ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/libev/ev.pod
(Generate patch)

Comparing libev/ev.pod (file contents):
Revision 1.211 by root, Mon Nov 3 14:34:16 2008 UTC vs.
Revision 1.216 by root, Thu Nov 13 15:55:38 2008 UTC

417i.e. keep at least one watcher active per fd at all times. Stopping and 417i.e. keep at least one watcher active per fd at all times. Stopping and
418starting a watcher (without re-setting it) also usually doesn't cause 418starting a watcher (without re-setting it) also usually doesn't cause
419extra overhead. A fork can both result in spurious notifications as well 419extra overhead. A fork can both result in spurious notifications as well
420as in libev having to destroy and recreate the epoll object, which can 420as in libev having to destroy and recreate the epoll object, which can
421take considerable time and thus should be avoided. 421take considerable time and thus should be avoided.
422
423All this means that, in practice, C<EVBACKEND_SELECT> can be as fast or
424faster than epoll for maybe up to a hundred file descriptors, depending on
425the usage. So sad.
422 426
423While nominally embeddable in other event loops, this feature is broken in 427While nominally embeddable in other event loops, this feature is broken in
424all kernel versions tested so far. 428all kernel versions tested so far.
425 429
426This backend maps C<EV_READ> and C<EV_WRITE> in the same way as 430This backend maps C<EV_READ> and C<EV_WRITE> in the same way as
1415 else 1419 else
1416 { 1420 {
1417 // callback was invoked, but there was some activity, re-arm 1421 // callback was invoked, but there was some activity, re-arm
1418 // the watcher to fire in last_activity + 60, which is 1422 // the watcher to fire in last_activity + 60, which is
1419 // guaranteed to be in the future, so "again" is positive: 1423 // guaranteed to be in the future, so "again" is positive:
1420 w->again = timeout - now; 1424 w->repeat = timeout - now;
1421 ev_timer_again (EV_A_ w); 1425 ev_timer_again (EV_A_ w);
1422 } 1426 }
1423 } 1427 }
1424 1428
1425To summarise the callback: first calculate the real timeout (defined 1429To summarise the callback: first calculate the real timeout (defined
1997 2001
1998There is no support for kqueue, as apparently it cannot be used to 2002There is no support for kqueue, as apparently it cannot be used to
1999implement this functionality, due to the requirement of having a file 2003implement this functionality, due to the requirement of having a file
2000descriptor open on the object at all times, and detecting renames, unlinks 2004descriptor open on the object at all times, and detecting renames, unlinks
2001etc. is difficult. 2005etc. is difficult.
2006
2007=head3 C<stat ()> is a synchronous operation
2008
2009Libev doesn't normally do any kind of I/O itself, and so is not blocking
2010the process. The exception are C<ev_stat> watchers - those call C<stat
2011()>, which is a synchronous operation.
2012
2013For local paths, this usually doesn't matter: unless the system is very
2014busy or the intervals between stat's are large, a stat call will be fast,
2015as the path data is suually in memory already (except when starting the
2016watcher).
2017
2018For networked file systems, calling C<stat ()> can block an indefinite
2019time due to network issues, and even under good conditions, a stat call
2020often takes multiple milliseconds.
2021
2022Therefore, it is best to avoid using C<ev_stat> watchers on networked
2023paths, although this is fully supported by libev.
2002 2024
2003=head3 The special problem of stat time resolution 2025=head3 The special problem of stat time resolution
2004 2026
2005The C<stat ()> system call only supports full-second resolution portably, 2027The C<stat ()> system call only supports full-second resolution portably,
2006and even on systems where the resolution is higher, most file systems 2028and even on systems where the resolution is higher, most file systems

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines