1 |
=head1 NAME |
2 |
|
3 |
AnyEvent - the DBI of event loop programming |
4 |
|
5 |
EV, Event, Glib, Tk, Perl, Event::Lib, Irssi, IO::Async, Qt and POE are |
6 |
various supported event loops/environments. |
7 |
|
8 |
=head1 SYNOPSIS |
9 |
|
10 |
use AnyEvent; |
11 |
|
12 |
# file descriptor readable |
13 |
my $w = AnyEvent->io (fh => $fh, poll => "r", cb => sub { ... }); |
14 |
|
15 |
# one-shot or repeating timers |
16 |
my $w = AnyEvent->timer (after => $seconds, cb => sub { ... }); |
17 |
my $w = AnyEvent->timer (after => $seconds, interval => $seconds, cb => ... |
18 |
|
19 |
print AnyEvent->now; # prints current event loop time |
20 |
print AnyEvent->time; # think Time::HiRes::time or simply CORE::time. |
21 |
|
22 |
# POSIX signal |
23 |
my $w = AnyEvent->signal (signal => "TERM", cb => sub { ... }); |
24 |
|
25 |
# child process exit |
26 |
my $w = AnyEvent->child (pid => $pid, cb => sub { |
27 |
my ($pid, $status) = @_; |
28 |
... |
29 |
}); |
30 |
|
31 |
# called when event loop idle (if applicable) |
32 |
my $w = AnyEvent->idle (cb => sub { ... }); |
33 |
|
34 |
my $w = AnyEvent->condvar; # stores whether a condition was flagged |
35 |
$w->send; # wake up current and all future recv's |
36 |
$w->recv; # enters "main loop" till $condvar gets ->send |
37 |
# use a condvar in callback mode: |
38 |
$w->cb (sub { $_[0]->recv }); |
39 |
|
40 |
=head1 INTRODUCTION/TUTORIAL |
41 |
|
42 |
This manpage is mainly a reference manual. If you are interested |
43 |
in a tutorial or some gentle introduction, have a look at the |
44 |
L<AnyEvent::Intro> manpage. |
45 |
|
46 |
=head1 SUPPORT |
47 |
|
48 |
There is a mailinglist for discussing all things AnyEvent, and an IRC |
49 |
channel, too. |
50 |
|
51 |
See the AnyEvent project page at the B<Schmorpforge Ta-Sa Software |
52 |
Repository>, at L<http://anyevent.schmorp.de>, for more info. |
53 |
|
54 |
=head1 WHY YOU SHOULD USE THIS MODULE (OR NOT) |
55 |
|
56 |
Glib, POE, IO::Async, Event... CPAN offers event models by the dozen |
57 |
nowadays. So what is different about AnyEvent? |
58 |
|
59 |
Executive Summary: AnyEvent is I<compatible>, AnyEvent is I<free of |
60 |
policy> and AnyEvent is I<small and efficient>. |
61 |
|
62 |
First and foremost, I<AnyEvent is not an event model> itself, it only |
63 |
interfaces to whatever event model the main program happens to use, in a |
64 |
pragmatic way. For event models and certain classes of immortals alike, |
65 |
the statement "there can only be one" is a bitter reality: In general, |
66 |
only one event loop can be active at the same time in a process. AnyEvent |
67 |
cannot change this, but it can hide the differences between those event |
68 |
loops. |
69 |
|
70 |
The goal of AnyEvent is to offer module authors the ability to do event |
71 |
programming (waiting for I/O or timer events) without subscribing to a |
72 |
religion, a way of living, and most importantly: without forcing your |
73 |
module users into the same thing by forcing them to use the same event |
74 |
model you use. |
75 |
|
76 |
For modules like POE or IO::Async (which is a total misnomer as it is |
77 |
actually doing all I/O I<synchronously>...), using them in your module is |
78 |
like joining a cult: After you joined, you are dependent on them and you |
79 |
cannot use anything else, as they are simply incompatible to everything |
80 |
that isn't them. What's worse, all the potential users of your |
81 |
module are I<also> forced to use the same event loop you use. |
82 |
|
83 |
AnyEvent is different: AnyEvent + POE works fine. AnyEvent + Glib works |
84 |
fine. AnyEvent + Tk works fine etc. etc. but none of these work together |
85 |
with the rest: POE + IO::Async? No go. Tk + Event? No go. Again: if |
86 |
your module uses one of those, every user of your module has to use it, |
87 |
too. But if your module uses AnyEvent, it works transparently with all |
88 |
event models it supports (including stuff like IO::Async, as long as those |
89 |
use one of the supported event loops. It is trivial to add new event loops |
90 |
to AnyEvent, too, so it is future-proof). |
91 |
|
92 |
In addition to being free of having to use I<the one and only true event |
93 |
model>, AnyEvent also is free of bloat and policy: with POE or similar |
94 |
modules, you get an enormous amount of code and strict rules you have to |
95 |
follow. AnyEvent, on the other hand, is lean and up to the point, by only |
96 |
offering the functionality that is necessary, in as thin as a wrapper as |
97 |
technically possible. |
98 |
|
99 |
Of course, AnyEvent comes with a big (and fully optional!) toolbox |
100 |
of useful functionality, such as an asynchronous DNS resolver, 100% |
101 |
non-blocking connects (even with TLS/SSL, IPv6 and on broken platforms |
102 |
such as Windows) and lots of real-world knowledge and workarounds for |
103 |
platform bugs and differences. |
104 |
|
105 |
Now, if you I<do want> lots of policy (this can arguably be somewhat |
106 |
useful) and you want to force your users to use the one and only event |
107 |
model, you should I<not> use this module. |
108 |
|
109 |
=head1 DESCRIPTION |
110 |
|
111 |
L<AnyEvent> provides an identical interface to multiple event loops. This |
112 |
allows module authors to utilise an event loop without forcing module |
113 |
users to use the same event loop (as only a single event loop can coexist |
114 |
peacefully at any one time). |
115 |
|
116 |
The interface itself is vaguely similar, but not identical to the L<Event> |
117 |
module. |
118 |
|
119 |
During the first call of any watcher-creation method, the module tries |
120 |
to detect the currently loaded event loop by probing whether one of the |
121 |
following modules is already loaded: L<EV>, |
122 |
L<Event>, L<Glib>, L<AnyEvent::Impl::Perl>, L<Tk>, L<Event::Lib>, L<Qt>, |
123 |
L<POE>. The first one found is used. If none are found, the module tries |
124 |
to load these modules (excluding Tk, Event::Lib, Qt and POE as the pure perl |
125 |
adaptor should always succeed) in the order given. The first one that can |
126 |
be successfully loaded will be used. If, after this, still none could be |
127 |
found, AnyEvent will fall back to a pure-perl event loop, which is not |
128 |
very efficient, but should work everywhere. |
129 |
|
130 |
Because AnyEvent first checks for modules that are already loaded, loading |
131 |
an event model explicitly before first using AnyEvent will likely make |
132 |
that model the default. For example: |
133 |
|
134 |
use Tk; |
135 |
use AnyEvent; |
136 |
|
137 |
# .. AnyEvent will likely default to Tk |
138 |
|
139 |
The I<likely> means that, if any module loads another event model and |
140 |
starts using it, all bets are off. Maybe you should tell their authors to |
141 |
use AnyEvent so their modules work together with others seamlessly... |
142 |
|
143 |
The pure-perl implementation of AnyEvent is called |
144 |
C<AnyEvent::Impl::Perl>. Like other event modules you can load it |
145 |
explicitly and enjoy the high availability of that event loop :) |
146 |
|
147 |
=head1 WATCHERS |
148 |
|
149 |
AnyEvent has the central concept of a I<watcher>, which is an object that |
150 |
stores relevant data for each kind of event you are waiting for, such as |
151 |
the callback to call, the file handle to watch, etc. |
152 |
|
153 |
These watchers are normal Perl objects with normal Perl lifetime. After |
154 |
creating a watcher it will immediately "watch" for events and invoke the |
155 |
callback when the event occurs (of course, only when the event model |
156 |
is in control). |
157 |
|
158 |
Note that B<callbacks must not permanently change global variables> |
159 |
potentially in use by the event loop (such as C<$_> or C<$[>) and that B<< |
160 |
callbacks must not C<die> >>. The former is good programming practise in |
161 |
Perl and the latter stems from the fact that exception handling differs |
162 |
widely between event loops. |
163 |
|
164 |
To disable the watcher you have to destroy it (e.g. by setting the |
165 |
variable you store it in to C<undef> or otherwise deleting all references |
166 |
to it). |
167 |
|
168 |
All watchers are created by calling a method on the C<AnyEvent> class. |
169 |
|
170 |
Many watchers either are used with "recursion" (repeating timers for |
171 |
example), or need to refer to their watcher object in other ways. |
172 |
|
173 |
An any way to achieve that is this pattern: |
174 |
|
175 |
my $w; $w = AnyEvent->type (arg => value ..., cb => sub { |
176 |
# you can use $w here, for example to undef it |
177 |
undef $w; |
178 |
}); |
179 |
|
180 |
Note that C<my $w; $w => combination. This is necessary because in Perl, |
181 |
my variables are only visible after the statement in which they are |
182 |
declared. |
183 |
|
184 |
=head2 I/O WATCHERS |
185 |
|
186 |
You can create an I/O watcher by calling the C<< AnyEvent->io >> method |
187 |
with the following mandatory key-value pairs as arguments: |
188 |
|
189 |
C<fh> is the Perl I<file handle> (or a naked file descriptor) to watch |
190 |
for events (AnyEvent might or might not keep a reference to this file |
191 |
handle). Note that only file handles pointing to things for which |
192 |
non-blocking operation makes sense are allowed. This includes sockets, |
193 |
most character devices, pipes, fifos and so on, but not for example files |
194 |
or block devices. |
195 |
|
196 |
C<poll> must be a string that is either C<r> or C<w>, which creates a |
197 |
watcher waiting for "r"eadable or "w"ritable events, respectively. |
198 |
|
199 |
C<cb> is the callback to invoke each time the file handle becomes ready. |
200 |
|
201 |
Although the callback might get passed parameters, their value and |
202 |
presence is undefined and you cannot rely on them. Portable AnyEvent |
203 |
callbacks cannot use arguments passed to I/O watcher callbacks. |
204 |
|
205 |
The I/O watcher might use the underlying file descriptor or a copy of it. |
206 |
You must not close a file handle as long as any watcher is active on the |
207 |
underlying file descriptor. |
208 |
|
209 |
Some event loops issue spurious readyness notifications, so you should |
210 |
always use non-blocking calls when reading/writing from/to your file |
211 |
handles. |
212 |
|
213 |
Example: wait for readability of STDIN, then read a line and disable the |
214 |
watcher. |
215 |
|
216 |
my $w; $w = AnyEvent->io (fh => \*STDIN, poll => 'r', cb => sub { |
217 |
chomp (my $input = <STDIN>); |
218 |
warn "read: $input\n"; |
219 |
undef $w; |
220 |
}); |
221 |
|
222 |
=head2 TIME WATCHERS |
223 |
|
224 |
You can create a time watcher by calling the C<< AnyEvent->timer >> |
225 |
method with the following mandatory arguments: |
226 |
|
227 |
C<after> specifies after how many seconds (fractional values are |
228 |
supported) the callback should be invoked. C<cb> is the callback to invoke |
229 |
in that case. |
230 |
|
231 |
Although the callback might get passed parameters, their value and |
232 |
presence is undefined and you cannot rely on them. Portable AnyEvent |
233 |
callbacks cannot use arguments passed to time watcher callbacks. |
234 |
|
235 |
The callback will normally be invoked once only. If you specify another |
236 |
parameter, C<interval>, as a strictly positive number (> 0), then the |
237 |
callback will be invoked regularly at that interval (in fractional |
238 |
seconds) after the first invocation. If C<interval> is specified with a |
239 |
false value, then it is treated as if it were missing. |
240 |
|
241 |
The callback will be rescheduled before invoking the callback, but no |
242 |
attempt is done to avoid timer drift in most backends, so the interval is |
243 |
only approximate. |
244 |
|
245 |
Example: fire an event after 7.7 seconds. |
246 |
|
247 |
my $w = AnyEvent->timer (after => 7.7, cb => sub { |
248 |
warn "timeout\n"; |
249 |
}); |
250 |
|
251 |
# to cancel the timer: |
252 |
undef $w; |
253 |
|
254 |
Example 2: fire an event after 0.5 seconds, then roughly every second. |
255 |
|
256 |
my $w = AnyEvent->timer (after => 0.5, interval => 1, cb => sub { |
257 |
warn "timeout\n"; |
258 |
}; |
259 |
|
260 |
=head3 TIMING ISSUES |
261 |
|
262 |
There are two ways to handle timers: based on real time (relative, "fire |
263 |
in 10 seconds") and based on wallclock time (absolute, "fire at 12 |
264 |
o'clock"). |
265 |
|
266 |
While most event loops expect timers to specified in a relative way, they |
267 |
use absolute time internally. This makes a difference when your clock |
268 |
"jumps", for example, when ntp decides to set your clock backwards from |
269 |
the wrong date of 2014-01-01 to 2008-01-01, a watcher that is supposed to |
270 |
fire "after" a second might actually take six years to finally fire. |
271 |
|
272 |
AnyEvent cannot compensate for this. The only event loop that is conscious |
273 |
about these issues is L<EV>, which offers both relative (ev_timer, based |
274 |
on true relative time) and absolute (ev_periodic, based on wallclock time) |
275 |
timers. |
276 |
|
277 |
AnyEvent always prefers relative timers, if available, matching the |
278 |
AnyEvent API. |
279 |
|
280 |
AnyEvent has two additional methods that return the "current time": |
281 |
|
282 |
=over 4 |
283 |
|
284 |
=item AnyEvent->time |
285 |
|
286 |
This returns the "current wallclock time" as a fractional number of |
287 |
seconds since the Epoch (the same thing as C<time> or C<Time::HiRes::time> |
288 |
return, and the result is guaranteed to be compatible with those). |
289 |
|
290 |
It progresses independently of any event loop processing, i.e. each call |
291 |
will check the system clock, which usually gets updated frequently. |
292 |
|
293 |
=item AnyEvent->now |
294 |
|
295 |
This also returns the "current wallclock time", but unlike C<time>, above, |
296 |
this value might change only once per event loop iteration, depending on |
297 |
the event loop (most return the same time as C<time>, above). This is the |
298 |
time that AnyEvent's timers get scheduled against. |
299 |
|
300 |
I<In almost all cases (in all cases if you don't care), this is the |
301 |
function to call when you want to know the current time.> |
302 |
|
303 |
This function is also often faster then C<< AnyEvent->time >>, and |
304 |
thus the preferred method if you want some timestamp (for example, |
305 |
L<AnyEvent::Handle> uses this to update it's activity timeouts). |
306 |
|
307 |
The rest of this section is only of relevance if you try to be very exact |
308 |
with your timing, you can skip it without bad conscience. |
309 |
|
310 |
For a practical example of when these times differ, consider L<Event::Lib> |
311 |
and L<EV> and the following set-up: |
312 |
|
313 |
The event loop is running and has just invoked one of your callback at |
314 |
time=500 (assume no other callbacks delay processing). In your callback, |
315 |
you wait a second by executing C<sleep 1> (blocking the process for a |
316 |
second) and then (at time=501) you create a relative timer that fires |
317 |
after three seconds. |
318 |
|
319 |
With L<Event::Lib>, C<< AnyEvent->time >> and C<< AnyEvent->now >> will |
320 |
both return C<501>, because that is the current time, and the timer will |
321 |
be scheduled to fire at time=504 (C<501> + C<3>). |
322 |
|
323 |
With L<EV>, C<< AnyEvent->time >> returns C<501> (as that is the current |
324 |
time), but C<< AnyEvent->now >> returns C<500>, as that is the time the |
325 |
last event processing phase started. With L<EV>, your timer gets scheduled |
326 |
to run at time=503 (C<500> + C<3>). |
327 |
|
328 |
In one sense, L<Event::Lib> is more exact, as it uses the current time |
329 |
regardless of any delays introduced by event processing. However, most |
330 |
callbacks do not expect large delays in processing, so this causes a |
331 |
higher drift (and a lot more system calls to get the current time). |
332 |
|
333 |
In another sense, L<EV> is more exact, as your timer will be scheduled at |
334 |
the same time, regardless of how long event processing actually took. |
335 |
|
336 |
In either case, if you care (and in most cases, you don't), then you |
337 |
can get whatever behaviour you want with any event loop, by taking the |
338 |
difference between C<< AnyEvent->time >> and C<< AnyEvent->now >> into |
339 |
account. |
340 |
|
341 |
=item AnyEvent->now_update |
342 |
|
343 |
Some event loops (such as L<EV> or L<AnyEvent::Impl::Perl>) cache |
344 |
the current time for each loop iteration (see the discussion of L<< |
345 |
AnyEvent->now >>, above). |
346 |
|
347 |
When a callback runs for a long time (or when the process sleeps), then |
348 |
this "current" time will differ substantially from the real time, which |
349 |
might affect timers and time-outs. |
350 |
|
351 |
When this is the case, you can call this method, which will update the |
352 |
event loop's idea of "current time". |
353 |
|
354 |
Note that updating the time I<might> cause some events to be handled. |
355 |
|
356 |
=back |
357 |
|
358 |
=head2 SIGNAL WATCHERS |
359 |
|
360 |
You can watch for signals using a signal watcher, C<signal> is the signal |
361 |
I<name> in uppercase and without any C<SIG> prefix, C<cb> is the Perl |
362 |
callback to be invoked whenever a signal occurs. |
363 |
|
364 |
Although the callback might get passed parameters, their value and |
365 |
presence is undefined and you cannot rely on them. Portable AnyEvent |
366 |
callbacks cannot use arguments passed to signal watcher callbacks. |
367 |
|
368 |
Multiple signal occurrences can be clumped together into one callback |
369 |
invocation, and callback invocation will be synchronous. Synchronous means |
370 |
that it might take a while until the signal gets handled by the process, |
371 |
but it is guaranteed not to interrupt any other callbacks. |
372 |
|
373 |
The main advantage of using these watchers is that you can share a signal |
374 |
between multiple watchers, and AnyEvent will ensure that signals will not |
375 |
interrupt your program at bad times. |
376 |
|
377 |
This watcher might use C<%SIG> (depending on the event loop used), |
378 |
so programs overwriting those signals directly will likely not work |
379 |
correctly. |
380 |
|
381 |
Example: exit on SIGINT |
382 |
|
383 |
my $w = AnyEvent->signal (signal => "INT", cb => sub { exit 1 }); |
384 |
|
385 |
=head3 Signal Races, Delays and Workarounds |
386 |
|
387 |
Many event loops (e.g. Glib, Tk, Qt, IO::Async) do not support attaching |
388 |
callbacks to signals in a generic way, which is a pity, as you cannot do |
389 |
race-free signal handling in perl. AnyEvent will try to do it's best, but |
390 |
in some cases, signals will be delayed. The maximum time a signal might |
391 |
be delayed is specified in C<$AnyEvent::MAX_SIGNAL_LATENCY> (default: 10 |
392 |
seconds). This variable can be changed only before the first signal |
393 |
watcher is created, and should be left alone otherwise. Higher values |
394 |
will cause fewer spurious wake-ups, which is better for power and CPU |
395 |
saving. All these problems can be avoided by installing the optional |
396 |
L<Async::Interrupt> module. This will not work with inherently broken |
397 |
event loops such as L<Event> or L<Event::Lib> (and not with L<POE> |
398 |
currently, as POE does it's own workaround with one-second latency). With |
399 |
those, you just have to suffer the delays. |
400 |
|
401 |
=head2 CHILD PROCESS WATCHERS |
402 |
|
403 |
You can also watch on a child process exit and catch its exit status. |
404 |
|
405 |
The child process is specified by the C<pid> argument (one some backends, |
406 |
using C<0> watches for any child process exit, on others this will |
407 |
croak). The watcher will be triggered only when the child process has |
408 |
finished and an exit status is available, not on any trace events |
409 |
(stopped/continued). |
410 |
|
411 |
The callback will be called with the pid and exit status (as returned by |
412 |
waitpid), so unlike other watcher types, you I<can> rely on child watcher |
413 |
callback arguments. |
414 |
|
415 |
This watcher type works by installing a signal handler for C<SIGCHLD>, |
416 |
and since it cannot be shared, nothing else should use SIGCHLD or reap |
417 |
random child processes (waiting for specific child processes, e.g. inside |
418 |
C<system>, is just fine). |
419 |
|
420 |
There is a slight catch to child watchers, however: you usually start them |
421 |
I<after> the child process was created, and this means the process could |
422 |
have exited already (and no SIGCHLD will be sent anymore). |
423 |
|
424 |
Not all event models handle this correctly (neither POE nor IO::Async do, |
425 |
see their AnyEvent::Impl manpages for details), but even for event models |
426 |
that I<do> handle this correctly, they usually need to be loaded before |
427 |
the process exits (i.e. before you fork in the first place). AnyEvent's |
428 |
pure perl event loop handles all cases correctly regardless of when you |
429 |
start the watcher. |
430 |
|
431 |
This means you cannot create a child watcher as the very first |
432 |
thing in an AnyEvent program, you I<have> to create at least one |
433 |
watcher before you C<fork> the child (alternatively, you can call |
434 |
C<AnyEvent::detect>). |
435 |
|
436 |
As most event loops do not support waiting for child events, they will be |
437 |
emulated by AnyEvent in most cases, in which the latency and race problems |
438 |
mentioned in the description of signal watchers apply. |
439 |
|
440 |
Example: fork a process and wait for it |
441 |
|
442 |
my $done = AnyEvent->condvar; |
443 |
|
444 |
my $pid = fork or exit 5; |
445 |
|
446 |
my $w = AnyEvent->child ( |
447 |
pid => $pid, |
448 |
cb => sub { |
449 |
my ($pid, $status) = @_; |
450 |
warn "pid $pid exited with status $status"; |
451 |
$done->send; |
452 |
}, |
453 |
); |
454 |
|
455 |
# do something else, then wait for process exit |
456 |
$done->recv; |
457 |
|
458 |
=head2 IDLE WATCHERS |
459 |
|
460 |
Sometimes there is a need to do something, but it is not so important |
461 |
to do it instantly, but only when there is nothing better to do. This |
462 |
"nothing better to do" is usually defined to be "no other events need |
463 |
attention by the event loop". |
464 |
|
465 |
Idle watchers ideally get invoked when the event loop has nothing |
466 |
better to do, just before it would block the process to wait for new |
467 |
events. Instead of blocking, the idle watcher is invoked. |
468 |
|
469 |
Most event loops unfortunately do not really support idle watchers (only |
470 |
EV, Event and Glib do it in a usable fashion) - for the rest, AnyEvent |
471 |
will simply call the callback "from time to time". |
472 |
|
473 |
Example: read lines from STDIN, but only process them when the |
474 |
program is otherwise idle: |
475 |
|
476 |
my @lines; # read data |
477 |
my $idle_w; |
478 |
my $io_w = AnyEvent->io (fh => \*STDIN, poll => 'r', cb => sub { |
479 |
push @lines, scalar <STDIN>; |
480 |
|
481 |
# start an idle watcher, if not already done |
482 |
$idle_w ||= AnyEvent->idle (cb => sub { |
483 |
# handle only one line, when there are lines left |
484 |
if (my $line = shift @lines) { |
485 |
print "handled when idle: $line"; |
486 |
} else { |
487 |
# otherwise disable the idle watcher again |
488 |
undef $idle_w; |
489 |
} |
490 |
}); |
491 |
}); |
492 |
|
493 |
=head2 CONDITION VARIABLES |
494 |
|
495 |
If you are familiar with some event loops you will know that all of them |
496 |
require you to run some blocking "loop", "run" or similar function that |
497 |
will actively watch for new events and call your callbacks. |
498 |
|
499 |
AnyEvent is slightly different: it expects somebody else to run the event |
500 |
loop and will only block when necessary (usually when told by the user). |
501 |
|
502 |
The instrument to do that is called a "condition variable", so called |
503 |
because they represent a condition that must become true. |
504 |
|
505 |
Now is probably a good time to look at the examples further below. |
506 |
|
507 |
Condition variables can be created by calling the C<< AnyEvent->condvar |
508 |
>> method, usually without arguments. The only argument pair allowed is |
509 |
C<cb>, which specifies a callback to be called when the condition variable |
510 |
becomes true, with the condition variable as the first argument (but not |
511 |
the results). |
512 |
|
513 |
After creation, the condition variable is "false" until it becomes "true" |
514 |
by calling the C<send> method (or calling the condition variable as if it |
515 |
were a callback, read about the caveats in the description for the C<< |
516 |
->send >> method). |
517 |
|
518 |
Condition variables are similar to callbacks, except that you can |
519 |
optionally wait for them. They can also be called merge points - points |
520 |
in time where multiple outstanding events have been processed. And yet |
521 |
another way to call them is transactions - each condition variable can be |
522 |
used to represent a transaction, which finishes at some point and delivers |
523 |
a result. And yet some people know them as "futures" - a promise to |
524 |
compute/deliver something that you can wait for. |
525 |
|
526 |
Condition variables are very useful to signal that something has finished, |
527 |
for example, if you write a module that does asynchronous http requests, |
528 |
then a condition variable would be the ideal candidate to signal the |
529 |
availability of results. The user can either act when the callback is |
530 |
called or can synchronously C<< ->recv >> for the results. |
531 |
|
532 |
You can also use them to simulate traditional event loops - for example, |
533 |
you can block your main program until an event occurs - for example, you |
534 |
could C<< ->recv >> in your main program until the user clicks the Quit |
535 |
button of your app, which would C<< ->send >> the "quit" event. |
536 |
|
537 |
Note that condition variables recurse into the event loop - if you have |
538 |
two pieces of code that call C<< ->recv >> in a round-robin fashion, you |
539 |
lose. Therefore, condition variables are good to export to your caller, but |
540 |
you should avoid making a blocking wait yourself, at least in callbacks, |
541 |
as this asks for trouble. |
542 |
|
543 |
Condition variables are represented by hash refs in perl, and the keys |
544 |
used by AnyEvent itself are all named C<_ae_XXX> to make subclassing |
545 |
easy (it is often useful to build your own transaction class on top of |
546 |
AnyEvent). To subclass, use C<AnyEvent::CondVar> as base class and call |
547 |
it's C<new> method in your own C<new> method. |
548 |
|
549 |
There are two "sides" to a condition variable - the "producer side" which |
550 |
eventually calls C<< -> send >>, and the "consumer side", which waits |
551 |
for the send to occur. |
552 |
|
553 |
Example: wait for a timer. |
554 |
|
555 |
# wait till the result is ready |
556 |
my $result_ready = AnyEvent->condvar; |
557 |
|
558 |
# do something such as adding a timer |
559 |
# or socket watcher the calls $result_ready->send |
560 |
# when the "result" is ready. |
561 |
# in this case, we simply use a timer: |
562 |
my $w = AnyEvent->timer ( |
563 |
after => 1, |
564 |
cb => sub { $result_ready->send }, |
565 |
); |
566 |
|
567 |
# this "blocks" (while handling events) till the callback |
568 |
# calls -<send |
569 |
$result_ready->recv; |
570 |
|
571 |
Example: wait for a timer, but take advantage of the fact that condition |
572 |
variables are also callable directly. |
573 |
|
574 |
my $done = AnyEvent->condvar; |
575 |
my $delay = AnyEvent->timer (after => 5, cb => $done); |
576 |
$done->recv; |
577 |
|
578 |
Example: Imagine an API that returns a condvar and doesn't support |
579 |
callbacks. This is how you make a synchronous call, for example from |
580 |
the main program: |
581 |
|
582 |
use AnyEvent::CouchDB; |
583 |
|
584 |
... |
585 |
|
586 |
my @info = $couchdb->info->recv; |
587 |
|
588 |
And this is how you would just set a callback to be called whenever the |
589 |
results are available: |
590 |
|
591 |
$couchdb->info->cb (sub { |
592 |
my @info = $_[0]->recv; |
593 |
}); |
594 |
|
595 |
=head3 METHODS FOR PRODUCERS |
596 |
|
597 |
These methods should only be used by the producing side, i.e. the |
598 |
code/module that eventually sends the signal. Note that it is also |
599 |
the producer side which creates the condvar in most cases, but it isn't |
600 |
uncommon for the consumer to create it as well. |
601 |
|
602 |
=over 4 |
603 |
|
604 |
=item $cv->send (...) |
605 |
|
606 |
Flag the condition as ready - a running C<< ->recv >> and all further |
607 |
calls to C<recv> will (eventually) return after this method has been |
608 |
called. If nobody is waiting the send will be remembered. |
609 |
|
610 |
If a callback has been set on the condition variable, it is called |
611 |
immediately from within send. |
612 |
|
613 |
Any arguments passed to the C<send> call will be returned by all |
614 |
future C<< ->recv >> calls. |
615 |
|
616 |
Condition variables are overloaded so one can call them directly (as if |
617 |
they were a code reference). Calling them directly is the same as calling |
618 |
C<send>. |
619 |
|
620 |
=item $cv->croak ($error) |
621 |
|
622 |
Similar to send, but causes all call's to C<< ->recv >> to invoke |
623 |
C<Carp::croak> with the given error message/object/scalar. |
624 |
|
625 |
This can be used to signal any errors to the condition variable |
626 |
user/consumer. Doing it this way instead of calling C<croak> directly |
627 |
delays the error detetcion, but has the overwhelmign advantage that it |
628 |
diagnoses the error at the place where the result is expected, and not |
629 |
deep in some event clalback without connection to the actual code causing |
630 |
the problem. |
631 |
|
632 |
=item $cv->begin ([group callback]) |
633 |
|
634 |
=item $cv->end |
635 |
|
636 |
These two methods can be used to combine many transactions/events into |
637 |
one. For example, a function that pings many hosts in parallel might want |
638 |
to use a condition variable for the whole process. |
639 |
|
640 |
Every call to C<< ->begin >> will increment a counter, and every call to |
641 |
C<< ->end >> will decrement it. If the counter reaches C<0> in C<< ->end |
642 |
>>, the (last) callback passed to C<begin> will be executed. That callback |
643 |
is I<supposed> to call C<< ->send >>, but that is not required. If no |
644 |
callback was set, C<send> will be called without any arguments. |
645 |
|
646 |
You can think of C<< $cv->send >> giving you an OR condition (one call |
647 |
sends), while C<< $cv->begin >> and C<< $cv->end >> giving you an AND |
648 |
condition (all C<begin> calls must be C<end>'ed before the condvar sends). |
649 |
|
650 |
Let's start with a simple example: you have two I/O watchers (for example, |
651 |
STDOUT and STDERR for a program), and you want to wait for both streams to |
652 |
close before activating a condvar: |
653 |
|
654 |
my $cv = AnyEvent->condvar; |
655 |
|
656 |
$cv->begin; # first watcher |
657 |
my $w1 = AnyEvent->io (fh => $fh1, cb => sub { |
658 |
defined sysread $fh1, my $buf, 4096 |
659 |
or $cv->end; |
660 |
}); |
661 |
|
662 |
$cv->begin; # second watcher |
663 |
my $w2 = AnyEvent->io (fh => $fh2, cb => sub { |
664 |
defined sysread $fh2, my $buf, 4096 |
665 |
or $cv->end; |
666 |
}); |
667 |
|
668 |
$cv->recv; |
669 |
|
670 |
This works because for every event source (EOF on file handle), there is |
671 |
one call to C<begin>, so the condvar waits for all calls to C<end> before |
672 |
sending. |
673 |
|
674 |
The ping example mentioned above is slightly more complicated, as the |
675 |
there are results to be passwd back, and the number of tasks that are |
676 |
begung can potentially be zero: |
677 |
|
678 |
my $cv = AnyEvent->condvar; |
679 |
|
680 |
my %result; |
681 |
$cv->begin (sub { $cv->send (\%result) }); |
682 |
|
683 |
for my $host (@list_of_hosts) { |
684 |
$cv->begin; |
685 |
ping_host_then_call_callback $host, sub { |
686 |
$result{$host} = ...; |
687 |
$cv->end; |
688 |
}; |
689 |
} |
690 |
|
691 |
$cv->end; |
692 |
|
693 |
This code fragment supposedly pings a number of hosts and calls |
694 |
C<send> after results for all then have have been gathered - in any |
695 |
order. To achieve this, the code issues a call to C<begin> when it starts |
696 |
each ping request and calls C<end> when it has received some result for |
697 |
it. Since C<begin> and C<end> only maintain a counter, the order in which |
698 |
results arrive is not relevant. |
699 |
|
700 |
There is an additional bracketing call to C<begin> and C<end> outside the |
701 |
loop, which serves two important purposes: first, it sets the callback |
702 |
to be called once the counter reaches C<0>, and second, it ensures that |
703 |
C<send> is called even when C<no> hosts are being pinged (the loop |
704 |
doesn't execute once). |
705 |
|
706 |
This is the general pattern when you "fan out" into multiple (but |
707 |
potentially none) subrequests: use an outer C<begin>/C<end> pair to set |
708 |
the callback and ensure C<end> is called at least once, and then, for each |
709 |
subrequest you start, call C<begin> and for each subrequest you finish, |
710 |
call C<end>. |
711 |
|
712 |
=back |
713 |
|
714 |
=head3 METHODS FOR CONSUMERS |
715 |
|
716 |
These methods should only be used by the consuming side, i.e. the |
717 |
code awaits the condition. |
718 |
|
719 |
=over 4 |
720 |
|
721 |
=item $cv->recv |
722 |
|
723 |
Wait (blocking if necessary) until the C<< ->send >> or C<< ->croak |
724 |
>> methods have been called on c<$cv>, while servicing other watchers |
725 |
normally. |
726 |
|
727 |
You can only wait once on a condition - additional calls are valid but |
728 |
will return immediately. |
729 |
|
730 |
If an error condition has been set by calling C<< ->croak >>, then this |
731 |
function will call C<croak>. |
732 |
|
733 |
In list context, all parameters passed to C<send> will be returned, |
734 |
in scalar context only the first one will be returned. |
735 |
|
736 |
Note that doing a blocking wait in a callback is not supported by any |
737 |
event loop, that is, recursive invocation of a blocking C<< ->recv |
738 |
>> is not allowed, and the C<recv> call will C<croak> if such a |
739 |
condition is detected. This condition can be slightly loosened by using |
740 |
L<Coro::AnyEvent>, which allows you to do a blocking C<< ->recv >> from |
741 |
any thread that doesn't run the event loop itself. |
742 |
|
743 |
Not all event models support a blocking wait - some die in that case |
744 |
(programs might want to do that to stay interactive), so I<if you are |
745 |
using this from a module, never require a blocking wait>. Instead, let the |
746 |
caller decide whether the call will block or not (for example, by coupling |
747 |
condition variables with some kind of request results and supporting |
748 |
callbacks so the caller knows that getting the result will not block, |
749 |
while still supporting blocking waits if the caller so desires). |
750 |
|
751 |
You can ensure that C<< -recv >> never blocks by setting a callback and |
752 |
only calling C<< ->recv >> from within that callback (or at a later |
753 |
time). This will work even when the event loop does not support blocking |
754 |
waits otherwise. |
755 |
|
756 |
=item $bool = $cv->ready |
757 |
|
758 |
Returns true when the condition is "true", i.e. whether C<send> or |
759 |
C<croak> have been called. |
760 |
|
761 |
=item $cb = $cv->cb ($cb->($cv)) |
762 |
|
763 |
This is a mutator function that returns the callback set and optionally |
764 |
replaces it before doing so. |
765 |
|
766 |
The callback will be called when the condition becomes "true", i.e. when |
767 |
C<send> or C<croak> are called, with the only argument being the condition |
768 |
variable itself. Calling C<recv> inside the callback or at any later time |
769 |
is guaranteed not to block. |
770 |
|
771 |
=back |
772 |
|
773 |
=head1 SUPPORTED EVENT LOOPS/BACKENDS |
774 |
|
775 |
The available backend classes are (every class has its own manpage): |
776 |
|
777 |
=over 4 |
778 |
|
779 |
=item Backends that are autoprobed when no other event loop can be found. |
780 |
|
781 |
EV is the preferred backend when no other event loop seems to be in |
782 |
use. If EV is not installed, then AnyEvent will try Event, and, failing |
783 |
that, will fall back to its own pure-perl implementation, which is |
784 |
available everywhere as it comes with AnyEvent itself. |
785 |
|
786 |
AnyEvent::Impl::EV based on EV (interface to libev, best choice). |
787 |
AnyEvent::Impl::Event based on Event, very stable, few glitches. |
788 |
AnyEvent::Impl::Perl pure-perl implementation, fast and portable. |
789 |
|
790 |
=item Backends that are transparently being picked up when they are used. |
791 |
|
792 |
These will be used when they are currently loaded when the first watcher |
793 |
is created, in which case it is assumed that the application is using |
794 |
them. This means that AnyEvent will automatically pick the right backend |
795 |
when the main program loads an event module before anything starts to |
796 |
create watchers. Nothing special needs to be done by the main program. |
797 |
|
798 |
AnyEvent::Impl::Glib based on Glib, slow but very stable. |
799 |
AnyEvent::Impl::Tk based on Tk, very broken. |
800 |
AnyEvent::Impl::EventLib based on Event::Lib, leaks memory and worse. |
801 |
AnyEvent::Impl::POE based on POE, very slow, some limitations. |
802 |
AnyEvent::Impl::Irssi used when running within irssi. |
803 |
|
804 |
=item Backends with special needs. |
805 |
|
806 |
Qt requires the Qt::Application to be instantiated first, but will |
807 |
otherwise be picked up automatically. As long as the main program |
808 |
instantiates the application before any AnyEvent watchers are created, |
809 |
everything should just work. |
810 |
|
811 |
AnyEvent::Impl::Qt based on Qt. |
812 |
|
813 |
Support for IO::Async can only be partial, as it is too broken and |
814 |
architecturally limited to even support the AnyEvent API. It also |
815 |
is the only event loop that needs the loop to be set explicitly, so |
816 |
it can only be used by a main program knowing about AnyEvent. See |
817 |
L<AnyEvent::Impl::Async> for the gory details. |
818 |
|
819 |
AnyEvent::Impl::IOAsync based on IO::Async, cannot be autoprobed. |
820 |
|
821 |
=item Event loops that are indirectly supported via other backends. |
822 |
|
823 |
Some event loops can be supported via other modules: |
824 |
|
825 |
There is no direct support for WxWidgets (L<Wx>) or L<Prima>. |
826 |
|
827 |
B<WxWidgets> has no support for watching file handles. However, you can |
828 |
use WxWidgets through the POE adaptor, as POE has a Wx backend that simply |
829 |
polls 20 times per second, which was considered to be too horrible to even |
830 |
consider for AnyEvent. |
831 |
|
832 |
B<Prima> is not supported as nobody seems to be using it, but it has a POE |
833 |
backend, so it can be supported through POE. |
834 |
|
835 |
AnyEvent knows about both L<Prima> and L<Wx>, however, and will try to |
836 |
load L<POE> when detecting them, in the hope that POE will pick them up, |
837 |
in which case everything will be automatic. |
838 |
|
839 |
=back |
840 |
|
841 |
=head1 GLOBAL VARIABLES AND FUNCTIONS |
842 |
|
843 |
These are not normally required to use AnyEvent, but can be useful to |
844 |
write AnyEvent extension modules. |
845 |
|
846 |
=over 4 |
847 |
|
848 |
=item $AnyEvent::MODEL |
849 |
|
850 |
Contains C<undef> until the first watcher is being created, before the |
851 |
backend has been autodetected. |
852 |
|
853 |
Afterwards it contains the event model that is being used, which is the |
854 |
name of the Perl class implementing the model. This class is usually one |
855 |
of the C<AnyEvent::Impl:xxx> modules, but can be any other class in the |
856 |
case AnyEvent has been extended at runtime (e.g. in I<rxvt-unicode> it |
857 |
will be C<urxvt::anyevent>). |
858 |
|
859 |
=item AnyEvent::detect |
860 |
|
861 |
Returns C<$AnyEvent::MODEL>, forcing autodetection of the event model |
862 |
if necessary. You should only call this function right before you would |
863 |
have created an AnyEvent watcher anyway, that is, as late as possible at |
864 |
runtime, and not e.g. while initialising of your module. |
865 |
|
866 |
If you need to do some initialisation before AnyEvent watchers are |
867 |
created, use C<post_detect>. |
868 |
|
869 |
=item $guard = AnyEvent::post_detect { BLOCK } |
870 |
|
871 |
Arranges for the code block to be executed as soon as the event model is |
872 |
autodetected (or immediately if this has already happened). |
873 |
|
874 |
The block will be executed I<after> the actual backend has been detected |
875 |
(C<$AnyEvent::MODEL> is set), but I<before> any watchers have been |
876 |
created, so it is possible to e.g. patch C<@AnyEvent::ISA> or do |
877 |
other initialisations - see the sources of L<AnyEvent::Strict> or |
878 |
L<AnyEvent::AIO> to see how this is used. |
879 |
|
880 |
The most common usage is to create some global watchers, without forcing |
881 |
event module detection too early, for example, L<AnyEvent::AIO> creates |
882 |
and installs the global L<IO::AIO> watcher in a C<post_detect> block to |
883 |
avoid autodetecting the event module at load time. |
884 |
|
885 |
If called in scalar or list context, then it creates and returns an object |
886 |
that automatically removes the callback again when it is destroyed (or |
887 |
C<undef> when the hook was immediately executed). See L<AnyEvent::AIO> for |
888 |
a case where this is useful. |
889 |
|
890 |
Example: Create a watcher for the IO::AIO module and store it in |
891 |
C<$WATCHER>. Only do so after the event loop is initialised, though. |
892 |
|
893 |
our WATCHER; |
894 |
|
895 |
my $guard = AnyEvent::post_detect { |
896 |
$WATCHER = AnyEvent->io (fh => IO::AIO::poll_fileno, poll => 'r', cb => \&IO::AIO::poll_cb); |
897 |
}; |
898 |
|
899 |
# the ||= is important in case post_detect immediately runs the block, |
900 |
# as to not clobber the newly-created watcher. assigning both watcher and |
901 |
# post_detect guard to the same variable has the advantage of users being |
902 |
# able to just C<undef $WATCHER> if the watcher causes them grief. |
903 |
|
904 |
$WATCHER ||= $guard; |
905 |
|
906 |
=item @AnyEvent::post_detect |
907 |
|
908 |
If there are any code references in this array (you can C<push> to it |
909 |
before or after loading AnyEvent), then they will called directly after |
910 |
the event loop has been chosen. |
911 |
|
912 |
You should check C<$AnyEvent::MODEL> before adding to this array, though: |
913 |
if it is defined then the event loop has already been detected, and the |
914 |
array will be ignored. |
915 |
|
916 |
Best use C<AnyEvent::post_detect { BLOCK }> when your application allows |
917 |
it,as it takes care of these details. |
918 |
|
919 |
This variable is mainly useful for modules that can do something useful |
920 |
when AnyEvent is used and thus want to know when it is initialised, but do |
921 |
not need to even load it by default. This array provides the means to hook |
922 |
into AnyEvent passively, without loading it. |
923 |
|
924 |
=back |
925 |
|
926 |
=head1 WHAT TO DO IN A MODULE |
927 |
|
928 |
As a module author, you should C<use AnyEvent> and call AnyEvent methods |
929 |
freely, but you should not load a specific event module or rely on it. |
930 |
|
931 |
Be careful when you create watchers in the module body - AnyEvent will |
932 |
decide which event module to use as soon as the first method is called, so |
933 |
by calling AnyEvent in your module body you force the user of your module |
934 |
to load the event module first. |
935 |
|
936 |
Never call C<< ->recv >> on a condition variable unless you I<know> that |
937 |
the C<< ->send >> method has been called on it already. This is |
938 |
because it will stall the whole program, and the whole point of using |
939 |
events is to stay interactive. |
940 |
|
941 |
It is fine, however, to call C<< ->recv >> when the user of your module |
942 |
requests it (i.e. if you create a http request object ad have a method |
943 |
called C<results> that returns the results, it should call C<< ->recv >> |
944 |
freely, as the user of your module knows what she is doing. always). |
945 |
|
946 |
=head1 WHAT TO DO IN THE MAIN PROGRAM |
947 |
|
948 |
There will always be a single main program - the only place that should |
949 |
dictate which event model to use. |
950 |
|
951 |
If it doesn't care, it can just "use AnyEvent" and use it itself, or not |
952 |
do anything special (it does not need to be event-based) and let AnyEvent |
953 |
decide which implementation to chose if some module relies on it. |
954 |
|
955 |
If the main program relies on a specific event model - for example, in |
956 |
Gtk2 programs you have to rely on the Glib module - you should load the |
957 |
event module before loading AnyEvent or any module that uses it: generally |
958 |
speaking, you should load it as early as possible. The reason is that |
959 |
modules might create watchers when they are loaded, and AnyEvent will |
960 |
decide on the event model to use as soon as it creates watchers, and it |
961 |
might chose the wrong one unless you load the correct one yourself. |
962 |
|
963 |
You can chose to use a pure-perl implementation by loading the |
964 |
C<AnyEvent::Impl::Perl> module, which gives you similar behaviour |
965 |
everywhere, but letting AnyEvent chose the model is generally better. |
966 |
|
967 |
=head2 MAINLOOP EMULATION |
968 |
|
969 |
Sometimes (often for short test scripts, or even standalone programs who |
970 |
only want to use AnyEvent), you do not want to run a specific event loop. |
971 |
|
972 |
In that case, you can use a condition variable like this: |
973 |
|
974 |
AnyEvent->condvar->recv; |
975 |
|
976 |
This has the effect of entering the event loop and looping forever. |
977 |
|
978 |
Note that usually your program has some exit condition, in which case |
979 |
it is better to use the "traditional" approach of storing a condition |
980 |
variable somewhere, waiting for it, and sending it when the program should |
981 |
exit cleanly. |
982 |
|
983 |
|
984 |
=head1 OTHER MODULES |
985 |
|
986 |
The following is a non-exhaustive list of additional modules that use |
987 |
AnyEvent as a client and can therefore be mixed easily with other AnyEvent |
988 |
modules and other event loops in the same program. Some of the modules |
989 |
come with AnyEvent, most are available via CPAN. |
990 |
|
991 |
=over 4 |
992 |
|
993 |
=item L<AnyEvent::Util> |
994 |
|
995 |
Contains various utility functions that replace often-used but blocking |
996 |
functions such as C<inet_aton> by event-/callback-based versions. |
997 |
|
998 |
=item L<AnyEvent::Socket> |
999 |
|
1000 |
Provides various utility functions for (internet protocol) sockets, |
1001 |
addresses and name resolution. Also functions to create non-blocking tcp |
1002 |
connections or tcp servers, with IPv6 and SRV record support and more. |
1003 |
|
1004 |
=item L<AnyEvent::Handle> |
1005 |
|
1006 |
Provide read and write buffers, manages watchers for reads and writes, |
1007 |
supports raw and formatted I/O, I/O queued and fully transparent and |
1008 |
non-blocking SSL/TLS (via L<AnyEvent::TLS>. |
1009 |
|
1010 |
=item L<AnyEvent::DNS> |
1011 |
|
1012 |
Provides rich asynchronous DNS resolver capabilities. |
1013 |
|
1014 |
=item L<AnyEvent::HTTP> |
1015 |
|
1016 |
A simple-to-use HTTP library that is capable of making a lot of concurrent |
1017 |
HTTP requests. |
1018 |
|
1019 |
=item L<AnyEvent::HTTPD> |
1020 |
|
1021 |
Provides a simple web application server framework. |
1022 |
|
1023 |
=item L<AnyEvent::FastPing> |
1024 |
|
1025 |
The fastest ping in the west. |
1026 |
|
1027 |
=item L<AnyEvent::DBI> |
1028 |
|
1029 |
Executes L<DBI> requests asynchronously in a proxy process. |
1030 |
|
1031 |
=item L<AnyEvent::AIO> |
1032 |
|
1033 |
Truly asynchronous I/O, should be in the toolbox of every event |
1034 |
programmer. AnyEvent::AIO transparently fuses L<IO::AIO> and AnyEvent |
1035 |
together. |
1036 |
|
1037 |
=item L<AnyEvent::BDB> |
1038 |
|
1039 |
Truly asynchronous Berkeley DB access. AnyEvent::BDB transparently fuses |
1040 |
L<BDB> and AnyEvent together. |
1041 |
|
1042 |
=item L<AnyEvent::GPSD> |
1043 |
|
1044 |
A non-blocking interface to gpsd, a daemon delivering GPS information. |
1045 |
|
1046 |
=item L<AnyEvent::IRC> |
1047 |
|
1048 |
AnyEvent based IRC client module family (replacing the older Net::IRC3). |
1049 |
|
1050 |
=item L<AnyEvent::XMPP> |
1051 |
|
1052 |
AnyEvent based XMPP (Jabber protocol) module family (replacing the older |
1053 |
Net::XMPP2>. |
1054 |
|
1055 |
=item L<AnyEvent::IGS> |
1056 |
|
1057 |
A non-blocking interface to the Internet Go Server protocol (used by |
1058 |
L<App::IGS>). |
1059 |
|
1060 |
=item L<Net::FCP> |
1061 |
|
1062 |
AnyEvent-based implementation of the Freenet Client Protocol, birthplace |
1063 |
of AnyEvent. |
1064 |
|
1065 |
=item L<Event::ExecFlow> |
1066 |
|
1067 |
High level API for event-based execution flow control. |
1068 |
|
1069 |
=item L<Coro> |
1070 |
|
1071 |
Has special support for AnyEvent via L<Coro::AnyEvent>. |
1072 |
|
1073 |
=back |
1074 |
|
1075 |
=cut |
1076 |
|
1077 |
package AnyEvent; |
1078 |
|
1079 |
# basically a tuned-down version of common::sense |
1080 |
sub common_sense { |
1081 |
# no warnings |
1082 |
${^WARNING_BITS} ^= ${^WARNING_BITS}; |
1083 |
# use strict vars subs |
1084 |
$^H |= 0x00000600; |
1085 |
} |
1086 |
|
1087 |
BEGIN { AnyEvent::common_sense } |
1088 |
|
1089 |
use Carp (); |
1090 |
|
1091 |
our $VERSION = 4.86; |
1092 |
our $MODEL; |
1093 |
|
1094 |
our $AUTOLOAD; |
1095 |
our @ISA; |
1096 |
|
1097 |
our @REGISTRY; |
1098 |
|
1099 |
our $WIN32; |
1100 |
|
1101 |
our $VERBOSE; |
1102 |
|
1103 |
BEGIN { |
1104 |
eval "sub WIN32(){ " . (($^O =~ /mswin32/i)*1) ." }"; |
1105 |
eval "sub TAINT(){ " . (${^TAINT}*1) . " }"; |
1106 |
|
1107 |
delete @ENV{grep /^PERL_ANYEVENT_/, keys %ENV} |
1108 |
if ${^TAINT}; |
1109 |
|
1110 |
$VERBOSE = $ENV{PERL_ANYEVENT_VERBOSE}*1; |
1111 |
|
1112 |
} |
1113 |
|
1114 |
our $MAX_SIGNAL_LATENCY = 10; |
1115 |
|
1116 |
our %PROTOCOL; # (ipv4|ipv6) => (1|2), higher numbers are preferred |
1117 |
|
1118 |
{ |
1119 |
my $idx; |
1120 |
$PROTOCOL{$_} = ++$idx |
1121 |
for reverse split /\s*,\s*/, |
1122 |
$ENV{PERL_ANYEVENT_PROTOCOLS} || "ipv4,ipv6"; |
1123 |
} |
1124 |
|
1125 |
my @models = ( |
1126 |
[EV:: => AnyEvent::Impl::EV:: , 1], |
1127 |
[Event:: => AnyEvent::Impl::Event::, 1], |
1128 |
[AnyEvent::Impl::Perl:: => AnyEvent::Impl::Perl:: , 1], |
1129 |
# everything below here will not (normally) be autoprobed |
1130 |
# as the pureperl backend should work everywhere |
1131 |
# and is usually faster |
1132 |
[Glib:: => AnyEvent::Impl::Glib:: , 1], # becomes extremely slow with many watchers |
1133 |
[Event::Lib:: => AnyEvent::Impl::EventLib::], # too buggy |
1134 |
[Irssi:: => AnyEvent::Impl::Irssi::], # Irssi has a bogus "Event" package |
1135 |
[Tk:: => AnyEvent::Impl::Tk::], # crashes with many handles |
1136 |
[Qt:: => AnyEvent::Impl::Qt::], # requires special main program |
1137 |
[POE::Kernel:: => AnyEvent::Impl::POE::], # lasciate ogni speranza |
1138 |
[Wx:: => AnyEvent::Impl::POE::], |
1139 |
[Prima:: => AnyEvent::Impl::POE::], |
1140 |
# IO::Async is just too broken - we would need workarounds for its |
1141 |
# byzantine signal and broken child handling, among others. |
1142 |
# IO::Async is rather hard to detect, as it doesn't have any |
1143 |
# obvious default class. |
1144 |
# [0, IO::Async:: => AnyEvent::Impl::IOAsync::], # requires special main program |
1145 |
# [0, IO::Async::Loop:: => AnyEvent::Impl::IOAsync::], # requires special main program |
1146 |
# [0, IO::Async::Notifier:: => AnyEvent::Impl::IOAsync::], # requires special main program |
1147 |
); |
1148 |
|
1149 |
our %method = map +($_ => 1), |
1150 |
qw(io timer time now now_update signal child idle condvar one_event DESTROY); |
1151 |
|
1152 |
our @post_detect; |
1153 |
|
1154 |
sub post_detect(&) { |
1155 |
my ($cb) = @_; |
1156 |
|
1157 |
if ($MODEL) { |
1158 |
$cb->(); |
1159 |
|
1160 |
undef |
1161 |
} else { |
1162 |
push @post_detect, $cb; |
1163 |
|
1164 |
defined wantarray |
1165 |
? bless \$cb, "AnyEvent::Util::postdetect" |
1166 |
: () |
1167 |
} |
1168 |
} |
1169 |
|
1170 |
sub AnyEvent::Util::postdetect::DESTROY { |
1171 |
@post_detect = grep $_ != ${$_[0]}, @post_detect; |
1172 |
} |
1173 |
|
1174 |
sub detect() { |
1175 |
unless ($MODEL) { |
1176 |
local $SIG{__DIE__}; |
1177 |
|
1178 |
if ($ENV{PERL_ANYEVENT_MODEL} =~ /^([a-zA-Z]+)$/) { |
1179 |
my $model = "AnyEvent::Impl::$1"; |
1180 |
if (eval "require $model") { |
1181 |
$MODEL = $model; |
1182 |
warn "AnyEvent: loaded model '$model' (forced by \$ENV{PERL_ANYEVENT_MODEL}), using it.\n" if $VERBOSE >= 2; |
1183 |
} else { |
1184 |
warn "AnyEvent: unable to load model '$model' (from \$ENV{PERL_ANYEVENT_MODEL}):\n$@" if $VERBOSE; |
1185 |
} |
1186 |
} |
1187 |
|
1188 |
# check for already loaded models |
1189 |
unless ($MODEL) { |
1190 |
for (@REGISTRY, @models) { |
1191 |
my ($package, $model) = @$_; |
1192 |
if (${"$package\::VERSION"} > 0) { |
1193 |
if (eval "require $model") { |
1194 |
$MODEL = $model; |
1195 |
warn "AnyEvent: autodetected model '$model', using it.\n" if $VERBOSE >= 2; |
1196 |
last; |
1197 |
} |
1198 |
} |
1199 |
} |
1200 |
|
1201 |
unless ($MODEL) { |
1202 |
# try to autoload a model |
1203 |
for (@REGISTRY, @models) { |
1204 |
my ($package, $model, $autoload) = @$_; |
1205 |
if ( |
1206 |
$autoload |
1207 |
and eval "require $package" |
1208 |
and ${"$package\::VERSION"} > 0 |
1209 |
and eval "require $model" |
1210 |
) { |
1211 |
$MODEL = $model; |
1212 |
warn "AnyEvent: autoloaded model '$model', using it.\n" if $VERBOSE >= 2; |
1213 |
last; |
1214 |
} |
1215 |
} |
1216 |
|
1217 |
$MODEL |
1218 |
or die "No event module selected for AnyEvent and autodetect failed. Install any one of these modules: EV, Event or Glib.\n"; |
1219 |
} |
1220 |
} |
1221 |
|
1222 |
push @{"$MODEL\::ISA"}, "AnyEvent::Base"; |
1223 |
|
1224 |
unshift @ISA, $MODEL; |
1225 |
|
1226 |
require AnyEvent::Strict if $ENV{PERL_ANYEVENT_STRICT}; |
1227 |
|
1228 |
(shift @post_detect)->() while @post_detect; |
1229 |
} |
1230 |
|
1231 |
$MODEL |
1232 |
} |
1233 |
|
1234 |
sub AUTOLOAD { |
1235 |
(my $func = $AUTOLOAD) =~ s/.*://; |
1236 |
|
1237 |
$method{$func} |
1238 |
or Carp::croak "$func: not a valid method for AnyEvent objects"; |
1239 |
|
1240 |
detect unless $MODEL; |
1241 |
|
1242 |
my $class = shift; |
1243 |
$class->$func (@_); |
1244 |
} |
1245 |
|
1246 |
# utility function to dup a filehandle. this is used by many backends |
1247 |
# to support binding more than one watcher per filehandle (they usually |
1248 |
# allow only one watcher per fd, so we dup it to get a different one). |
1249 |
sub _dupfh($$;$$) { |
1250 |
my ($poll, $fh, $r, $w) = @_; |
1251 |
|
1252 |
# cygwin requires the fh mode to be matching, unix doesn't |
1253 |
my ($rw, $mode) = $poll eq "r" ? ($r, "<&") : ($w, ">&"); |
1254 |
|
1255 |
open my $fh2, $mode, $fh |
1256 |
or die "AnyEvent->io: cannot dup() filehandle in mode '$poll': $!,"; |
1257 |
|
1258 |
# we assume CLOEXEC is already set by perl in all important cases |
1259 |
|
1260 |
($fh2, $rw) |
1261 |
} |
1262 |
|
1263 |
package AnyEvent::Base; |
1264 |
|
1265 |
# default implementations for many methods |
1266 |
|
1267 |
sub _time { |
1268 |
# probe for availability of Time::HiRes |
1269 |
if (eval "use Time::HiRes (); Time::HiRes::time (); 1") { |
1270 |
warn "AnyEvent: using Time::HiRes for sub-second timing accuracy.\n" if $VERBOSE >= 8; |
1271 |
*_time = \&Time::HiRes::time; |
1272 |
# if (eval "use POSIX (); (POSIX::times())... |
1273 |
} else { |
1274 |
warn "AnyEvent: using built-in time(), WARNING, no sub-second resolution!\n" if $VERBOSE; |
1275 |
*_time = sub { time }; # epic fail |
1276 |
} |
1277 |
|
1278 |
&_time |
1279 |
} |
1280 |
|
1281 |
sub time { _time } |
1282 |
sub now { _time } |
1283 |
sub now_update { } |
1284 |
|
1285 |
# default implementation for ->condvar |
1286 |
|
1287 |
sub condvar { |
1288 |
bless { @_ == 3 ? (_ae_cb => $_[2]) : () }, "AnyEvent::CondVar" |
1289 |
} |
1290 |
|
1291 |
# default implementation for ->signal |
1292 |
|
1293 |
our $HAVE_ASYNC_INTERRUPT; |
1294 |
our ($SIGPIPE_R, $SIGPIPE_W, %SIG_CB, %SIG_EV, $SIG_IO); |
1295 |
our (%SIG_ASY, %SIG_ASY_W); |
1296 |
our ($SIG_COUNT, $SIG_TW); |
1297 |
|
1298 |
sub _signal_exec { |
1299 |
$HAVE_ASYNC_INTERRUPT |
1300 |
? $SIGPIPE_R->drain |
1301 |
: sysread $SIGPIPE_R, my $dummy, 9; |
1302 |
|
1303 |
while (%SIG_EV) { |
1304 |
for (keys %SIG_EV) { |
1305 |
delete $SIG_EV{$_}; |
1306 |
$_->() for values %{ $SIG_CB{$_} || {} }; |
1307 |
} |
1308 |
} |
1309 |
} |
1310 |
|
1311 |
# install a dumym wakeupw atcher to reduce signal catching latency |
1312 |
sub _sig_add() { |
1313 |
unless ($SIG_COUNT++) { |
1314 |
# try to align timer on a full-second boundary, if possible |
1315 |
my $NOW = AnyEvent->now; |
1316 |
|
1317 |
$SIG_TW = AnyEvent->timer ( |
1318 |
after => $MAX_SIGNAL_LATENCY - ($NOW - int $NOW), |
1319 |
interval => $MAX_SIGNAL_LATENCY, |
1320 |
cb => sub { }, # just for the PERL_ASYNC_CHECK |
1321 |
); |
1322 |
} |
1323 |
} |
1324 |
|
1325 |
sub _sig_del { |
1326 |
undef $SIG_TW |
1327 |
unless --$SIG_COUNT; |
1328 |
} |
1329 |
|
1330 |
sub _signal { |
1331 |
my (undef, %arg) = @_; |
1332 |
|
1333 |
my $signal = uc $arg{signal} |
1334 |
or Carp::croak "required option 'signal' is missing"; |
1335 |
|
1336 |
$SIG_CB{$signal}{$arg{cb}} = $arg{cb}; |
1337 |
|
1338 |
if ($HAVE_ASYNC_INTERRUPT) { |
1339 |
# async::interrupt |
1340 |
|
1341 |
$SIG_ASY{$signal} ||= do { |
1342 |
my $asy = new Async::Interrupt |
1343 |
cb => sub { undef $SIG_EV{$signal} }, |
1344 |
signal => $signal, |
1345 |
pipe => [$SIGPIPE_R->filenos], |
1346 |
; |
1347 |
$asy->pipe_autodrain (0); |
1348 |
|
1349 |
$asy |
1350 |
}; |
1351 |
|
1352 |
} else { |
1353 |
# pure perl |
1354 |
|
1355 |
$SIG{$signal} ||= sub { |
1356 |
local $!; |
1357 |
syswrite $SIGPIPE_W, "\x00", 1 unless %SIG_EV; |
1358 |
undef $SIG_EV{$signal}; |
1359 |
}; |
1360 |
|
1361 |
# can't do signal processing without introducing races in pure perl, |
1362 |
# so limit the signal latency. |
1363 |
_sig_add; |
1364 |
} |
1365 |
|
1366 |
bless [$signal, $arg{cb}], "AnyEvent::Base::signal" |
1367 |
} |
1368 |
|
1369 |
sub signal { |
1370 |
# probe for availability of Async::Interrupt |
1371 |
if (!$ENV{PERL_ANYEVENT_AVOID_ASYNC_INTERRUPT} && eval "use Async::Interrupt 0.6 (); 1") { |
1372 |
warn "AnyEvent: using Async::Interrupt for race-free signal handling.\n" if $VERBOSE >= 8; |
1373 |
|
1374 |
$HAVE_ASYNC_INTERRUPT = 1; |
1375 |
$SIGPIPE_R = new Async::Interrupt::EventPipe; |
1376 |
$SIG_IO = AnyEvent->io (fh => $SIGPIPE_R->fileno, poll => "r", cb => \&_signal_exec); |
1377 |
|
1378 |
} else { |
1379 |
warn "AnyEvent: using emulated perl signal handling with latency timer.\n" if $VERBOSE >= 8; |
1380 |
|
1381 |
require Fcntl; |
1382 |
|
1383 |
if (AnyEvent::WIN32) { |
1384 |
require AnyEvent::Util; |
1385 |
|
1386 |
($SIGPIPE_R, $SIGPIPE_W) = AnyEvent::Util::portable_pipe (); |
1387 |
AnyEvent::Util::fh_nonblocking ($SIGPIPE_R) if $SIGPIPE_R; |
1388 |
AnyEvent::Util::fh_nonblocking ($SIGPIPE_W) if $SIGPIPE_W; # just in case |
1389 |
} else { |
1390 |
pipe $SIGPIPE_R, $SIGPIPE_W; |
1391 |
fcntl $SIGPIPE_R, &Fcntl::F_SETFL, &Fcntl::O_NONBLOCK if $SIGPIPE_R; |
1392 |
fcntl $SIGPIPE_W, &Fcntl::F_SETFL, &Fcntl::O_NONBLOCK if $SIGPIPE_W; # just in case |
1393 |
|
1394 |
# not strictly required, as $^F is normally 2, but let's make sure... |
1395 |
fcntl $SIGPIPE_R, &Fcntl::F_SETFD, &Fcntl::FD_CLOEXEC; |
1396 |
fcntl $SIGPIPE_W, &Fcntl::F_SETFD, &Fcntl::FD_CLOEXEC; |
1397 |
} |
1398 |
|
1399 |
$SIGPIPE_R |
1400 |
or Carp::croak "AnyEvent: unable to create a signal reporting pipe: $!\n"; |
1401 |
|
1402 |
$SIG_IO = AnyEvent->io (fh => $SIGPIPE_R, poll => "r", cb => \&_signal_exec); |
1403 |
} |
1404 |
|
1405 |
*signal = \&_signal; |
1406 |
&signal |
1407 |
} |
1408 |
|
1409 |
sub AnyEvent::Base::signal::DESTROY { |
1410 |
my ($signal, $cb) = @{$_[0]}; |
1411 |
|
1412 |
_sig_del; |
1413 |
|
1414 |
delete $SIG_CB{$signal}{$cb}; |
1415 |
|
1416 |
$HAVE_ASYNC_INTERRUPT |
1417 |
? delete $SIG_ASY{$signal} |
1418 |
: # delete doesn't work with older perls - they then |
1419 |
# print weird messages, or just unconditionally exit |
1420 |
# instead of getting the default action. |
1421 |
undef $SIG{$signal} |
1422 |
unless keys %{ $SIG_CB{$signal} }; |
1423 |
} |
1424 |
|
1425 |
# default implementation for ->child |
1426 |
|
1427 |
our %PID_CB; |
1428 |
our $CHLD_W; |
1429 |
our $CHLD_DELAY_W; |
1430 |
our $WNOHANG; |
1431 |
|
1432 |
sub _emit_childstatus($$) { |
1433 |
my (undef, $rpid, $rstatus) = @_; |
1434 |
|
1435 |
$_->($rpid, $rstatus) |
1436 |
for values %{ $PID_CB{$rpid} || {} }, |
1437 |
values %{ $PID_CB{0} || {} }; |
1438 |
} |
1439 |
|
1440 |
sub _sigchld { |
1441 |
my $pid; |
1442 |
|
1443 |
AnyEvent->_emit_childstatus ($pid, $?) |
1444 |
while ($pid = waitpid -1, $WNOHANG) > 0; |
1445 |
} |
1446 |
|
1447 |
sub child { |
1448 |
my (undef, %arg) = @_; |
1449 |
|
1450 |
defined (my $pid = $arg{pid} + 0) |
1451 |
or Carp::croak "required option 'pid' is missing"; |
1452 |
|
1453 |
$PID_CB{$pid}{$arg{cb}} = $arg{cb}; |
1454 |
|
1455 |
# WNOHANG is almost cetrainly 1 everywhere |
1456 |
$WNOHANG ||= $^O =~ /^(?:openbsd|netbsd|linux|freebsd|cygwin|MSWin32)$/ |
1457 |
? 1 |
1458 |
: eval { local $SIG{__DIE__}; require POSIX; &POSIX::WNOHANG } || 1; |
1459 |
|
1460 |
unless ($CHLD_W) { |
1461 |
$CHLD_W = AnyEvent->signal (signal => 'CHLD', cb => \&_sigchld); |
1462 |
# child could be a zombie already, so make at least one round |
1463 |
&_sigchld; |
1464 |
} |
1465 |
|
1466 |
bless [$pid, $arg{cb}], "AnyEvent::Base::child" |
1467 |
} |
1468 |
|
1469 |
sub AnyEvent::Base::child::DESTROY { |
1470 |
my ($pid, $cb) = @{$_[0]}; |
1471 |
|
1472 |
delete $PID_CB{$pid}{$cb}; |
1473 |
delete $PID_CB{$pid} unless keys %{ $PID_CB{$pid} }; |
1474 |
|
1475 |
undef $CHLD_W unless keys %PID_CB; |
1476 |
} |
1477 |
|
1478 |
# idle emulation is done by simply using a timer, regardless |
1479 |
# of whether the process is idle or not, and not letting |
1480 |
# the callback use more than 50% of the time. |
1481 |
sub idle { |
1482 |
my (undef, %arg) = @_; |
1483 |
|
1484 |
my ($cb, $w, $rcb) = $arg{cb}; |
1485 |
|
1486 |
$rcb = sub { |
1487 |
if ($cb) { |
1488 |
$w = _time; |
1489 |
&$cb; |
1490 |
$w = _time - $w; |
1491 |
|
1492 |
# never use more then 50% of the time for the idle watcher, |
1493 |
# within some limits |
1494 |
$w = 0.0001 if $w < 0.0001; |
1495 |
$w = 5 if $w > 5; |
1496 |
|
1497 |
$w = AnyEvent->timer (after => $w, cb => $rcb); |
1498 |
} else { |
1499 |
# clean up... |
1500 |
undef $w; |
1501 |
undef $rcb; |
1502 |
} |
1503 |
}; |
1504 |
|
1505 |
$w = AnyEvent->timer (after => 0.05, cb => $rcb); |
1506 |
|
1507 |
bless \\$cb, "AnyEvent::Base::idle" |
1508 |
} |
1509 |
|
1510 |
sub AnyEvent::Base::idle::DESTROY { |
1511 |
undef $${$_[0]}; |
1512 |
} |
1513 |
|
1514 |
package AnyEvent::CondVar; |
1515 |
|
1516 |
our @ISA = AnyEvent::CondVar::Base::; |
1517 |
|
1518 |
package AnyEvent::CondVar::Base; |
1519 |
|
1520 |
#use overload |
1521 |
# '&{}' => sub { my $self = shift; sub { $self->send (@_) } }, |
1522 |
# fallback => 1; |
1523 |
|
1524 |
# save 300+ kilobytes by dirtily hardcoding overloading |
1525 |
${"AnyEvent::CondVar::Base::OVERLOAD"}{dummy}++; # Register with magic by touching. |
1526 |
*{'AnyEvent::CondVar::Base::()'} = sub { }; # "Make it findable via fetchmethod." |
1527 |
*{'AnyEvent::CondVar::Base::(&{}'} = sub { my $self = shift; sub { $self->send (@_) } }; # &{} |
1528 |
${'AnyEvent::CondVar::Base::()'} = 1; # fallback |
1529 |
|
1530 |
our $WAITING; |
1531 |
|
1532 |
sub _send { |
1533 |
# nop |
1534 |
} |
1535 |
|
1536 |
sub send { |
1537 |
my $cv = shift; |
1538 |
$cv->{_ae_sent} = [@_]; |
1539 |
(delete $cv->{_ae_cb})->($cv) if $cv->{_ae_cb}; |
1540 |
$cv->_send; |
1541 |
} |
1542 |
|
1543 |
sub croak { |
1544 |
$_[0]{_ae_croak} = $_[1]; |
1545 |
$_[0]->send; |
1546 |
} |
1547 |
|
1548 |
sub ready { |
1549 |
$_[0]{_ae_sent} |
1550 |
} |
1551 |
|
1552 |
sub _wait { |
1553 |
$WAITING |
1554 |
and !$_[0]{_ae_sent} |
1555 |
and Carp::croak "AnyEvent::CondVar: recursive blocking wait detected"; |
1556 |
|
1557 |
local $WAITING = 1; |
1558 |
AnyEvent->one_event while !$_[0]{_ae_sent}; |
1559 |
} |
1560 |
|
1561 |
sub recv { |
1562 |
$_[0]->_wait; |
1563 |
|
1564 |
Carp::croak $_[0]{_ae_croak} if $_[0]{_ae_croak}; |
1565 |
wantarray ? @{ $_[0]{_ae_sent} } : $_[0]{_ae_sent}[0] |
1566 |
} |
1567 |
|
1568 |
sub cb { |
1569 |
$_[0]{_ae_cb} = $_[1] if @_ > 1; |
1570 |
$_[0]{_ae_cb} |
1571 |
} |
1572 |
|
1573 |
sub begin { |
1574 |
++$_[0]{_ae_counter}; |
1575 |
$_[0]{_ae_end_cb} = $_[1] if @_ > 1; |
1576 |
} |
1577 |
|
1578 |
sub end { |
1579 |
return if --$_[0]{_ae_counter}; |
1580 |
&{ $_[0]{_ae_end_cb} || sub { $_[0]->send } }; |
1581 |
} |
1582 |
|
1583 |
# undocumented/compatibility with pre-3.4 |
1584 |
*broadcast = \&send; |
1585 |
*wait = \&_wait; |
1586 |
|
1587 |
=head1 ERROR AND EXCEPTION HANDLING |
1588 |
|
1589 |
In general, AnyEvent does not do any error handling - it relies on the |
1590 |
caller to do that if required. The L<AnyEvent::Strict> module (see also |
1591 |
the C<PERL_ANYEVENT_STRICT> environment variable, below) provides strict |
1592 |
checking of all AnyEvent methods, however, which is highly useful during |
1593 |
development. |
1594 |
|
1595 |
As for exception handling (i.e. runtime errors and exceptions thrown while |
1596 |
executing a callback), this is not only highly event-loop specific, but |
1597 |
also not in any way wrapped by this module, as this is the job of the main |
1598 |
program. |
1599 |
|
1600 |
The pure perl event loop simply re-throws the exception (usually |
1601 |
within C<< condvar->recv >>), the L<Event> and L<EV> modules call C<< |
1602 |
$Event/EV::DIED->() >>, L<Glib> uses C<< install_exception_handler >> and |
1603 |
so on. |
1604 |
|
1605 |
=head1 ENVIRONMENT VARIABLES |
1606 |
|
1607 |
The following environment variables are used by this module or its |
1608 |
submodules. |
1609 |
|
1610 |
Note that AnyEvent will remove I<all> environment variables starting with |
1611 |
C<PERL_ANYEVENT_> from C<%ENV> when it is loaded while taint mode is |
1612 |
enabled. |
1613 |
|
1614 |
=over 4 |
1615 |
|
1616 |
=item C<PERL_ANYEVENT_VERBOSE> |
1617 |
|
1618 |
By default, AnyEvent will be completely silent except in fatal |
1619 |
conditions. You can set this environment variable to make AnyEvent more |
1620 |
talkative. |
1621 |
|
1622 |
When set to C<1> or higher, causes AnyEvent to warn about unexpected |
1623 |
conditions, such as not being able to load the event model specified by |
1624 |
C<PERL_ANYEVENT_MODEL>. |
1625 |
|
1626 |
When set to C<2> or higher, cause AnyEvent to report to STDERR which event |
1627 |
model it chooses. |
1628 |
|
1629 |
When set to C<8> or higher, then AnyEvent will report extra information on |
1630 |
which optional modules it loads and how it implements certain features. |
1631 |
|
1632 |
=item C<PERL_ANYEVENT_STRICT> |
1633 |
|
1634 |
AnyEvent does not do much argument checking by default, as thorough |
1635 |
argument checking is very costly. Setting this variable to a true value |
1636 |
will cause AnyEvent to load C<AnyEvent::Strict> and then to thoroughly |
1637 |
check the arguments passed to most method calls. If it finds any problems, |
1638 |
it will croak. |
1639 |
|
1640 |
In other words, enables "strict" mode. |
1641 |
|
1642 |
Unlike C<use strict> (or it's modern cousin, C<< use L<common::sense> |
1643 |
>>, it is definitely recommended to keep it off in production. Keeping |
1644 |
C<PERL_ANYEVENT_STRICT=1> in your environment while developing programs |
1645 |
can be very useful, however. |
1646 |
|
1647 |
=item C<PERL_ANYEVENT_MODEL> |
1648 |
|
1649 |
This can be used to specify the event model to be used by AnyEvent, before |
1650 |
auto detection and -probing kicks in. It must be a string consisting |
1651 |
entirely of ASCII letters. The string C<AnyEvent::Impl::> gets prepended |
1652 |
and the resulting module name is loaded and if the load was successful, |
1653 |
used as event model. If it fails to load AnyEvent will proceed with |
1654 |
auto detection and -probing. |
1655 |
|
1656 |
This functionality might change in future versions. |
1657 |
|
1658 |
For example, to force the pure perl model (L<AnyEvent::Impl::Perl>) you |
1659 |
could start your program like this: |
1660 |
|
1661 |
PERL_ANYEVENT_MODEL=Perl perl ... |
1662 |
|
1663 |
=item C<PERL_ANYEVENT_PROTOCOLS> |
1664 |
|
1665 |
Used by both L<AnyEvent::DNS> and L<AnyEvent::Socket> to determine preferences |
1666 |
for IPv4 or IPv6. The default is unspecified (and might change, or be the result |
1667 |
of auto probing). |
1668 |
|
1669 |
Must be set to a comma-separated list of protocols or address families, |
1670 |
current supported: C<ipv4> and C<ipv6>. Only protocols mentioned will be |
1671 |
used, and preference will be given to protocols mentioned earlier in the |
1672 |
list. |
1673 |
|
1674 |
This variable can effectively be used for denial-of-service attacks |
1675 |
against local programs (e.g. when setuid), although the impact is likely |
1676 |
small, as the program has to handle conenction and other failures anyways. |
1677 |
|
1678 |
Examples: C<PERL_ANYEVENT_PROTOCOLS=ipv4,ipv6> - prefer IPv4 over IPv6, |
1679 |
but support both and try to use both. C<PERL_ANYEVENT_PROTOCOLS=ipv4> |
1680 |
- only support IPv4, never try to resolve or contact IPv6 |
1681 |
addresses. C<PERL_ANYEVENT_PROTOCOLS=ipv6,ipv4> support either IPv4 or |
1682 |
IPv6, but prefer IPv6 over IPv4. |
1683 |
|
1684 |
=item C<PERL_ANYEVENT_EDNS0> |
1685 |
|
1686 |
Used by L<AnyEvent::DNS> to decide whether to use the EDNS0 extension |
1687 |
for DNS. This extension is generally useful to reduce DNS traffic, but |
1688 |
some (broken) firewalls drop such DNS packets, which is why it is off by |
1689 |
default. |
1690 |
|
1691 |
Setting this variable to C<1> will cause L<AnyEvent::DNS> to announce |
1692 |
EDNS0 in its DNS requests. |
1693 |
|
1694 |
=item C<PERL_ANYEVENT_MAX_FORKS> |
1695 |
|
1696 |
The maximum number of child processes that C<AnyEvent::Util::fork_call> |
1697 |
will create in parallel. |
1698 |
|
1699 |
=item C<PERL_ANYEVENT_MAX_OUTSTANDING_DNS> |
1700 |
|
1701 |
The default value for the C<max_outstanding> parameter for the default DNS |
1702 |
resolver - this is the maximum number of parallel DNS requests that are |
1703 |
sent to the DNS server. |
1704 |
|
1705 |
=item C<PERL_ANYEVENT_RESOLV_CONF> |
1706 |
|
1707 |
The file to use instead of F</etc/resolv.conf> (or OS-specific |
1708 |
configuration) in the default resolver. When set to the empty string, no |
1709 |
default config will be used. |
1710 |
|
1711 |
=item C<PERL_ANYEVENT_CA_FILE>, C<PERL_ANYEVENT_CA_PATH>. |
1712 |
|
1713 |
When neither C<ca_file> nor C<ca_path> was specified during |
1714 |
L<AnyEvent::TLS> context creation, and either of these environment |
1715 |
variables exist, they will be used to specify CA certificate locations |
1716 |
instead of a system-dependent default. |
1717 |
|
1718 |
=item C<PERL_ANYEVENT_AVOID_GUARD> and C<PERL_ANYEVENT_AVOID_ASYNC_INTERRUPT> |
1719 |
|
1720 |
When these are set to C<1>, then the respective modules are not |
1721 |
loaded. Mostly good for testing AnyEvent itself. |
1722 |
|
1723 |
=back |
1724 |
|
1725 |
=head1 SUPPLYING YOUR OWN EVENT MODEL INTERFACE |
1726 |
|
1727 |
This is an advanced topic that you do not normally need to use AnyEvent in |
1728 |
a module. This section is only of use to event loop authors who want to |
1729 |
provide AnyEvent compatibility. |
1730 |
|
1731 |
If you need to support another event library which isn't directly |
1732 |
supported by AnyEvent, you can supply your own interface to it by |
1733 |
pushing, before the first watcher gets created, the package name of |
1734 |
the event module and the package name of the interface to use onto |
1735 |
C<@AnyEvent::REGISTRY>. You can do that before and even without loading |
1736 |
AnyEvent, so it is reasonably cheap. |
1737 |
|
1738 |
Example: |
1739 |
|
1740 |
push @AnyEvent::REGISTRY, [urxvt => urxvt::anyevent::]; |
1741 |
|
1742 |
This tells AnyEvent to (literally) use the C<urxvt::anyevent::> |
1743 |
package/class when it finds the C<urxvt> package/module is already loaded. |
1744 |
|
1745 |
When AnyEvent is loaded and asked to find a suitable event model, it |
1746 |
will first check for the presence of urxvt by trying to C<use> the |
1747 |
C<urxvt::anyevent> module. |
1748 |
|
1749 |
The class should provide implementations for all watcher types. See |
1750 |
L<AnyEvent::Impl::EV> (source code), L<AnyEvent::Impl::Glib> (Source code) |
1751 |
and so on for actual examples. Use C<perldoc -m AnyEvent::Impl::Glib> to |
1752 |
see the sources. |
1753 |
|
1754 |
If you don't provide C<signal> and C<child> watchers than AnyEvent will |
1755 |
provide suitable (hopefully) replacements. |
1756 |
|
1757 |
The above example isn't fictitious, the I<rxvt-unicode> (a.k.a. urxvt) |
1758 |
terminal emulator uses the above line as-is. An interface isn't included |
1759 |
in AnyEvent because it doesn't make sense outside the embedded interpreter |
1760 |
inside I<rxvt-unicode>, and it is updated and maintained as part of the |
1761 |
I<rxvt-unicode> distribution. |
1762 |
|
1763 |
I<rxvt-unicode> also cheats a bit by not providing blocking access to |
1764 |
condition variables: code blocking while waiting for a condition will |
1765 |
C<die>. This still works with most modules/usages, and blocking calls must |
1766 |
not be done in an interactive application, so it makes sense. |
1767 |
|
1768 |
=head1 EXAMPLE PROGRAM |
1769 |
|
1770 |
The following program uses an I/O watcher to read data from STDIN, a timer |
1771 |
to display a message once per second, and a condition variable to quit the |
1772 |
program when the user enters quit: |
1773 |
|
1774 |
use AnyEvent; |
1775 |
|
1776 |
my $cv = AnyEvent->condvar; |
1777 |
|
1778 |
my $io_watcher = AnyEvent->io ( |
1779 |
fh => \*STDIN, |
1780 |
poll => 'r', |
1781 |
cb => sub { |
1782 |
warn "io event <$_[0]>\n"; # will always output <r> |
1783 |
chomp (my $input = <STDIN>); # read a line |
1784 |
warn "read: $input\n"; # output what has been read |
1785 |
$cv->send if $input =~ /^q/i; # quit program if /^q/i |
1786 |
}, |
1787 |
); |
1788 |
|
1789 |
my $time_watcher; # can only be used once |
1790 |
|
1791 |
sub new_timer { |
1792 |
$timer = AnyEvent->timer (after => 1, cb => sub { |
1793 |
warn "timeout\n"; # print 'timeout' about every second |
1794 |
&new_timer; # and restart the time |
1795 |
}); |
1796 |
} |
1797 |
|
1798 |
new_timer; # create first timer |
1799 |
|
1800 |
$cv->recv; # wait until user enters /^q/i |
1801 |
|
1802 |
=head1 REAL-WORLD EXAMPLE |
1803 |
|
1804 |
Consider the L<Net::FCP> module. It features (among others) the following |
1805 |
API calls, which are to freenet what HTTP GET requests are to http: |
1806 |
|
1807 |
my $data = $fcp->client_get ($url); # blocks |
1808 |
|
1809 |
my $transaction = $fcp->txn_client_get ($url); # does not block |
1810 |
$transaction->cb ( sub { ... } ); # set optional result callback |
1811 |
my $data = $transaction->result; # possibly blocks |
1812 |
|
1813 |
The C<client_get> method works like C<LWP::Simple::get>: it requests the |
1814 |
given URL and waits till the data has arrived. It is defined to be: |
1815 |
|
1816 |
sub client_get { $_[0]->txn_client_get ($_[1])->result } |
1817 |
|
1818 |
And in fact is automatically generated. This is the blocking API of |
1819 |
L<Net::FCP>, and it works as simple as in any other, similar, module. |
1820 |
|
1821 |
More complicated is C<txn_client_get>: It only creates a transaction |
1822 |
(completion, result, ...) object and initiates the transaction. |
1823 |
|
1824 |
my $txn = bless { }, Net::FCP::Txn::; |
1825 |
|
1826 |
It also creates a condition variable that is used to signal the completion |
1827 |
of the request: |
1828 |
|
1829 |
$txn->{finished} = AnyAvent->condvar; |
1830 |
|
1831 |
It then creates a socket in non-blocking mode. |
1832 |
|
1833 |
socket $txn->{fh}, ...; |
1834 |
fcntl $txn->{fh}, F_SETFL, O_NONBLOCK; |
1835 |
connect $txn->{fh}, ... |
1836 |
and !$!{EWOULDBLOCK} |
1837 |
and !$!{EINPROGRESS} |
1838 |
and Carp::croak "unable to connect: $!\n"; |
1839 |
|
1840 |
Then it creates a write-watcher which gets called whenever an error occurs |
1841 |
or the connection succeeds: |
1842 |
|
1843 |
$txn->{w} = AnyEvent->io (fh => $txn->{fh}, poll => 'w', cb => sub { $txn->fh_ready_w }); |
1844 |
|
1845 |
And returns this transaction object. The C<fh_ready_w> callback gets |
1846 |
called as soon as the event loop detects that the socket is ready for |
1847 |
writing. |
1848 |
|
1849 |
The C<fh_ready_w> method makes the socket blocking again, writes the |
1850 |
request data and replaces the watcher by a read watcher (waiting for reply |
1851 |
data). The actual code is more complicated, but that doesn't matter for |
1852 |
this example: |
1853 |
|
1854 |
fcntl $txn->{fh}, F_SETFL, 0; |
1855 |
syswrite $txn->{fh}, $txn->{request} |
1856 |
or die "connection or write error"; |
1857 |
$txn->{w} = AnyEvent->io (fh => $txn->{fh}, poll => 'r', cb => sub { $txn->fh_ready_r }); |
1858 |
|
1859 |
Again, C<fh_ready_r> waits till all data has arrived, and then stores the |
1860 |
result and signals any possible waiters that the request has finished: |
1861 |
|
1862 |
sysread $txn->{fh}, $txn->{buf}, length $txn->{$buf}; |
1863 |
|
1864 |
if (end-of-file or data complete) { |
1865 |
$txn->{result} = $txn->{buf}; |
1866 |
$txn->{finished}->send; |
1867 |
$txb->{cb}->($txn) of $txn->{cb}; # also call callback |
1868 |
} |
1869 |
|
1870 |
The C<result> method, finally, just waits for the finished signal (if the |
1871 |
request was already finished, it doesn't wait, of course, and returns the |
1872 |
data: |
1873 |
|
1874 |
$txn->{finished}->recv; |
1875 |
return $txn->{result}; |
1876 |
|
1877 |
The actual code goes further and collects all errors (C<die>s, exceptions) |
1878 |
that occurred during request processing. The C<result> method detects |
1879 |
whether an exception as thrown (it is stored inside the $txn object) |
1880 |
and just throws the exception, which means connection errors and other |
1881 |
problems get reported tot he code that tries to use the result, not in a |
1882 |
random callback. |
1883 |
|
1884 |
All of this enables the following usage styles: |
1885 |
|
1886 |
1. Blocking: |
1887 |
|
1888 |
my $data = $fcp->client_get ($url); |
1889 |
|
1890 |
2. Blocking, but running in parallel: |
1891 |
|
1892 |
my @datas = map $_->result, |
1893 |
map $fcp->txn_client_get ($_), |
1894 |
@urls; |
1895 |
|
1896 |
Both blocking examples work without the module user having to know |
1897 |
anything about events. |
1898 |
|
1899 |
3a. Event-based in a main program, using any supported event module: |
1900 |
|
1901 |
use EV; |
1902 |
|
1903 |
$fcp->txn_client_get ($url)->cb (sub { |
1904 |
my $txn = shift; |
1905 |
my $data = $txn->result; |
1906 |
... |
1907 |
}); |
1908 |
|
1909 |
EV::loop; |
1910 |
|
1911 |
3b. The module user could use AnyEvent, too: |
1912 |
|
1913 |
use AnyEvent; |
1914 |
|
1915 |
my $quit = AnyEvent->condvar; |
1916 |
|
1917 |
$fcp->txn_client_get ($url)->cb (sub { |
1918 |
... |
1919 |
$quit->send; |
1920 |
}); |
1921 |
|
1922 |
$quit->recv; |
1923 |
|
1924 |
|
1925 |
=head1 BENCHMARKS |
1926 |
|
1927 |
To give you an idea of the performance and overheads that AnyEvent adds |
1928 |
over the event loops themselves and to give you an impression of the speed |
1929 |
of various event loops I prepared some benchmarks. |
1930 |
|
1931 |
=head2 BENCHMARKING ANYEVENT OVERHEAD |
1932 |
|
1933 |
Here is a benchmark of various supported event models used natively and |
1934 |
through AnyEvent. The benchmark creates a lot of timers (with a zero |
1935 |
timeout) and I/O watchers (watching STDOUT, a pty, to become writable, |
1936 |
which it is), lets them fire exactly once and destroys them again. |
1937 |
|
1938 |
Source code for this benchmark is found as F<eg/bench> in the AnyEvent |
1939 |
distribution. |
1940 |
|
1941 |
=head3 Explanation of the columns |
1942 |
|
1943 |
I<watcher> is the number of event watchers created/destroyed. Since |
1944 |
different event models feature vastly different performances, each event |
1945 |
loop was given a number of watchers so that overall runtime is acceptable |
1946 |
and similar between tested event loop (and keep them from crashing): Glib |
1947 |
would probably take thousands of years if asked to process the same number |
1948 |
of watchers as EV in this benchmark. |
1949 |
|
1950 |
I<bytes> is the number of bytes (as measured by the resident set size, |
1951 |
RSS) consumed by each watcher. This method of measuring captures both C |
1952 |
and Perl-based overheads. |
1953 |
|
1954 |
I<create> is the time, in microseconds (millionths of seconds), that it |
1955 |
takes to create a single watcher. The callback is a closure shared between |
1956 |
all watchers, to avoid adding memory overhead. That means closure creation |
1957 |
and memory usage is not included in the figures. |
1958 |
|
1959 |
I<invoke> is the time, in microseconds, used to invoke a simple |
1960 |
callback. The callback simply counts down a Perl variable and after it was |
1961 |
invoked "watcher" times, it would C<< ->send >> a condvar once to |
1962 |
signal the end of this phase. |
1963 |
|
1964 |
I<destroy> is the time, in microseconds, that it takes to destroy a single |
1965 |
watcher. |
1966 |
|
1967 |
=head3 Results |
1968 |
|
1969 |
name watchers bytes create invoke destroy comment |
1970 |
EV/EV 400000 224 0.47 0.35 0.27 EV native interface |
1971 |
EV/Any 100000 224 2.88 0.34 0.27 EV + AnyEvent watchers |
1972 |
CoroEV/Any 100000 224 2.85 0.35 0.28 coroutines + Coro::Signal |
1973 |
Perl/Any 100000 452 4.13 0.73 0.95 pure perl implementation |
1974 |
Event/Event 16000 517 32.20 31.80 0.81 Event native interface |
1975 |
Event/Any 16000 590 35.85 31.55 1.06 Event + AnyEvent watchers |
1976 |
IOAsync/Any 16000 989 38.10 32.77 11.13 via IO::Async::Loop::IO_Poll |
1977 |
IOAsync/Any 16000 990 37.59 29.50 10.61 via IO::Async::Loop::Epoll |
1978 |
Glib/Any 16000 1357 102.33 12.31 51.00 quadratic behaviour |
1979 |
Tk/Any 2000 1860 27.20 66.31 14.00 SEGV with >> 2000 watchers |
1980 |
POE/Event 2000 6328 109.99 751.67 14.02 via POE::Loop::Event |
1981 |
POE/Select 2000 6027 94.54 809.13 579.80 via POE::Loop::Select |
1982 |
|
1983 |
=head3 Discussion |
1984 |
|
1985 |
The benchmark does I<not> measure scalability of the event loop very |
1986 |
well. For example, a select-based event loop (such as the pure perl one) |
1987 |
can never compete with an event loop that uses epoll when the number of |
1988 |
file descriptors grows high. In this benchmark, all events become ready at |
1989 |
the same time, so select/poll-based implementations get an unnatural speed |
1990 |
boost. |
1991 |
|
1992 |
Also, note that the number of watchers usually has a nonlinear effect on |
1993 |
overall speed, that is, creating twice as many watchers doesn't take twice |
1994 |
the time - usually it takes longer. This puts event loops tested with a |
1995 |
higher number of watchers at a disadvantage. |
1996 |
|
1997 |
To put the range of results into perspective, consider that on the |
1998 |
benchmark machine, handling an event takes roughly 1600 CPU cycles with |
1999 |
EV, 3100 CPU cycles with AnyEvent's pure perl loop and almost 3000000 CPU |
2000 |
cycles with POE. |
2001 |
|
2002 |
C<EV> is the sole leader regarding speed and memory use, which are both |
2003 |
maximal/minimal, respectively. Even when going through AnyEvent, it uses |
2004 |
far less memory than any other event loop and is still faster than Event |
2005 |
natively. |
2006 |
|
2007 |
The pure perl implementation is hit in a few sweet spots (both the |
2008 |
constant timeout and the use of a single fd hit optimisations in the perl |
2009 |
interpreter and the backend itself). Nevertheless this shows that it |
2010 |
adds very little overhead in itself. Like any select-based backend its |
2011 |
performance becomes really bad with lots of file descriptors (and few of |
2012 |
them active), of course, but this was not subject of this benchmark. |
2013 |
|
2014 |
The C<Event> module has a relatively high setup and callback invocation |
2015 |
cost, but overall scores in on the third place. |
2016 |
|
2017 |
C<IO::Async> performs admirably well, about on par with C<Event>, even |
2018 |
when using its pure perl backend. |
2019 |
|
2020 |
C<Glib>'s memory usage is quite a bit higher, but it features a |
2021 |
faster callback invocation and overall ends up in the same class as |
2022 |
C<Event>. However, Glib scales extremely badly, doubling the number of |
2023 |
watchers increases the processing time by more than a factor of four, |
2024 |
making it completely unusable when using larger numbers of watchers |
2025 |
(note that only a single file descriptor was used in the benchmark, so |
2026 |
inefficiencies of C<poll> do not account for this). |
2027 |
|
2028 |
The C<Tk> adaptor works relatively well. The fact that it crashes with |
2029 |
more than 2000 watchers is a big setback, however, as correctness takes |
2030 |
precedence over speed. Nevertheless, its performance is surprising, as the |
2031 |
file descriptor is dup()ed for each watcher. This shows that the dup() |
2032 |
employed by some adaptors is not a big performance issue (it does incur a |
2033 |
hidden memory cost inside the kernel which is not reflected in the figures |
2034 |
above). |
2035 |
|
2036 |
C<POE>, regardless of underlying event loop (whether using its pure perl |
2037 |
select-based backend or the Event module, the POE-EV backend couldn't |
2038 |
be tested because it wasn't working) shows abysmal performance and |
2039 |
memory usage with AnyEvent: Watchers use almost 30 times as much memory |
2040 |
as EV watchers, and 10 times as much memory as Event (the high memory |
2041 |
requirements are caused by requiring a session for each watcher). Watcher |
2042 |
invocation speed is almost 900 times slower than with AnyEvent's pure perl |
2043 |
implementation. |
2044 |
|
2045 |
The design of the POE adaptor class in AnyEvent can not really account |
2046 |
for the performance issues, though, as session creation overhead is |
2047 |
small compared to execution of the state machine, which is coded pretty |
2048 |
optimally within L<AnyEvent::Impl::POE> (and while everybody agrees that |
2049 |
using multiple sessions is not a good approach, especially regarding |
2050 |
memory usage, even the author of POE could not come up with a faster |
2051 |
design). |
2052 |
|
2053 |
=head3 Summary |
2054 |
|
2055 |
=over 4 |
2056 |
|
2057 |
=item * Using EV through AnyEvent is faster than any other event loop |
2058 |
(even when used without AnyEvent), but most event loops have acceptable |
2059 |
performance with or without AnyEvent. |
2060 |
|
2061 |
=item * The overhead AnyEvent adds is usually much smaller than the overhead of |
2062 |
the actual event loop, only with extremely fast event loops such as EV |
2063 |
adds AnyEvent significant overhead. |
2064 |
|
2065 |
=item * You should avoid POE like the plague if you want performance or |
2066 |
reasonable memory usage. |
2067 |
|
2068 |
=back |
2069 |
|
2070 |
=head2 BENCHMARKING THE LARGE SERVER CASE |
2071 |
|
2072 |
This benchmark actually benchmarks the event loop itself. It works by |
2073 |
creating a number of "servers": each server consists of a socket pair, a |
2074 |
timeout watcher that gets reset on activity (but never fires), and an I/O |
2075 |
watcher waiting for input on one side of the socket. Each time the socket |
2076 |
watcher reads a byte it will write that byte to a random other "server". |
2077 |
|
2078 |
The effect is that there will be a lot of I/O watchers, only part of which |
2079 |
are active at any one point (so there is a constant number of active |
2080 |
fds for each loop iteration, but which fds these are is random). The |
2081 |
timeout is reset each time something is read because that reflects how |
2082 |
most timeouts work (and puts extra pressure on the event loops). |
2083 |
|
2084 |
In this benchmark, we use 10000 socket pairs (20000 sockets), of which 100 |
2085 |
(1%) are active. This mirrors the activity of large servers with many |
2086 |
connections, most of which are idle at any one point in time. |
2087 |
|
2088 |
Source code for this benchmark is found as F<eg/bench2> in the AnyEvent |
2089 |
distribution. |
2090 |
|
2091 |
=head3 Explanation of the columns |
2092 |
|
2093 |
I<sockets> is the number of sockets, and twice the number of "servers" (as |
2094 |
each server has a read and write socket end). |
2095 |
|
2096 |
I<create> is the time it takes to create a socket pair (which is |
2097 |
nontrivial) and two watchers: an I/O watcher and a timeout watcher. |
2098 |
|
2099 |
I<request>, the most important value, is the time it takes to handle a |
2100 |
single "request", that is, reading the token from the pipe and forwarding |
2101 |
it to another server. This includes deleting the old timeout and creating |
2102 |
a new one that moves the timeout into the future. |
2103 |
|
2104 |
=head3 Results |
2105 |
|
2106 |
name sockets create request |
2107 |
EV 20000 69.01 11.16 |
2108 |
Perl 20000 73.32 35.87 |
2109 |
IOAsync 20000 157.00 98.14 epoll |
2110 |
IOAsync 20000 159.31 616.06 poll |
2111 |
Event 20000 212.62 257.32 |
2112 |
Glib 20000 651.16 1896.30 |
2113 |
POE 20000 349.67 12317.24 uses POE::Loop::Event |
2114 |
|
2115 |
=head3 Discussion |
2116 |
|
2117 |
This benchmark I<does> measure scalability and overall performance of the |
2118 |
particular event loop. |
2119 |
|
2120 |
EV is again fastest. Since it is using epoll on my system, the setup time |
2121 |
is relatively high, though. |
2122 |
|
2123 |
Perl surprisingly comes second. It is much faster than the C-based event |
2124 |
loops Event and Glib. |
2125 |
|
2126 |
IO::Async performs very well when using its epoll backend, and still quite |
2127 |
good compared to Glib when using its pure perl backend. |
2128 |
|
2129 |
Event suffers from high setup time as well (look at its code and you will |
2130 |
understand why). Callback invocation also has a high overhead compared to |
2131 |
the C<< $_->() for .. >>-style loop that the Perl event loop uses. Event |
2132 |
uses select or poll in basically all documented configurations. |
2133 |
|
2134 |
Glib is hit hard by its quadratic behaviour w.r.t. many watchers. It |
2135 |
clearly fails to perform with many filehandles or in busy servers. |
2136 |
|
2137 |
POE is still completely out of the picture, taking over 1000 times as long |
2138 |
as EV, and over 100 times as long as the Perl implementation, even though |
2139 |
it uses a C-based event loop in this case. |
2140 |
|
2141 |
=head3 Summary |
2142 |
|
2143 |
=over 4 |
2144 |
|
2145 |
=item * The pure perl implementation performs extremely well. |
2146 |
|
2147 |
=item * Avoid Glib or POE in large projects where performance matters. |
2148 |
|
2149 |
=back |
2150 |
|
2151 |
=head2 BENCHMARKING SMALL SERVERS |
2152 |
|
2153 |
While event loops should scale (and select-based ones do not...) even to |
2154 |
large servers, most programs we (or I :) actually write have only a few |
2155 |
I/O watchers. |
2156 |
|
2157 |
In this benchmark, I use the same benchmark program as in the large server |
2158 |
case, but it uses only eight "servers", of which three are active at any |
2159 |
one time. This should reflect performance for a small server relatively |
2160 |
well. |
2161 |
|
2162 |
The columns are identical to the previous table. |
2163 |
|
2164 |
=head3 Results |
2165 |
|
2166 |
name sockets create request |
2167 |
EV 16 20.00 6.54 |
2168 |
Perl 16 25.75 12.62 |
2169 |
Event 16 81.27 35.86 |
2170 |
Glib 16 32.63 15.48 |
2171 |
POE 16 261.87 276.28 uses POE::Loop::Event |
2172 |
|
2173 |
=head3 Discussion |
2174 |
|
2175 |
The benchmark tries to test the performance of a typical small |
2176 |
server. While knowing how various event loops perform is interesting, keep |
2177 |
in mind that their overhead in this case is usually not as important, due |
2178 |
to the small absolute number of watchers (that is, you need efficiency and |
2179 |
speed most when you have lots of watchers, not when you only have a few of |
2180 |
them). |
2181 |
|
2182 |
EV is again fastest. |
2183 |
|
2184 |
Perl again comes second. It is noticeably faster than the C-based event |
2185 |
loops Event and Glib, although the difference is too small to really |
2186 |
matter. |
2187 |
|
2188 |
POE also performs much better in this case, but is is still far behind the |
2189 |
others. |
2190 |
|
2191 |
=head3 Summary |
2192 |
|
2193 |
=over 4 |
2194 |
|
2195 |
=item * C-based event loops perform very well with small number of |
2196 |
watchers, as the management overhead dominates. |
2197 |
|
2198 |
=back |
2199 |
|
2200 |
=head2 THE IO::Lambda BENCHMARK |
2201 |
|
2202 |
Recently I was told about the benchmark in the IO::Lambda manpage, which |
2203 |
could be misinterpreted to make AnyEvent look bad. In fact, the benchmark |
2204 |
simply compares IO::Lambda with POE, and IO::Lambda looks better (which |
2205 |
shouldn't come as a surprise to anybody). As such, the benchmark is |
2206 |
fine, and mostly shows that the AnyEvent backend from IO::Lambda isn't |
2207 |
very optimal. But how would AnyEvent compare when used without the extra |
2208 |
baggage? To explore this, I wrote the equivalent benchmark for AnyEvent. |
2209 |
|
2210 |
The benchmark itself creates an echo-server, and then, for 500 times, |
2211 |
connects to the echo server, sends a line, waits for the reply, and then |
2212 |
creates the next connection. This is a rather bad benchmark, as it doesn't |
2213 |
test the efficiency of the framework or much non-blocking I/O, but it is a |
2214 |
benchmark nevertheless. |
2215 |
|
2216 |
name runtime |
2217 |
Lambda/select 0.330 sec |
2218 |
+ optimized 0.122 sec |
2219 |
Lambda/AnyEvent 0.327 sec |
2220 |
+ optimized 0.138 sec |
2221 |
Raw sockets/select 0.077 sec |
2222 |
POE/select, components 0.662 sec |
2223 |
POE/select, raw sockets 0.226 sec |
2224 |
POE/select, optimized 0.404 sec |
2225 |
|
2226 |
AnyEvent/select/nb 0.085 sec |
2227 |
AnyEvent/EV/nb 0.068 sec |
2228 |
+state machine 0.134 sec |
2229 |
|
2230 |
The benchmark is also a bit unfair (my fault): the IO::Lambda/POE |
2231 |
benchmarks actually make blocking connects and use 100% blocking I/O, |
2232 |
defeating the purpose of an event-based solution. All of the newly |
2233 |
written AnyEvent benchmarks use 100% non-blocking connects (using |
2234 |
AnyEvent::Socket::tcp_connect and the asynchronous pure perl DNS |
2235 |
resolver), so AnyEvent is at a disadvantage here, as non-blocking connects |
2236 |
generally require a lot more bookkeeping and event handling than blocking |
2237 |
connects (which involve a single syscall only). |
2238 |
|
2239 |
The last AnyEvent benchmark additionally uses L<AnyEvent::Handle>, which |
2240 |
offers similar expressive power as POE and IO::Lambda, using conventional |
2241 |
Perl syntax. This means that both the echo server and the client are 100% |
2242 |
non-blocking, further placing it at a disadvantage. |
2243 |
|
2244 |
As you can see, the AnyEvent + EV combination even beats the |
2245 |
hand-optimised "raw sockets benchmark", while AnyEvent + its pure perl |
2246 |
backend easily beats IO::Lambda and POE. |
2247 |
|
2248 |
And even the 100% non-blocking version written using the high-level (and |
2249 |
slow :) L<AnyEvent::Handle> abstraction beats both POE and IO::Lambda by a |
2250 |
large margin, even though it does all of DNS, tcp-connect and socket I/O |
2251 |
in a non-blocking way. |
2252 |
|
2253 |
The two AnyEvent benchmarks programs can be found as F<eg/ae0.pl> and |
2254 |
F<eg/ae2.pl> in the AnyEvent distribution, the remaining benchmarks are |
2255 |
part of the IO::lambda distribution and were used without any changes. |
2256 |
|
2257 |
|
2258 |
=head1 SIGNALS |
2259 |
|
2260 |
AnyEvent currently installs handlers for these signals: |
2261 |
|
2262 |
=over 4 |
2263 |
|
2264 |
=item SIGCHLD |
2265 |
|
2266 |
A handler for C<SIGCHLD> is installed by AnyEvent's child watcher |
2267 |
emulation for event loops that do not support them natively. Also, some |
2268 |
event loops install a similar handler. |
2269 |
|
2270 |
Additionally, when AnyEvent is loaded and SIGCHLD is set to IGNORE, then |
2271 |
AnyEvent will reset it to default, to avoid losing child exit statuses. |
2272 |
|
2273 |
=item SIGPIPE |
2274 |
|
2275 |
A no-op handler is installed for C<SIGPIPE> when C<$SIG{PIPE}> is C<undef> |
2276 |
when AnyEvent gets loaded. |
2277 |
|
2278 |
The rationale for this is that AnyEvent users usually do not really depend |
2279 |
on SIGPIPE delivery (which is purely an optimisation for shell use, or |
2280 |
badly-written programs), but C<SIGPIPE> can cause spurious and rare |
2281 |
program exits as a lot of people do not expect C<SIGPIPE> when writing to |
2282 |
some random socket. |
2283 |
|
2284 |
The rationale for installing a no-op handler as opposed to ignoring it is |
2285 |
that this way, the handler will be restored to defaults on exec. |
2286 |
|
2287 |
Feel free to install your own handler, or reset it to defaults. |
2288 |
|
2289 |
=back |
2290 |
|
2291 |
=cut |
2292 |
|
2293 |
undef $SIG{CHLD} |
2294 |
if $SIG{CHLD} eq 'IGNORE'; |
2295 |
|
2296 |
$SIG{PIPE} = sub { } |
2297 |
unless defined $SIG{PIPE}; |
2298 |
|
2299 |
=head1 RECOMMENDED/OPTIONAL MODULES |
2300 |
|
2301 |
One of AnyEvent's main goals is to be 100% Pure-Perl(tm): only perl (and |
2302 |
it's built-in modules) are required to use it. |
2303 |
|
2304 |
That does not mean that AnyEvent won't take advantage of some additional |
2305 |
modules if they are installed. |
2306 |
|
2307 |
This section epxlains which additional modules will be used, and how they |
2308 |
affect AnyEvent's operetion. |
2309 |
|
2310 |
=over 4 |
2311 |
|
2312 |
=item L<Async::Interrupt> |
2313 |
|
2314 |
This slightly arcane module is used to implement fast signal handling: To |
2315 |
my knowledge, there is no way to do completely race-free and quick |
2316 |
signal handling in pure perl. To ensure that signals still get |
2317 |
delivered, AnyEvent will start an interval timer to wake up perl (and |
2318 |
catch the signals) with some delay (default is 10 seconds, look for |
2319 |
C<$AnyEvent::MAX_SIGNAL_LATENCY>). |
2320 |
|
2321 |
If this module is available, then it will be used to implement signal |
2322 |
catching, which means that signals will not be delayed, and the event loop |
2323 |
will not be interrupted regularly, which is more efficient (And good for |
2324 |
battery life on laptops). |
2325 |
|
2326 |
This affects not just the pure-perl event loop, but also other event loops |
2327 |
that have no signal handling on their own (e.g. Glib, Tk, Qt). |
2328 |
|
2329 |
Some event loops (POE, Event, Event::Lib) offer signal watchers natively, |
2330 |
and either employ their own workarounds (POE) or use AnyEvent's workaround |
2331 |
(using C<$AnyEvent::MAX_SIGNAL_LATENCY>). Installing L<Async::Interrupt> |
2332 |
does nothing for those backends. |
2333 |
|
2334 |
=item L<EV> |
2335 |
|
2336 |
This module isn't really "optional", as it is simply one of the backend |
2337 |
event loops that AnyEvent can use. However, it is simply the best event |
2338 |
loop available in terms of features, speed and stability: It supports |
2339 |
the AnyEvent API optimally, implements all the watcher types in XS, does |
2340 |
automatic timer adjustments even when no monotonic clock is available, |
2341 |
can take avdantage of advanced kernel interfaces such as C<epoll> and |
2342 |
C<kqueue>, and is the fastest backend I<by far>. You can even embed |
2343 |
L<Glib>/L<Gtk2> in it (or vice versa, see L<EV::Glib> and L<Glib::EV>). |
2344 |
|
2345 |
=item L<Guard> |
2346 |
|
2347 |
The guard module, when used, will be used to implement |
2348 |
C<AnyEvent::Util::guard>. This speeds up guards considerably (and uses a |
2349 |
lot less memory), but otherwise doesn't affect guard operation much. It is |
2350 |
purely used for performance. |
2351 |
|
2352 |
=item L<JSON> and L<JSON::XS> |
2353 |
|
2354 |
This module is required when you want to read or write JSON data via |
2355 |
L<AnyEvent::Handle>. It is also written in pure-perl, but can take |
2356 |
advantage of the ultra-high-speed L<JSON::XS> module when it is installed. |
2357 |
|
2358 |
In fact, L<AnyEvent::Handle> will use L<JSON::XS> by default if it is |
2359 |
installed. |
2360 |
|
2361 |
=item L<Net::SSLeay> |
2362 |
|
2363 |
Implementing TLS/SSL in Perl is certainly interesting, but not very |
2364 |
worthwhile: If this module is installed, then L<AnyEvent::Handle> (with |
2365 |
the help of L<AnyEvent::TLS>), gains the ability to do TLS/SSL. |
2366 |
|
2367 |
=item L<Time::HiRes> |
2368 |
|
2369 |
This module is part of perl since release 5.008. It will be used when the |
2370 |
chosen event library does not come with a timing source on it's own. The |
2371 |
pure-perl event loop (L<AnyEvent::Impl::Perl>) will additionally use it to |
2372 |
try to use a monotonic clock for timing stability. |
2373 |
|
2374 |
=back |
2375 |
|
2376 |
|
2377 |
=head1 FORK |
2378 |
|
2379 |
Most event libraries are not fork-safe. The ones who are usually are |
2380 |
because they rely on inefficient but fork-safe C<select> or C<poll> |
2381 |
calls. Only L<EV> is fully fork-aware. |
2382 |
|
2383 |
If you have to fork, you must either do so I<before> creating your first |
2384 |
watcher OR you must not use AnyEvent at all in the child OR you must do |
2385 |
something completely out of the scope of AnyEvent. |
2386 |
|
2387 |
|
2388 |
=head1 SECURITY CONSIDERATIONS |
2389 |
|
2390 |
AnyEvent can be forced to load any event model via |
2391 |
$ENV{PERL_ANYEVENT_MODEL}. While this cannot (to my knowledge) be used to |
2392 |
execute arbitrary code or directly gain access, it can easily be used to |
2393 |
make the program hang or malfunction in subtle ways, as AnyEvent watchers |
2394 |
will not be active when the program uses a different event model than |
2395 |
specified in the variable. |
2396 |
|
2397 |
You can make AnyEvent completely ignore this variable by deleting it |
2398 |
before the first watcher gets created, e.g. with a C<BEGIN> block: |
2399 |
|
2400 |
BEGIN { delete $ENV{PERL_ANYEVENT_MODEL} } |
2401 |
|
2402 |
use AnyEvent; |
2403 |
|
2404 |
Similar considerations apply to $ENV{PERL_ANYEVENT_VERBOSE}, as that can |
2405 |
be used to probe what backend is used and gain other information (which is |
2406 |
probably even less useful to an attacker than PERL_ANYEVENT_MODEL), and |
2407 |
$ENV{PERL_ANYEVENT_STRICT}. |
2408 |
|
2409 |
Note that AnyEvent will remove I<all> environment variables starting with |
2410 |
C<PERL_ANYEVENT_> from C<%ENV> when it is loaded while taint mode is |
2411 |
enabled. |
2412 |
|
2413 |
|
2414 |
=head1 BUGS |
2415 |
|
2416 |
Perl 5.8 has numerous memleaks that sometimes hit this module and are hard |
2417 |
to work around. If you suffer from memleaks, first upgrade to Perl 5.10 |
2418 |
and check wether the leaks still show up. (Perl 5.10.0 has other annoying |
2419 |
memleaks, such as leaking on C<map> and C<grep> but it is usually not as |
2420 |
pronounced). |
2421 |
|
2422 |
|
2423 |
=head1 SEE ALSO |
2424 |
|
2425 |
Utility functions: L<AnyEvent::Util>. |
2426 |
|
2427 |
Event modules: L<EV>, L<EV::Glib>, L<Glib::EV>, L<Event>, L<Glib::Event>, |
2428 |
L<Glib>, L<Tk>, L<Event::Lib>, L<Qt>, L<POE>. |
2429 |
|
2430 |
Implementations: L<AnyEvent::Impl::EV>, L<AnyEvent::Impl::Event>, |
2431 |
L<AnyEvent::Impl::Glib>, L<AnyEvent::Impl::Tk>, L<AnyEvent::Impl::Perl>, |
2432 |
L<AnyEvent::Impl::EventLib>, L<AnyEvent::Impl::Qt>, |
2433 |
L<AnyEvent::Impl::POE>, L<AnyEvent::Impl::IOAsync>, L<Anyevent::Impl::Irssi>. |
2434 |
|
2435 |
Non-blocking file handles, sockets, TCP clients and |
2436 |
servers: L<AnyEvent::Handle>, L<AnyEvent::Socket>, L<AnyEvent::TLS>. |
2437 |
|
2438 |
Asynchronous DNS: L<AnyEvent::DNS>. |
2439 |
|
2440 |
Coroutine support: L<Coro>, L<Coro::AnyEvent>, L<Coro::EV>, |
2441 |
L<Coro::Event>, |
2442 |
|
2443 |
Nontrivial usage examples: L<AnyEvent::GPSD>, L<AnyEvent::XMPP>, |
2444 |
L<AnyEvent::HTTP>. |
2445 |
|
2446 |
|
2447 |
=head1 AUTHOR |
2448 |
|
2449 |
Marc Lehmann <schmorp@schmorp.de> |
2450 |
http://home.schmorp.de/ |
2451 |
|
2452 |
=cut |
2453 |
|
2454 |
1 |
2455 |
|