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

Comparing IO-AIO/AIO.pm (file contents):
Revision 1.193 by root, Thu May 26 04:15:37 2011 UTC vs.
Revision 1.197 by root, Wed Jun 29 10:49:35 2011 UTC

… …
168use common::sense; 168use common::sense;
169 169
170use base 'Exporter'; 170use base 'Exporter';
171 171
172BEGIN { 172BEGIN {
173 our $VERSION = '3.8'; 173 our $VERSION = '3.9';
174 174
175 our @AIO_REQ = qw(aio_sendfile aio_read aio_write aio_open aio_close 175 our @AIO_REQ = qw(aio_sendfile aio_read aio_write aio_open aio_close
176 aio_stat aio_lstat aio_unlink aio_rmdir aio_readdir aio_readdirx 176 aio_stat aio_lstat aio_unlink aio_rmdir aio_readdir aio_readdirx
177 aio_scandir aio_symlink aio_readlink aio_sync aio_fsync 177 aio_scandir aio_symlink aio_readlink aio_sync aio_fsync
178 aio_fdatasync aio_sync_file_range aio_pathsync aio_readahead 178 aio_fdatasync aio_sync_file_range aio_pathsync aio_readahead
… …
368 } else { 368 } else {
369 die "open failed: $!\n"; 369 die "open failed: $!\n";
370 } 370 }
371 }; 371 };
372 372
373In addition to all the common open modes/flags (C<O_RDONLY>, C<O_WRONLY>,
374C<O_RDWR>, C<O_CREAT>, C<O_TRUNC>, C<O_EXCL> and C<O_APPEND>), the
375following POSIX and non-POSIX constants are available (missing ones on
376your system are, as usual, C<0>):
377
378C<O_ASYNC>, C<O_DIRECT>, C<O_NOATIME>, C<O_CLOEXEC>, C<O_NOCTTY>, C<O_NOFOLLOW>,
379C<O_NONBLOCK>, C<O_EXEC>, C<O_SEARCH>, C<O_DIRECTORY>, C<O_DSYNC>,
380C<O_RSYNC>, C<O_SYNC> and C<O_TTY_INIT>.
381
373 382
374=item aio_close $fh, $callback->($status) 383=item aio_close $fh, $callback->($status)
375 384
376Asynchronously close a file and call the callback with the result 385Asynchronously close a file and call the callback with the result
377code. 386code.
… …
427 436
428Tries to copy C<$length> bytes from C<$in_fh> to C<$out_fh>. It starts 437Tries to copy C<$length> bytes from C<$in_fh> to C<$out_fh>. It starts
429reading at byte offset C<$in_offset>, and starts writing at the current 438reading at byte offset C<$in_offset>, and starts writing at the current
430file offset of C<$out_fh>. Because of that, it is not safe to issue more 439file offset of C<$out_fh>. Because of that, it is not safe to issue more
431than one C<aio_sendfile> per C<$out_fh>, as they will interfere with each 440than one C<aio_sendfile> per C<$out_fh>, as they will interfere with each
432other. 441other. The same C<$in_fh> works fine though, as this function does not
442move or use the file offset of C<$in_fh>.
433 443
434Please note that C<aio_sendfile> can read more bytes from C<$in_fh> than 444Please note that C<aio_sendfile> can read more bytes from C<$in_fh> than
435are written, and there is no way to find out how many bytes have been read 445are written, and there is no way to find out how many more bytes have been
436from C<aio_sendfile> alone, as C<aio_sendfile> only provides the number of 446read from C<aio_sendfile> alone, as C<aio_sendfile> only provides the
437bytes written to C<$out_fh>. Only if the result value equals C<$length> 447number of bytes written to C<$out_fh>. Only if the result value equals
438one can assume that C<$length> bytes have been read. 448C<$length> one can assume that C<$length> bytes have been read.
439 449
440Unlike with other C<aio_> functions, it makes a lot of sense to use 450Unlike with other C<aio_> functions, it makes a lot of sense to use
441C<aio_sendfile> on non-blocking sockets, as long as one end (typically 451C<aio_sendfile> on non-blocking sockets, as long as one end (typically
442the C<$in_fh>) is a file - the file I/O will then be asynchronous, while 452the C<$in_fh>) is a file - the file I/O will then be asynchronous, while
443the socket I/O will be non-blocking. Note, however, that you can run into 453the socket I/O will be non-blocking. Note, however, that you can run
444a trap where C<aio_sendfile> reads some data with readahead, then fails 454into a trap where C<aio_sendfile> reads some data with readahead, then
445to write all data, and when the socket is ready the next time, the data 455fails to write all data, and when the socket is ready the next time, the
446in the cache is already lost, forcing C<aio_sendfile> to again hit the 456data in the cache is already lost, forcing C<aio_sendfile> to again hit
447disk. Explicit C<aio_read> + C<aio_write> let's you control resource usage 457the disk. Explicit C<aio_read> + C<aio_write> let's you better control
448much better. 458resource usage.
449 459
450This call tries to make use of a native C<sendfile> syscall to provide 460This call tries to make use of a native C<sendfile>-like syscall to
451zero-copy operation. For this to work, C<$out_fh> should refer to a 461provide zero-copy operation. For this to work, C<$out_fh> should refer to
452socket, and C<$in_fh> should refer to an mmap'able file. 462a socket, and C<$in_fh> should refer to an mmap'able file.
453 463
454If a native sendfile cannot be found or it fails with C<ENOSYS>, 464If a native sendfile cannot be found or it fails with C<ENOSYS>,
455C<ENOTSUP>, C<EOPNOTSUPP>, C<EAFNOSUPPORT>, C<EPROTOTYPE> or C<ENOTSOCK>, 465C<EINVAL>, C<ENOTSUP>, C<EOPNOTSUPP>, C<EAFNOSUPPORT>, C<EPROTOTYPE> or
456it will be emulated, so you can call C<aio_sendfile> on any type of 466C<ENOTSOCK>, it will be emulated, so you can call C<aio_sendfile> on any
457filehandle regardless of the limitations of the operating system. 467type of filehandle regardless of the limitations of the operating system.
468
469As native sendfile syscalls (as practically any non-POSIX interface hacked
470together in a hurry to improve benchmark numbers) tend to be rather buggy
471on many systems, this implementation tries to work around some known bugs
472in Linux and FreeBSD kernels (probably others, too), but that might fail,
473so you really really should check the return value of C<aio_sendfile> -
474fewre bytes than expected might have been transferred.
458 475
459 476
460=item aio_readahead $fh,$offset,$length, $callback->($retval) 477=item aio_readahead $fh,$offset,$length, $callback->($retval)
461 478
462C<aio_readahead> populates the page cache with data from a file so that 479C<aio_readahead> populates the page cache with data from a file so that
… …
853 if ($_[0] && $! == EXDEV) { 870 if ($_[0] && $! == EXDEV) {
854 aioreq_pri $pri; 871 aioreq_pri $pri;
855 add $grp aio_copy $src, $dst, sub { 872 add $grp aio_copy $src, $dst, sub {
856 $grp->result ($_[0]); 873 $grp->result ($_[0]);
857 874
858 if (!$_[0]) { 875 unless ($_[0]) {
859 aioreq_pri $pri; 876 aioreq_pri $pri;
860 add $grp aio_unlink $src; 877 add $grp aio_unlink $src;
861 } 878 }
862 }; 879 };
863 } else { 880 } else {
… …
1555Sets the minimum idle timeout (default 10) after which worker threads are 1572Sets the minimum idle timeout (default 10) after which worker threads are
1556allowed to exit. SEe C<IO::AIO::max_idle>. 1573allowed to exit. SEe C<IO::AIO::max_idle>.
1557 1574
1558=item IO::AIO::max_outstanding $maxreqs 1575=item IO::AIO::max_outstanding $maxreqs
1559 1576
1577Sets the maximum number of outstanding requests to C<$nreqs>. If
1578you do queue up more than this number of requests, the next call to
1579C<IO::AIO::poll_cb> (and other functions calling C<poll_cb>, such as
1580C<IO::AIO::flush> or C<IO::AIO::poll>) will block until the limit is no
1581longer exceeded.
1582
1583In other words, this setting does not enforce a queue limit, but can be
1584used to make poll functions block if the limit is exceeded.
1585
1560This is a very bad function to use in interactive programs because it 1586This is a very bad function to use in interactive programs because it
1561blocks, and a bad way to reduce concurrency because it is inexact: Better 1587blocks, and a bad way to reduce concurrency because it is inexact: Better
1562use an C<aio_group> together with a feed callback. 1588use an C<aio_group> together with a feed callback.
1563 1589
1564Sets the maximum number of outstanding requests to C<$nreqs>. If you 1590It's main use is in scripts without an event loop - when you want to stat
1565do queue up more than this number of requests, the next call to the 1591a lot of files, you can write somehting like this:
1566C<poll_cb> (and C<poll_some> and other functions calling C<poll_cb>)
1567function will block until the limit is no longer exceeded.
1568 1592
1569The default value is very large, so there is no practical limit on the 1593 IO::AIO::max_outstanding 32;
1570number of outstanding requests.
1571 1594
1572You can still queue as many requests as you want. Therefore, 1595 for my $path (...) {
1573C<max_outstanding> is mainly useful in simple scripts (with low values) or 1596 aio_stat $path , ...;
1574as a stop gap to shield against fatal memory overflow (with large values). 1597 IO::AIO::poll_cb;
1598 }
1599
1600 IO::AIO::flush;
1601
1602The call to C<poll_cb> inside the loop will normally return instantly, but
1603as soon as more thna C<32> reqeusts are in-flight, it will block until
1604some requests have been handled. This keeps the loop from pushing a large
1605number of C<aio_stat> requests onto the queue.
1606
1607The default value for C<max_outstanding> is very large, so there is no
1608practical limit on the number of outstanding requests.
1575 1609
1576=back 1610=back
1577 1611
1578=head3 STATISTICAL INFORMATION 1612=head3 STATISTICAL INFORMATION
1579 1613
… …
1755 Danga::Socket->AddOtherFds (IO::AIO::poll_fileno => 1789 Danga::Socket->AddOtherFds (IO::AIO::poll_fileno =>
1756 \&IO::AIO::poll_cb); 1790 \&IO::AIO::poll_cb);
1757 1791
1758=head2 FORK BEHAVIOUR 1792=head2 FORK BEHAVIOUR
1759 1793
1760This module should do "the right thing" when the process using it forks: 1794Usage of pthreads in a program changes the semantics of fork
1795considerably. Specifically, only async-safe functions can be called after
1796fork. Perl doesn't know about this, so in general, you cannot call fork
1797with defined behaviour in perl. IO::AIO uses pthreads, so this applies,
1798but many other extensions and (for inexplicable reasons) perl itself often
1799is linked against pthreads, so this limitation applies.
1761 1800
1762Before the fork, IO::AIO enters a quiescent state where no requests 1801Some operating systems have extensions that allow safe use of fork, and
1763can be added in other threads and no results will be processed. After 1802this module should do "the right thing" on those, and tries on others. At
1764the fork the parent simply leaves the quiescent state and continues 1803the time of this writing (2011) only GNU/Linux supports these extensions
1765request/result processing, while the child frees the request/result queue 1804to POSIX.
1766(so that the requests started before the fork will only be handled in the
1767parent). Threads will be started on demand until the limit set in the
1768parent process has been reached again.
1769
1770In short: the parent will, after a short pause, continue as if fork had
1771not been called, while the child will act as if IO::AIO has not been used
1772yet.
1773 1805
1774=head2 MEMORY USAGE 1806=head2 MEMORY USAGE
1775 1807
1776Per-request usage: 1808Per-request usage:
1777 1809

Diff Legend

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