similar or |
| 58 |
identical, and they all accept an additional (and optional) $callback |
| 59 |
argument which must be a code reference. This code reference will get |
| 60 |
called with the syscall return code (e.g. most syscalls return -1 on |
| 61 |
error, unlike perl, which usually delivers "false") as it's sole |
| 62 |
argument when the given syscall has been executed asynchronously. |
| 63 |
|
| 64 |
All functions expecting a filehandle keep a copy of the filehandle |
| 65 |
internally until the request has finished. |
| 66 |
|
| 67 |
The pathnames you pass to these routines *must* be absolute and encoded |
| 68 |
in byte form. The reason for the former is that at the time the request |
| 69 |
is being executed, the current working directory could have changed. |
| 70 |
Alternatively, you can make sure that you never change the current |
| 71 |
working directory. |
| 72 |
|
| 73 |
To encode pathnames to byte form, either make sure you either: a) always |
| 74 |
pass in filenames you got from outside (command line, readdir etc.), b) |
| 75 |
are ASCII or ISO 8859-1, c) use the Encode module and encode your |
| 76 |
pathnames to the locale (or other) encoding in effect in the user |
| 77 |
environment, d) use Glib::filename_from_unicode on unicode filenames or |
| 78 |
e) use something else. |
| 79 |
|
| 80 |
aio_open $pathname, $flags, $mode, $callback->($fh) |
| 81 |
Asynchronously open or create a file and call the callback with a |
| 82 |
newly created filehandle for the file. |
| 83 |
|
| 84 |
The pathname passed to "aio_open" must be absolute. See API NOTES, |
| 85 |
above, for an explanation. |
| 86 |
|
| 87 |
The $flags argument is a bitmask. See the "Fcntl" module for a list. |
| 88 |
They are the same as used by "sysopen". |
| 89 |
|
| 90 |
Likewise, $mode specifies the mode of the newly created file, if it |
| 91 |
didn't exist and "O_CREAT" has been given, just like perl's |
| 92 |
"sysopen", except that it is mandatory (i.e. use 0 if you don't |
| 93 |
create new files, and 0666 or 0777 if you do). |
| 94 |
|
| 95 |
Example: |
| 96 |
|
| 97 |
aio_open "/etc/passwd", O_RDONLY, 0, sub { |
| 98 |
if ($_[0]) { |
| 99 |
print "open successful, fh is $_[0]\n"; |
| 100 |
... |
| 101 |
} else { |
| 102 |
die "open failed: $!\n"; |
| 103 |
} |
| 104 |
}; |
| 105 |
|
| 106 |
aio_close $fh, $callback->($status) |
| 107 |
Asynchronously close a file and call the callback with the result |
| 108 |
code. *WARNING:* although accepted, you should not pass in a perl |
| 109 |
filehandle here, as perl will likely close the file descriptor |
| 110 |
another time when the filehandle is destroyed. Normally, you can |
| 111 |
safely call perls "close" or just let filehandles go out of scope. |
| 112 |
|
| 113 |
This is supposed to be a bug in the API, so that might change. It's |
| 114 |
therefore best to avoid this function. |
| 115 |
|
| 116 |
aio_read $fh,$offset,$length, $data,$dataoffset, $callback->($retval) |
| 117 |
aio_write $fh,$offset,$length, $data,$dataoffset, $callback->($retval) |
| 118 |
Reads or writes "length" bytes from the specified "fh" and "offset" |
| 119 |
into the scalar given by "data" and offset "dataoffset" and calls |
| 120 |
the callback without the actual number of bytes read (or -1 on |
| 121 |
error, just like the syscall). |
| 122 |
|
| 123 |
The $data scalar *MUST NOT* be modified in any way while the request |
| 124 |
is outstanding. Modifying it can result in segfaults or WW3 (if the |
| 125 |
necessary/optional hardware is installed). |
| 126 |
|
| 127 |
Example: Read 15 bytes at offset 7 into scalar $buffer, starting at |
| 128 |
offset 0 within the scalar: |
| 129 |
|
| 130 |
aio_read $fh, 7, 15, $buffer, 0, sub { |
| 131 |
$_[0] > 0 or die "read error: $!"; |
| 132 |
print "read $_[0] bytes: <$buffer>\n"; |
| 133 |
}; |
| 134 |
|
| 135 |
aio_sendfile $out_fh, $in_fh, $in_offset, $length, $callback->($retval) |
| 136 |
Tries to copy $length bytes from $in_fh to $out_fh. It starts |
| 137 |
reading at byte offset $in_offset, and starts writing at the current |
| 138 |
file offset of $out_fh. Because of that, it is not safe to issue |
| 139 |
more than one "aio_sendfile" per $out_fh, as they will interfere |
| 140 |
with each other. |
| 141 |
|
| 142 |
This call tries to make use of a native "sendfile" syscall to |
| 143 |
provide zero-copy operation. For this to work, $out_fh should refer |
| 144 |
to a socket, and $in_fh should refer to mmap'able file. |
| 145 |
|
| 146 |
If the native sendfile call fails or is not implemented, it will be |
| 147 |
emulated, so you can call "aio_sendfile" on any type of filehandle |
| 148 |
regardless of the limitations of the operating system. |
| 149 |
|
| 150 |
Please note, however, that "aio_sendfile" can read more bytes from |
| 151 |
$in_fh than are written, and there is no way to find out how many |
| 152 |
bytes have been read from "aio_sendfile" alone, as "aio_sendfile" |
| 153 |
only provides the number of bytes written to $out_fh. Only if the |
| 154 |
result value equals $length one can assume that $length bytes have |
| 155 |
been read. |
| 156 |
|
| 157 |
aio_readahead $fh,$offset,$length, $callback->($retval) |
| 158 |
"aio_readahead" populates the page cache with data from a file so |
| 159 |
that subsequent reads from that file will not block on disk I/O. The |
| 160 |
$offset argument specifies the starting point from which data is to |
| 161 |
be read and $length specifies the number of bytes to be read. I/O is |
| 162 |
performed in whole pages, so that offset is effectively rounded down |
| 163 |
to a page boundary and bytes are read up to the next page boundary |
| 164 |
greater than or equal to (off-set+length). "aio_readahead" does not |
| 165 |
read beyond the end of the file. The current file offset of the file |
| 166 |
is left unchanged. |
| 167 |
|
| 168 |
If that syscall doesn't exist (likely if your OS isn't Linux) it |
| 169 |
will be emulated by simply reading the data, which would have a |
| 170 |
similar effect. |
| 171 |
|
| 172 |
aio_stat $fh_or_path, $callback->($status) |
| 173 |
aio_lstat $fh, $callback->($status) |
| 174 |
Works like perl's "stat" or "lstat" in void context. The callback |
| 175 |
will be called after the stat and the results will be available |
| 176 |
using "stat _" or "-s _" etc... |
| 177 |
|
| 178 |
The pathname passed to "aio_stat" must be absolute. See API NOTES, |
| 179 |
above, for an explanation. |
| 180 |
|
| 181 |
Currently, the stats are always 64-bit-stats, i.e. instead of |
| 182 |
returning an error when stat'ing a large file, the results will be |
| 183 |
silently truncated unless perl itself is compiled with large file |
| 184 |
support. |
| 185 |
|
| 186 |
Example: Print the length of /etc/passwd: |
| 187 |
|
| 188 |
aio_stat "/etc/passwd", sub { |
| 189 |
$_[0] and die "stat failed: $!"; |
| 190 |
print "size is ", -s _, "\n"; |
| 191 |
}; |
| 192 |
|
| 193 |
aio_unlink $pathname, $callback->($status) |
| 194 |
Asynchronously unlink (delete) a file and call the callback with the |
| 195 |
result code. |
| 196 |
|
| 197 |
aio_rmdir $pathname, $callback->($status) |
| 198 |
Asynchronously rmdir (delete) a directory and call the callback with |
| 199 |
the result code. |
| 200 |
|
| 201 |
aio_readdir $pathname $callback->($entries) |
| 202 |
Unlike the POSIX call of the same name, "aio_readdir" reads an |
| 203 |
entire directory (i.e. opendir + readdir + closedir). The entries |
| 204 |
will not be sorted, and will NOT include the "." and ".." entries. |
| 205 |
|
| 206 |
The callback a single argument which is either "undef" or an |
| 207 |
array-ref with the filenames. |
| 208 |
|
| 209 |
aio_scandir $path, $maxreq, $callback->($dirs, $nondirs) |
| 210 |
Scans a directory (similar to "aio_readdir") and tries to separate |
| 211 |
the entries of directory $path into two sets of names, ones you can |
| 212 |
recurse into (directories), and ones you cannot recurse into |
| 213 |
(everything else). |
| 214 |
|
| 215 |
"aio_scandir" is a composite request that consists of many |
| 216 |
aio-primitives. $maxreq specifies the maximum number of outstanding |
| 217 |
aio requests that this function generates. If it is "<= 0", then a |
| 218 |
suitable default will be chosen (currently 8). |
| 219 |
|
| 220 |
On error, the callback is called without arguments, otherwise it |
| 221 |
receives two array-refs with path-relative entry names. |
| 222 |
|
| 223 |
Example: |
| 224 |
|
| 225 |
aio_scandir $dir, 0, sub { |
| 226 |
my ($dirs, $nondirs) = @_; |
| 227 |
print "real directories: @$dirs\n"; |
| 228 |
print "everything else: @$nondirs\n"; |
| 229 |
}; |
| 230 |
|
| 231 |
Implementation notes. |
| 232 |
|
| 233 |
The "aio_readdir" cannot be avoided, but "stat()"'ing every entry |
| 234 |
can. |
| 235 |
|
| 236 |
After reading the directory, the modification time, size etc. of the |
| 237 |
directory before and after the readdir is checked, and if they |
| 238 |
match, the link count will be used to decide how many entries are |
| 239 |
directories (if >= 2). Otherwise, no knowledge of the number of |
| 240 |
subdirectories will be assumed. |
| 241 |
|
| 242 |
Then entires will be sorted into likely directories (everything |
| 243 |
without a non-initial dot) and likely non-directories (everything |
| 244 |
else). Then every entry + "/." will be "stat"'ed, likely directories |
| 245 |
first. This is often faster because filesystems might detect the |
| 246 |
type of the entry without reading the inode data (e.g. ext2s |
| 247 |
filetype feature). If that succeeds, it assumes that the entry is a |
| 248 |
directory or a symlink to directory (which will be checked |
| 249 |
seperately). |
| 250 |
|
| 251 |
If the known number of directories has been reached, the rest of the |
| 252 |
entries is assumed to be non-directories. |
| 253 |
|
| 254 |
aio_fsync $fh, $callback->($status) |
| 255 |
Asynchronously call fsync on the given filehandle and call the |
| 256 |
callback with the fsync result code. |
| 257 |
|
| 258 |
aio_fdatasync $fh, $callback->($status) |
| 259 |
Asynchronously call fdatasync on the given filehandle and call the |
| 260 |
callback with the fdatasync result code. |
| 261 |
|
| 262 |
If this call isn't available because your OS lacks it or it couldn't |
| 263 |
be detected, it will be emulated by calling "fsync" instead. |
| 264 |
|
| 265 |
SUPPORT FUNCTIONS |
| 266 |
$fileno = IO::AIO::poll_fileno |
| 267 |
Return the *request result pipe file descriptor*. This filehandle |
| 268 |
must be polled for reading by some mechanism outside this module |
| 269 |
(e.g. Event or select, see below or the SYNOPSIS). If the pipe |
| 270 |
becomes readable you have to call "poll_cb" to check the results. |
| 271 |
|
| 272 |
See "poll_cb" for an example. |
| 273 |
|
| 274 |
IO::AIO::poll_cb |
| 275 |
Process all outstanding events on the result pipe. You have to call |
| 276 |
this regularly. Returns the number of events processed. Returns |
| 277 |
immediately when no events are outstanding. |
| 278 |
|
| 279 |
Example: Install an Event watcher that automatically calls |
| 280 |
IO::AIO::poll_cb with high priority: |
| 281 |
|
| 282 |
Event->io (fd => IO::AIO::poll_fileno, |
| 283 |
poll => 'r', async => 1, |
| 284 |
cb => \&IO::AIO::poll_cb); |
| 285 |
|
| 286 |
IO::AIO::poll_wait |
| 287 |
Wait till the result filehandle becomes ready for reading (simply |
| 288 |
does a "select" on the filehandle. This is useful if you want to |
| 289 |
synchronously wait for some requests to finish). |
| 290 |
|
| 291 |
See "nreqs" for an example. |
| 292 |
|
| 293 |
IO::AIO::nreqs |
| 294 |
Returns the number of requests currently outstanding (i.e. for which |
| 295 |
their callback has not been invoked yet). |
| 296 |
|
| 297 |
Example: wait till there are no outstanding requests anymore: |
| 298 |
|
| 299 |
IO::AIO::poll_wait, IO::AIO::poll_cb |
| 300 |
while IO::AIO::nreqs; |
| 301 |
|
| 302 |
IO::AIO::flush |
| 303 |
Wait till all outstanding AIO requests have been handled. |
| 304 |
|
| 305 |
Strictly equivalent to: |
| 306 |
|
| 307 |
IO::AIO::poll_wait, IO::AIO::poll_cb |
| 308 |
while IO::AIO::nreqs; |
| 309 |
|
| 310 |
IO::AIO::poll |
| 311 |
Waits until some requests have been handled. |
| 312 |
|
| 313 |
Strictly equivalent to: |
| 314 |
|
| 315 |
IO::AIO::poll_wait, IO::AIO::poll_cb |
| 316 |
if IO::AIO::nreqs; |
| 317 |
|
| 318 |
IO::AIO::min_parallel $nthreads |
| 319 |
Set the minimum number of AIO threads to $nthreads. The current |
| 320 |
default is 4, which means four asynchronous operations can be done |
| 321 |
at one time (the number of outstanding operations, however, is |
| 322 |
unlimited). |
| 323 |
|
| 324 |
IO::AIO starts threads only on demand, when an AIO request is queued |
| 325 |
and no free thread exists. |
| 326 |
|
| 327 |
It is recommended to keep the number of threads low, as some Linux |
| 328 |
kernel versions will scale negatively with the number of threads |
| 329 |
(higher parallelity => MUCH higher latency). With current Linux 2.6 |
| 330 |
versions, 4-32 threads should be fine. |
| 331 |
|
| 332 |
Under most circumstances you don't need to call this function, as |
| 333 |
the module selects a default that is suitable for low to moderate |
| 334 |
load. |
| 335 |
|
| 336 |
IO::AIO::max_parallel $nthreads |
| 337 |
Sets the maximum number of AIO threads to $nthreads. If more than |
| 338 |
the specified number of threads are currently running, this function |
| 339 |
kills them. This function blocks until the limit is reached. |
| 340 |
|
| 341 |
While $nthreads are zero, aio requests get queued but not executed |
| 342 |
until the number of threads has been increased again. |
| 343 |
|
| 344 |
This module automatically runs "max_parallel 0" at program end, to |
| 345 |
ensure that all threads are killed and that there are no outstanding |
| 346 |
requests. |
| 347 |
|
| 348 |
Under normal circumstances you don't need to call this function. |
| 349 |
|
| 350 |
$oldnreqs = IO::AIO::max_outstanding $nreqs |
| 351 |
Sets the maximum number of outstanding requests to $nreqs. If you |
| 352 |
try to queue up more than this number of requests, the caller will |
| 353 |
block until some requests have been handled. |
| 354 |
|
| 355 |
The default is very large, so normally there is no practical limit. |
| 356 |
If you queue up many requests in a loop it often improves speed if |
| 357 |
you set this to a relatively low number, such as 100. |
| 358 |
|
| 359 |
Under normal circumstances you don't need to call this function. |
| 360 |
|
| 361 |
FORK BEHAVIOUR |
| 362 |
Before the fork, IO::AIO enters a quiescent state where no requests can |
| 363 |
be added in other threads and no results will be processed. After the |
| 364 |
fork the parent simply leaves the quiescent state and continues |
| 365 |
request/result processing, while the child clears the request/result |
| 366 |
queue (so the requests started before the fork will only be handled in |
| 367 |
the parent). Threats will be started on demand until the limit ste in |
| 368 |
the parent process has been reached again. |
| 369 |
|
| 370 |
SEE ALSO |
| 371 |
Coro, Linux::AIO. |
| 372 |
|
| 373 |
AUTHOR |
| 374 |
Marc Lehmann <schmorp@schmorp.de> |
| 375 |
http://home.schmorp.de/ |
| 376 |
|
