ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/IO-AIO/README
(Generate patch)

Comparing IO-AIO/README (file contents):
Revision 1.45 by root, Thu Dec 30 07:19:31 2010 UTC vs.
Revision 1.49 by root, Mon Jul 18 03:09:06 2011 UTC

… …
168 aio_unlink $pathname, $callback->($status) 168 aio_unlink $pathname, $callback->($status)
169 aio_mknod $path, $mode, $dev, $callback->($status) 169 aio_mknod $path, $mode, $dev, $callback->($status)
170 aio_link $srcpath, $dstpath, $callback->($status) 170 aio_link $srcpath, $dstpath, $callback->($status)
171 aio_symlink $srcpath, $dstpath, $callback->($status) 171 aio_symlink $srcpath, $dstpath, $callback->($status)
172 aio_readlink $path, $callback->($link) 172 aio_readlink $path, $callback->($link)
173 aio_realpath $path, $callback->($link)
173 aio_rename $srcpath, $dstpath, $callback->($status) 174 aio_rename $srcpath, $dstpath, $callback->($status)
174 aio_mkdir $pathname, $mode, $callback->($status) 175 aio_mkdir $pathname, $mode, $callback->($status)
175 aio_rmdir $pathname, $callback->($status) 176 aio_rmdir $pathname, $callback->($status)
176 aio_readdir $pathname, $callback->($entries) 177 aio_readdir $pathname, $callback->($entries)
177 aio_readdirx $pathname, $flags, $callback->($entries, $flags) 178 aio_readdirx $pathname, $flags, $callback->($entries, $flags)
… …
204 IO::AIO::max_poll_reqs $nreqs 205 IO::AIO::max_poll_reqs $nreqs
205 IO::AIO::max_poll_time $seconds 206 IO::AIO::max_poll_time $seconds
206 IO::AIO::min_parallel $nthreads 207 IO::AIO::min_parallel $nthreads
207 IO::AIO::max_parallel $nthreads 208 IO::AIO::max_parallel $nthreads
208 IO::AIO::max_idle $nthreads 209 IO::AIO::max_idle $nthreads
210 IO::AIO::idle_timeout $seconds
209 IO::AIO::max_outstanding $maxreqs 211 IO::AIO::max_outstanding $maxreqs
210 IO::AIO::nreqs 212 IO::AIO::nreqs
211 IO::AIO::nready 213 IO::AIO::nready
212 IO::AIO::npending 214 IO::AIO::npending
213 215
… …
305 } else { 307 } else {
306 die "open failed: $!\n"; 308 die "open failed: $!\n";
307 } 309 }
308 }; 310 };
309 311
312 In addition to all the common open modes/flags ("O_RDONLY",
313 "O_WRONLY", "O_RDWR", "O_CREAT", "O_TRUNC", "O_EXCL" and
314 "O_APPEND"), the following POSIX and non-POSIX constants are
315 available (missing ones on your system are, as usual, 0):
316
317 "O_ASYNC", "O_DIRECT", "O_NOATIME", "O_CLOEXEC", "O_NOCTTY",
318 "O_NOFOLLOW", "O_NONBLOCK", "O_EXEC", "O_SEARCH", "O_DIRECTORY",
319 "O_DSYNC", "O_RSYNC", "O_SYNC" and "O_TTY_INIT".
320
310 aio_close $fh, $callback->($status) 321 aio_close $fh, $callback->($status)
311 Asynchronously close a file and call the callback with the result 322 Asynchronously close a file and call the callback with the result
312 code. 323 code.
313 324
314 Unfortunately, you can't do this to perl. Perl *insists* very 325 Unfortunately, you can't do this to perl. Perl *insists* very
… …
357 aio_sendfile $out_fh, $in_fh, $in_offset, $length, $callback->($retval) 368 aio_sendfile $out_fh, $in_fh, $in_offset, $length, $callback->($retval)
358 Tries to copy $length bytes from $in_fh to $out_fh. It starts 369 Tries to copy $length bytes from $in_fh to $out_fh. It starts
359 reading at byte offset $in_offset, and starts writing at the current 370 reading at byte offset $in_offset, and starts writing at the current
360 file offset of $out_fh. Because of that, it is not safe to issue 371 file offset of $out_fh. Because of that, it is not safe to issue
361 more than one "aio_sendfile" per $out_fh, as they will interfere 372 more than one "aio_sendfile" per $out_fh, as they will interfere
362 with each other. 373 with each other. The same $in_fh works fine though, as this function
374 does not move or use the file offset of $in_fh.
363 375
364 Please note that "aio_sendfile" can read more bytes from $in_fh than 376 Please note that "aio_sendfile" can read more bytes from $in_fh than
365 are written, and there is no way to find out how many bytes have 377 are written, and there is no way to find out how many more bytes
366 been read from "aio_sendfile" alone, as "aio_sendfile" only provides 378 have been read from "aio_sendfile" alone, as "aio_sendfile" only
367 the number of bytes written to $out_fh. Only if the result value 379 provides the number of bytes written to $out_fh. Only if the result
368 equals $length one can assume that $length bytes have been read. 380 value equals $length one can assume that $length bytes have been
381 read.
369 382
370 Unlike with other "aio_" functions, it makes a lot of sense to use 383 Unlike with other "aio_" functions, it makes a lot of sense to use
371 "aio_sendfile" on non-blocking sockets, as long as one end 384 "aio_sendfile" on non-blocking sockets, as long as one end
372 (typically the $in_fh) is a file - the file I/O will then be 385 (typically the $in_fh) is a file - the file I/O will then be
373 asynchronous, while the socket I/O will be non-blocking. Note, 386 asynchronous, while the socket I/O will be non-blocking. Note,
374 however, that you can run into a trap where "aio_sendfile" reads 387 however, that you can run into a trap where "aio_sendfile" reads
375 some data with readahead, then fails to write all data, and when the 388 some data with readahead, then fails to write all data, and when the
376 socket is ready the next time, the data in the cache is already 389 socket is ready the next time, the data in the cache is already
377 lost, forcing "aio_sendfile" to again hit the disk. Explicit 390 lost, forcing "aio_sendfile" to again hit the disk. Explicit
378 "aio_read" + "aio_write" let's you control resource usage much 391 "aio_read" + "aio_write" let's you better control resource usage.
379 better.
380 392
381 This call tries to make use of a native "sendfile" syscall to 393 This call tries to make use of a native "sendfile"-like syscall to
382 provide zero-copy operation. For this to work, $out_fh should refer 394 provide zero-copy operation. For this to work, $out_fh should refer
383 to a socket, and $in_fh should refer to an mmap'able file. 395 to a socket, and $in_fh should refer to an mmap'able file.
384 396
385 If a native sendfile cannot be found or it fails with "ENOSYS", 397 If a native sendfile cannot be found or it fails with "ENOSYS",
386 "ENOTSUP", "EOPNOTSUPP", "EAFNOSUPPORT", "EPROTOTYPE" or "ENOTSOCK", 398 "EINVAL", "ENOTSUP", "EOPNOTSUPP", "EAFNOSUPPORT", "EPROTOTYPE" or
387 it will be emulated, so you can call "aio_sendfile" on any type of 399 "ENOTSOCK", it will be emulated, so you can call "aio_sendfile" on
388 filehandle regardless of the limitations of the operating system. 400 any type of filehandle regardless of the limitations of the
401 operating system.
402
403 As native sendfile syscalls (as practically any non-POSIX interface
404 hacked together in a hurry to improve benchmark numbers) tend to be
405 rather buggy on many systems, this implementation tries to work
406 around some known bugs in Linux and FreeBSD kernels (probably
407 others, too), but that might fail, so you really really should check
408 the return value of "aio_sendfile" - fewre bytes than expected might
409 have been transferred.
389 410
390 aio_readahead $fh,$offset,$length, $callback->($retval) 411 aio_readahead $fh,$offset,$length, $callback->($retval)
391 "aio_readahead" populates the page cache with data from a file so 412 "aio_readahead" populates the page cache with data from a file so
392 that subsequent reads from that file will not block on disk I/O. The 413 that subsequent reads from that file will not block on disk I/O. The
393 $offset argument specifies the starting point from which data is to 414 $offset argument specifies the starting point from which data is to
… …
413 434
414 Currently, the stats are always 64-bit-stats, i.e. instead of 435 Currently, the stats are always 64-bit-stats, i.e. instead of
415 returning an error when stat'ing a large file, the results will be 436 returning an error when stat'ing a large file, the results will be
416 silently truncated unless perl itself is compiled with large file 437 silently truncated unless perl itself is compiled with large file
417 support. 438 support.
439
440 To help interpret the mode and dev/rdev stat values, IO::AIO offers
441 the following constants and functions (if not implemented, the
442 constants will be 0 and the functions will either "croak" or fall
443 back on traditional behaviour).
444
445 "S_IFMT", "S_IFIFO", "S_IFCHR", "S_IFBLK", "S_IFLNK", "S_IFREG",
446 "S_IFDIR", "S_IFWHT", "S_IFSOCK", "IO::AIO::major $dev_t",
447 "IO::AIO::minor $dev_t", "IO::AIO::makedev $major, $minor".
418 448
419 Example: Print the length of /etc/passwd: 449 Example: Print the length of /etc/passwd:
420 450
421 aio_stat "/etc/passwd", sub { 451 aio_stat "/etc/passwd", sub {
422 $_[0] and die "stat failed: $!"; 452 $_[0] and die "stat failed: $!";
… …
511 541
512 The only (POSIX-) portable way of calling this function is: 542 The only (POSIX-) portable way of calling this function is:
513 543
514 aio_mknod $path, IO::AIO::S_IFIFO | $mode, 0, sub { ... 544 aio_mknod $path, IO::AIO::S_IFIFO | $mode, 0, sub { ...
515 545
546 See "aio_stat" for info about some potentially helpful extra
547 constants and functions.
548
516 aio_link $srcpath, $dstpath, $callback->($status) 549 aio_link $srcpath, $dstpath, $callback->($status)
517 Asynchronously create a new link to the existing object at $srcpath 550 Asynchronously create a new link to the existing object at $srcpath
518 at the path $dstpath and call the callback with the result code. 551 at the path $dstpath and call the callback with the result code.
519 552
520 aio_symlink $srcpath, $dstpath, $callback->($status) 553 aio_symlink $srcpath, $dstpath, $callback->($status)
… …
525 aio_readlink $path, $callback->($link) 558 aio_readlink $path, $callback->($link)
526 Asynchronously read the symlink specified by $path and pass it to 559 Asynchronously read the symlink specified by $path and pass it to
527 the callback. If an error occurs, nothing or undef gets passed to 560 the callback. If an error occurs, nothing or undef gets passed to
528 the callback. 561 the callback.
529 562
563 aio_realpath $path, $callback->($path)
564 Asynchronously make the path absolute and resolve any symlinks in
565 $path. The resulting path only consists of directories (Same as
566 Cwd::realpath).
567
568 This request can be used to get the absolute path of the current
569 working directory by passing it a path of . (a single dot).
570
530 aio_rename $srcpath, $dstpath, $callback->($status) 571 aio_rename $srcpath, $dstpath, $callback->($status)
531 Asynchronously rename the object at $srcpath to $dstpath, just as 572 Asynchronously rename the object at $srcpath to $dstpath, just as
532 rename(2) and call the callback with the result code. 573 rename(2) and call the callback with the result code.
533 574
534 aio_mkdir $pathname, $mode, $callback->($status) 575 aio_mkdir $pathname, $mode, $callback->($status)
… …
556 The flags are a combination of the following constants, ORed 597 The flags are a combination of the following constants, ORed
557 together (the flags will also be passed to the callback, possibly 598 together (the flags will also be passed to the callback, possibly
558 modified): 599 modified):
559 600
560 IO::AIO::READDIR_DENTS 601 IO::AIO::READDIR_DENTS
561 When this flag is off, then the callback gets an arrayref with 602 When this flag is off, then the callback gets an arrayref
562 of names only (as with "aio_readdir"), otherwise it gets an 603 consisting of names only (as with "aio_readdir"), otherwise it
563 arrayref with "[$name, $type, $inode]" arrayrefs, each 604 gets an arrayref with "[$name, $type, $inode]" arrayrefs, each
564 describing a single directory entry in more detail. 605 describing a single directory entry in more detail.
565 606
566 $name is the name of the entry. 607 $name is the name of the entry.
567 608
568 $type is one of the "IO::AIO::DT_xxx" constants: 609 $type is one of the "IO::AIO::DT_xxx" constants:
… …
581 unspecified content on systems that do not deliver the inode 622 unspecified content on systems that do not deliver the inode
582 information. 623 information.
583 624
584 IO::AIO::READDIR_DIRS_FIRST 625 IO::AIO::READDIR_DIRS_FIRST
585 When this flag is set, then the names will be returned in an 626 When this flag is set, then the names will be returned in an
586 order where likely directories come first. This is useful when 627 order where likely directories come first, in optimal stat
587 you need to quickly find directories, or you want to find all 628 order. This is useful when you need to quickly find directories,
588 directories while avoiding to stat() each entry. 629 or you want to find all directories while avoiding to stat()
630 each entry.
589 631
590 If the system returns type information in readdir, then this is 632 If the system returns type information in readdir, then this is
591 used to find directories directly. Otherwise, likely directories 633 used to find directories directly. Otherwise, likely directories
592 are files beginning with ".", or otherwise files with no dots, 634 are names beginning with ".", or otherwise names with no dots,
593 of which files with short names are tried first. 635 of which names with short names are tried first.
594 636
595 IO::AIO::READDIR_STAT_ORDER 637 IO::AIO::READDIR_STAT_ORDER
596 When this flag is set, then the names will be returned in an 638 When this flag is set, then the names will be returned in an
597 order suitable for stat()'ing each one. That is, when you plan 639 order suitable for stat()'ing each one. That is, when you plan
598 to stat() all files in the given directory, then the returned 640 to stat() all files in the given directory, then the returned
… …
1009 1051
1010 See "poll_cb" for an example. 1052 See "poll_cb" for an example.
1011 1053
1012 IO::AIO::poll_cb 1054 IO::AIO::poll_cb
1013 Process some outstanding events on the result pipe. You have to call 1055 Process some outstanding events on the result pipe. You have to call
1014 this regularly. Returns 0 if all events could be processed, or -1 if 1056 this regularly. Returns 0 if all events could be processed (or there
1015 it returned earlier for whatever reason. Returns immediately when no 1057 were no events to process), or -1 if it returned earlier for
1016 events are outstanding. The amount of events processed depends on 1058 whatever reason. Returns immediately when no events are outstanding.
1017 the settings of "IO::AIO::max_poll_req" and 1059 The amount of events processed depends on the settings of
1018 "IO::AIO::max_poll_time". 1060 "IO::AIO::max_poll_req" and "IO::AIO::max_poll_time".
1019 1061
1020 If not all requests were processed for whatever reason, the 1062 If not all requests were processed for whatever reason, the
1021 filehandle will still be ready when "poll_cb" returns, so normally 1063 filehandle will still be ready when "poll_cb" returns, so normally
1022 you don't have to do anything special to have it called later. 1064 you don't have to do anything special to have it called later.
1065
1066 Apart from calling "IO::AIO::poll_cb" when the event filehandle
1067 becomes ready, it can be beneficial to call this function from loops
1068 which submit a lot of requests, to make sure the results get
1069 processed when they become available and not just when the loop is
1070 finished and the event loop takes over again. This function returns
1071 very fast when there are no outstanding requests.
1023 1072
1024 Example: Install an Event watcher that automatically calls 1073 Example: Install an Event watcher that automatically calls
1025 IO::AIO::poll_cb with high priority (more examples can be found in 1074 IO::AIO::poll_cb with high priority (more examples can be found in
1026 the SYNOPSIS section, at the top of this document): 1075 the SYNOPSIS section, at the top of this document):
1027 1076
… …
1121 1170
1122 Under normal circumstances you don't need to call this function. 1171 Under normal circumstances you don't need to call this function.
1123 1172
1124 IO::AIO::max_idle $nthreads 1173 IO::AIO::max_idle $nthreads
1125 Limit the number of threads (default: 4) that are allowed to idle 1174 Limit the number of threads (default: 4) that are allowed to idle
1126 (i.e., threads that did not get a request to process within 10 1175 (i.e., threads that did not get a request to process within the idle
1127 seconds). That means if a thread becomes idle while $nthreads other 1176 timeout (default: 10 seconds). That means if a thread becomes idle
1128 threads are also idle, it will free its resources and exit. 1177 while $nthreads other threads are also idle, it will free its
1178 resources and exit.
1129 1179
1130 This is useful when you allow a large number of threads (e.g. 100 or 1180 This is useful when you allow a large number of threads (e.g. 100 or
1131 1000) to allow for extremely high load situations, but want to free 1181 1000) to allow for extremely high load situations, but want to free
1132 resources under normal circumstances (1000 threads can easily 1182 resources under normal circumstances (1000 threads can easily
1133 consume 30MB of RAM). 1183 consume 30MB of RAM).
1134 1184
1135 The default is probably ok in most situations, especially if thread 1185 The default is probably ok in most situations, especially if thread
1136 creation is fast. If thread creation is very slow on your system you 1186 creation is fast. If thread creation is very slow on your system you
1137 might want to use larger values. 1187 might want to use larger values.
1138 1188
1189 IO::AIO::idle_timeout $seconds
1190 Sets the minimum idle timeout (default 10) after which worker
1191 threads are allowed to exit. SEe "IO::AIO::max_idle".
1192
1139 IO::AIO::max_outstanding $maxreqs 1193 IO::AIO::max_outstanding $maxreqs
1194 Sets the maximum number of outstanding requests to $nreqs. If you do
1195 queue up more than this number of requests, the next call to
1196 "IO::AIO::poll_cb" (and other functions calling "poll_cb", such as
1197 "IO::AIO::flush" or "IO::AIO::poll") will block until the limit is
1198 no longer exceeded.
1199
1200 In other words, this setting does not enforce a queue limit, but can
1201 be used to make poll functions block if the limit is exceeded.
1202
1140 This is a very bad function to use in interactive programs because 1203 This is a very bad function to use in interactive programs because
1141 it blocks, and a bad way to reduce concurrency because it is 1204 it blocks, and a bad way to reduce concurrency because it is
1142 inexact: Better use an "aio_group" together with a feed callback. 1205 inexact: Better use an "aio_group" together with a feed callback.
1143 1206
1144 Sets the maximum number of outstanding requests to $nreqs. If you do 1207 It's main use is in scripts without an event loop - when you want to
1145 queue up more than this number of requests, the next call to the 1208 stat a lot of files, you can write somehting like this:
1146 "poll_cb" (and "poll_some" and other functions calling "poll_cb")
1147 function will block until the limit is no longer exceeded.
1148 1209
1149 The default value is very large, so there is no practical limit on 1210 IO::AIO::max_outstanding 32;
1211
1212 for my $path (...) {
1213 aio_stat $path , ...;
1214 IO::AIO::poll_cb;
1215 }
1216
1217 IO::AIO::flush;
1218
1219 The call to "poll_cb" inside the loop will normally return
1220 instantly, but as soon as more thna 32 reqeusts are in-flight, it
1221 will block until some requests have been handled. This keeps the
1222 loop from pushing a large number of "aio_stat" requests onto the
1223 queue.
1224
1225 The default value for "max_outstanding" is very large, so there is
1150 the number of outstanding requests. 1226 no practical limit on the number of outstanding requests.
1151
1152 You can still queue as many requests as you want. Therefore,
1153 "max_outstanding" is mainly useful in simple scripts (with low
1154 values) or as a stop gap to shield against fatal memory overflow
1155 (with large values).
1156 1227
1157 STATISTICAL INFORMATION 1228 STATISTICAL INFORMATION
1158 IO::AIO::nreqs 1229 IO::AIO::nreqs
1159 Returns the number of requests currently in the ready, execute or 1230 Returns the number of requests currently in the ready, execute or
1160 pending states (i.e. for which their callback has not been invoked 1231 pending states (i.e. for which their callback has not been invoked
… …
1306 # Danga::Socket integration 1377 # Danga::Socket integration
1307 Danga::Socket->AddOtherFds (IO::AIO::poll_fileno => 1378 Danga::Socket->AddOtherFds (IO::AIO::poll_fileno =>
1308 \&IO::AIO::poll_cb); 1379 \&IO::AIO::poll_cb);
1309 1380
1310 FORK BEHAVIOUR 1381 FORK BEHAVIOUR
1311 This module should do "the right thing" when the process using it forks: 1382 Usage of pthreads in a program changes the semantics of fork
1383 considerably. Specifically, only async-safe functions can be called
1384 after fork. Perl doesn't know about this, so in general, you cannot call
1385 fork with defined behaviour in perl if pthreads are involved. IO::AIO
1386 uses pthreads, so this applies, but many other extensions and (for
1387 inexplicable reasons) perl itself often is linked against pthreads, so
1388 this limitation applies to quite a lot of perls.
1312 1389
1313 Before the fork, IO::AIO enters a quiescent state where no requests can 1390 This module no longer tries to fight your OS, or POSIX. That means
1314 be added in other threads and no results will be processed. After the 1391 IO::AIO only works in the process that loaded it. Forking is fully
1315 fork the parent simply leaves the quiescent state and continues 1392 supported, but using IO::AIO in the child is not.
1316 request/result processing, while the child frees the request/result
1317 queue (so that the requests started before the fork will only be handled
1318 in the parent). Threads will be started on demand until the limit set in
1319 the parent process has been reached again.
1320 1393
1321 In short: the parent will, after a short pause, continue as if fork had 1394 You might get around by not *using* IO::AIO before (or after) forking.
1322 not been called, while the child will act as if IO::AIO has not been 1395 You could also try to call the IO::AIO::reinit function in the child:
1323 used yet. 1396
1397 IO::AIO::reinit
1398 Abondons all current requests and I/O threads and simply
1399 reinitialises all data structures. This is not an operation
1400 suppported by any standards, but happens to work on GNU/Linux and
1401 some newer BSD systems.
1402
1403 The only reasonable use for this function is to call it after
1404 forking, if "IO::AIO" was used in the parent. Calling it while
1405 IO::AIO is active in the process will result in undefined behaviour.
1406 Calling it at any time will also result in any undefined (by POSIX)
1407 behaviour.
1324 1408
1325 MEMORY USAGE 1409 MEMORY USAGE
1326 Per-request usage: 1410 Per-request usage:
1327 1411
1328 Each aio request uses - depending on your architecture - around 100-200 1412 Each aio request uses - depending on your architecture - around 100-200

Diff Legend

– Removed lines
+ Added lines
< Changed lines
> Changed lines