1 |
=head1 NAME |
2 |
|
3 |
AnyEvent - provide framework for multiple event loops |
4 |
|
5 |
EV, Event, Glib, Tk, Perl, Event::Lib, Qt, POE - various supported event loops |
6 |
|
7 |
=head1 SYNOPSIS |
8 |
|
9 |
use AnyEvent; |
10 |
|
11 |
my $w = AnyEvent->io (fh => $fh, poll => "r|w", cb => sub { ... }); |
12 |
|
13 |
my $w = AnyEvent->timer (after => $seconds, cb => sub { ... }); |
14 |
my $w = AnyEvent->timer (after => $seconds, interval => $seconds, cb => ... |
15 |
|
16 |
print AnyEvent->now; # prints current event loop time |
17 |
print AnyEvent->time; # think Time::HiRes::time or simply CORE::time. |
18 |
|
19 |
my $w = AnyEvent->signal (signal => "TERM", cb => sub { ... }); |
20 |
|
21 |
my $w = AnyEvent->child (pid => $pid, cb => sub { |
22 |
my ($pid, $status) = @_; |
23 |
... |
24 |
}); |
25 |
|
26 |
my $w = AnyEvent->condvar; # stores whether a condition was flagged |
27 |
$w->send; # wake up current and all future recv's |
28 |
$w->recv; # enters "main loop" till $condvar gets ->send |
29 |
# use a condvar in callback mode: |
30 |
$w->cb (sub { $_[0]->recv }); |
31 |
|
32 |
=head1 INTRODUCTION/TUTORIAL |
33 |
|
34 |
This manpage is mainly a reference manual. If you are interested |
35 |
in a tutorial or some gentle introduction, have a look at the |
36 |
L<AnyEvent::Intro> manpage. |
37 |
|
38 |
=head1 WHY YOU SHOULD USE THIS MODULE (OR NOT) |
39 |
|
40 |
Glib, POE, IO::Async, Event... CPAN offers event models by the dozen |
41 |
nowadays. So what is different about AnyEvent? |
42 |
|
43 |
Executive Summary: AnyEvent is I<compatible>, AnyEvent is I<free of |
44 |
policy> and AnyEvent is I<small and efficient>. |
45 |
|
46 |
First and foremost, I<AnyEvent is not an event model> itself, it only |
47 |
interfaces to whatever event model the main program happens to use, in a |
48 |
pragmatic way. For event models and certain classes of immortals alike, |
49 |
the statement "there can only be one" is a bitter reality: In general, |
50 |
only one event loop can be active at the same time in a process. AnyEvent |
51 |
cannot change this, but it can hide the differences between those event |
52 |
loops. |
53 |
|
54 |
The goal of AnyEvent is to offer module authors the ability to do event |
55 |
programming (waiting for I/O or timer events) without subscribing to a |
56 |
religion, a way of living, and most importantly: without forcing your |
57 |
module users into the same thing by forcing them to use the same event |
58 |
model you use. |
59 |
|
60 |
For modules like POE or IO::Async (which is a total misnomer as it is |
61 |
actually doing all I/O I<synchronously>...), using them in your module is |
62 |
like joining a cult: After you joined, you are dependent on them and you |
63 |
cannot use anything else, as they are simply incompatible to everything |
64 |
that isn't them. What's worse, all the potential users of your |
65 |
module are I<also> forced to use the same event loop you use. |
66 |
|
67 |
AnyEvent is different: AnyEvent + POE works fine. AnyEvent + Glib works |
68 |
fine. AnyEvent + Tk works fine etc. etc. but none of these work together |
69 |
with the rest: POE + IO::Async? No go. Tk + Event? No go. Again: if |
70 |
your module uses one of those, every user of your module has to use it, |
71 |
too. But if your module uses AnyEvent, it works transparently with all |
72 |
event models it supports (including stuff like IO::Async, as long as those |
73 |
use one of the supported event loops. It is trivial to add new event loops |
74 |
to AnyEvent, too, so it is future-proof). |
75 |
|
76 |
In addition to being free of having to use I<the one and only true event |
77 |
model>, AnyEvent also is free of bloat and policy: with POE or similar |
78 |
modules, you get an enormous amount of code and strict rules you have to |
79 |
follow. AnyEvent, on the other hand, is lean and up to the point, by only |
80 |
offering the functionality that is necessary, in as thin as a wrapper as |
81 |
technically possible. |
82 |
|
83 |
Of course, AnyEvent comes with a big (and fully optional!) toolbox |
84 |
of useful functionality, such as an asynchronous DNS resolver, 100% |
85 |
non-blocking connects (even with TLS/SSL, IPv6 and on broken platforms |
86 |
such as Windows) and lots of real-world knowledge and workarounds for |
87 |
platform bugs and differences. |
88 |
|
89 |
Now, if you I<do want> lots of policy (this can arguably be somewhat |
90 |
useful) and you want to force your users to use the one and only event |
91 |
model, you should I<not> use this module. |
92 |
|
93 |
=head1 DESCRIPTION |
94 |
|
95 |
L<AnyEvent> provides an identical interface to multiple event loops. This |
96 |
allows module authors to utilise an event loop without forcing module |
97 |
users to use the same event loop (as only a single event loop can coexist |
98 |
peacefully at any one time). |
99 |
|
100 |
The interface itself is vaguely similar, but not identical to the L<Event> |
101 |
module. |
102 |
|
103 |
During the first call of any watcher-creation method, the module tries |
104 |
to detect the currently loaded event loop by probing whether one of the |
105 |
following modules is already loaded: L<EV>, |
106 |
L<Event>, L<Glib>, L<AnyEvent::Impl::Perl>, L<Tk>, L<Event::Lib>, L<Qt>, |
107 |
L<POE>. The first one found is used. If none are found, the module tries |
108 |
to load these modules (excluding Tk, Event::Lib, Qt and POE as the pure perl |
109 |
adaptor should always succeed) in the order given. The first one that can |
110 |
be successfully loaded will be used. If, after this, still none could be |
111 |
found, AnyEvent will fall back to a pure-perl event loop, which is not |
112 |
very efficient, but should work everywhere. |
113 |
|
114 |
Because AnyEvent first checks for modules that are already loaded, loading |
115 |
an event model explicitly before first using AnyEvent will likely make |
116 |
that model the default. For example: |
117 |
|
118 |
use Tk; |
119 |
use AnyEvent; |
120 |
|
121 |
# .. AnyEvent will likely default to Tk |
122 |
|
123 |
The I<likely> means that, if any module loads another event model and |
124 |
starts using it, all bets are off. Maybe you should tell their authors to |
125 |
use AnyEvent so their modules work together with others seamlessly... |
126 |
|
127 |
The pure-perl implementation of AnyEvent is called |
128 |
C<AnyEvent::Impl::Perl>. Like other event modules you can load it |
129 |
explicitly and enjoy the high availability of that event loop :) |
130 |
|
131 |
=head1 WATCHERS |
132 |
|
133 |
AnyEvent has the central concept of a I<watcher>, which is an object that |
134 |
stores relevant data for each kind of event you are waiting for, such as |
135 |
the callback to call, the file handle to watch, etc. |
136 |
|
137 |
These watchers are normal Perl objects with normal Perl lifetime. After |
138 |
creating a watcher it will immediately "watch" for events and invoke the |
139 |
callback when the event occurs (of course, only when the event model |
140 |
is in control). |
141 |
|
142 |
Note that B<callbacks must not permanently change global variables> |
143 |
potentially in use by the event loop (such as C<$_> or C<$[>) and that B<< |
144 |
callbacks must not C<die> >>. The former is good programming practise in |
145 |
Perl and the latter stems from the fact that exception handling differs |
146 |
widely between event loops. |
147 |
|
148 |
To disable the watcher you have to destroy it (e.g. by setting the |
149 |
variable you store it in to C<undef> or otherwise deleting all references |
150 |
to it). |
151 |
|
152 |
All watchers are created by calling a method on the C<AnyEvent> class. |
153 |
|
154 |
Many watchers either are used with "recursion" (repeating timers for |
155 |
example), or need to refer to their watcher object in other ways. |
156 |
|
157 |
An any way to achieve that is this pattern: |
158 |
|
159 |
my $w; $w = AnyEvent->type (arg => value ..., cb => sub { |
160 |
# you can use $w here, for example to undef it |
161 |
undef $w; |
162 |
}); |
163 |
|
164 |
Note that C<my $w; $w => combination. This is necessary because in Perl, |
165 |
my variables are only visible after the statement in which they are |
166 |
declared. |
167 |
|
168 |
=head2 I/O WATCHERS |
169 |
|
170 |
You can create an I/O watcher by calling the C<< AnyEvent->io >> method |
171 |
with the following mandatory key-value pairs as arguments: |
172 |
|
173 |
C<fh> is the Perl I<file handle> (I<not> file descriptor) to watch |
174 |
for events (AnyEvent might or might not keep a reference to this file |
175 |
handle). Note that only file handles pointing to things for which |
176 |
non-blocking operation makes sense are allowed. This includes sockets, |
177 |
most character devices, pipes, fifos and so on, but not for example files |
178 |
or block devices. |
179 |
|
180 |
C<poll> must be a string that is either C<r> or C<w>, which creates a |
181 |
watcher waiting for "r"eadable or "w"ritable events, respectively. |
182 |
|
183 |
C<cb> is the callback to invoke each time the file handle becomes ready. |
184 |
|
185 |
Although the callback might get passed parameters, their value and |
186 |
presence is undefined and you cannot rely on them. Portable AnyEvent |
187 |
callbacks cannot use arguments passed to I/O watcher callbacks. |
188 |
|
189 |
The I/O watcher might use the underlying file descriptor or a copy of it. |
190 |
You must not close a file handle as long as any watcher is active on the |
191 |
underlying file descriptor. |
192 |
|
193 |
Some event loops issue spurious readyness notifications, so you should |
194 |
always use non-blocking calls when reading/writing from/to your file |
195 |
handles. |
196 |
|
197 |
Example: wait for readability of STDIN, then read a line and disable the |
198 |
watcher. |
199 |
|
200 |
my $w; $w = AnyEvent->io (fh => \*STDIN, poll => 'r', cb => sub { |
201 |
chomp (my $input = <STDIN>); |
202 |
warn "read: $input\n"; |
203 |
undef $w; |
204 |
}); |
205 |
|
206 |
=head2 TIME WATCHERS |
207 |
|
208 |
You can create a time watcher by calling the C<< AnyEvent->timer >> |
209 |
method with the following mandatory arguments: |
210 |
|
211 |
C<after> specifies after how many seconds (fractional values are |
212 |
supported) the callback should be invoked. C<cb> is the callback to invoke |
213 |
in that case. |
214 |
|
215 |
Although the callback might get passed parameters, their value and |
216 |
presence is undefined and you cannot rely on them. Portable AnyEvent |
217 |
callbacks cannot use arguments passed to time watcher callbacks. |
218 |
|
219 |
The callback will normally be invoked once only. If you specify another |
220 |
parameter, C<interval>, as a strictly positive number (> 0), then the |
221 |
callback will be invoked regularly at that interval (in fractional |
222 |
seconds) after the first invocation. If C<interval> is specified with a |
223 |
false value, then it is treated as if it were missing. |
224 |
|
225 |
The callback will be rescheduled before invoking the callback, but no |
226 |
attempt is done to avoid timer drift in most backends, so the interval is |
227 |
only approximate. |
228 |
|
229 |
Example: fire an event after 7.7 seconds. |
230 |
|
231 |
my $w = AnyEvent->timer (after => 7.7, cb => sub { |
232 |
warn "timeout\n"; |
233 |
}); |
234 |
|
235 |
# to cancel the timer: |
236 |
undef $w; |
237 |
|
238 |
Example 2: fire an event after 0.5 seconds, then roughly every second. |
239 |
|
240 |
my $w = AnyEvent->timer (after => 0.5, interval => 1, cb => sub { |
241 |
warn "timeout\n"; |
242 |
}; |
243 |
|
244 |
=head3 TIMING ISSUES |
245 |
|
246 |
There are two ways to handle timers: based on real time (relative, "fire |
247 |
in 10 seconds") and based on wallclock time (absolute, "fire at 12 |
248 |
o'clock"). |
249 |
|
250 |
While most event loops expect timers to specified in a relative way, they |
251 |
use absolute time internally. This makes a difference when your clock |
252 |
"jumps", for example, when ntp decides to set your clock backwards from |
253 |
the wrong date of 2014-01-01 to 2008-01-01, a watcher that is supposed to |
254 |
fire "after" a second might actually take six years to finally fire. |
255 |
|
256 |
AnyEvent cannot compensate for this. The only event loop that is conscious |
257 |
about these issues is L<EV>, which offers both relative (ev_timer, based |
258 |
on true relative time) and absolute (ev_periodic, based on wallclock time) |
259 |
timers. |
260 |
|
261 |
AnyEvent always prefers relative timers, if available, matching the |
262 |
AnyEvent API. |
263 |
|
264 |
AnyEvent has two additional methods that return the "current time": |
265 |
|
266 |
=over 4 |
267 |
|
268 |
=item AnyEvent->time |
269 |
|
270 |
This returns the "current wallclock time" as a fractional number of |
271 |
seconds since the Epoch (the same thing as C<time> or C<Time::HiRes::time> |
272 |
return, and the result is guaranteed to be compatible with those). |
273 |
|
274 |
It progresses independently of any event loop processing, i.e. each call |
275 |
will check the system clock, which usually gets updated frequently. |
276 |
|
277 |
=item AnyEvent->now |
278 |
|
279 |
This also returns the "current wallclock time", but unlike C<time>, above, |
280 |
this value might change only once per event loop iteration, depending on |
281 |
the event loop (most return the same time as C<time>, above). This is the |
282 |
time that AnyEvent's timers get scheduled against. |
283 |
|
284 |
I<In almost all cases (in all cases if you don't care), this is the |
285 |
function to call when you want to know the current time.> |
286 |
|
287 |
This function is also often faster then C<< AnyEvent->time >>, and |
288 |
thus the preferred method if you want some timestamp (for example, |
289 |
L<AnyEvent::Handle> uses this to update it's activity timeouts). |
290 |
|
291 |
The rest of this section is only of relevance if you try to be very exact |
292 |
with your timing, you can skip it without bad conscience. |
293 |
|
294 |
For a practical example of when these times differ, consider L<Event::Lib> |
295 |
and L<EV> and the following set-up: |
296 |
|
297 |
The event loop is running and has just invoked one of your callback at |
298 |
time=500 (assume no other callbacks delay processing). In your callback, |
299 |
you wait a second by executing C<sleep 1> (blocking the process for a |
300 |
second) and then (at time=501) you create a relative timer that fires |
301 |
after three seconds. |
302 |
|
303 |
With L<Event::Lib>, C<< AnyEvent->time >> and C<< AnyEvent->now >> will |
304 |
both return C<501>, because that is the current time, and the timer will |
305 |
be scheduled to fire at time=504 (C<501> + C<3>). |
306 |
|
307 |
With L<EV>, C<< AnyEvent->time >> returns C<501> (as that is the current |
308 |
time), but C<< AnyEvent->now >> returns C<500>, as that is the time the |
309 |
last event processing phase started. With L<EV>, your timer gets scheduled |
310 |
to run at time=503 (C<500> + C<3>). |
311 |
|
312 |
In one sense, L<Event::Lib> is more exact, as it uses the current time |
313 |
regardless of any delays introduced by event processing. However, most |
314 |
callbacks do not expect large delays in processing, so this causes a |
315 |
higher drift (and a lot more system calls to get the current time). |
316 |
|
317 |
In another sense, L<EV> is more exact, as your timer will be scheduled at |
318 |
the same time, regardless of how long event processing actually took. |
319 |
|
320 |
In either case, if you care (and in most cases, you don't), then you |
321 |
can get whatever behaviour you want with any event loop, by taking the |
322 |
difference between C<< AnyEvent->time >> and C<< AnyEvent->now >> into |
323 |
account. |
324 |
|
325 |
=back |
326 |
|
327 |
=head2 SIGNAL WATCHERS |
328 |
|
329 |
You can watch for signals using a signal watcher, C<signal> is the signal |
330 |
I<name> in uppercase and without any C<SIG> prefix, C<cb> is the Perl |
331 |
callback to be invoked whenever a signal occurs. |
332 |
|
333 |
Although the callback might get passed parameters, their value and |
334 |
presence is undefined and you cannot rely on them. Portable AnyEvent |
335 |
callbacks cannot use arguments passed to signal watcher callbacks. |
336 |
|
337 |
Multiple signal occurrences can be clumped together into one callback |
338 |
invocation, and callback invocation will be synchronous. Synchronous means |
339 |
that it might take a while until the signal gets handled by the process, |
340 |
but it is guaranteed not to interrupt any other callbacks. |
341 |
|
342 |
The main advantage of using these watchers is that you can share a signal |
343 |
between multiple watchers. |
344 |
|
345 |
This watcher might use C<%SIG>, so programs overwriting those signals |
346 |
directly will likely not work correctly. |
347 |
|
348 |
Example: exit on SIGINT |
349 |
|
350 |
my $w = AnyEvent->signal (signal => "INT", cb => sub { exit 1 }); |
351 |
|
352 |
=head2 CHILD PROCESS WATCHERS |
353 |
|
354 |
You can also watch on a child process exit and catch its exit status. |
355 |
|
356 |
The child process is specified by the C<pid> argument (if set to C<0>, it |
357 |
watches for any child process exit). The watcher will triggered only when |
358 |
the child process has finished and an exit status is available, not on |
359 |
any trace events (stopped/continued). |
360 |
|
361 |
The callback will be called with the pid and exit status (as returned by |
362 |
waitpid), so unlike other watcher types, you I<can> rely on child watcher |
363 |
callback arguments. |
364 |
|
365 |
This watcher type works by installing a signal handler for C<SIGCHLD>, |
366 |
and since it cannot be shared, nothing else should use SIGCHLD or reap |
367 |
random child processes (waiting for specific child processes, e.g. inside |
368 |
C<system>, is just fine). |
369 |
|
370 |
There is a slight catch to child watchers, however: you usually start them |
371 |
I<after> the child process was created, and this means the process could |
372 |
have exited already (and no SIGCHLD will be sent anymore). |
373 |
|
374 |
Not all event models handle this correctly (POE doesn't), but even for |
375 |
event models that I<do> handle this correctly, they usually need to be |
376 |
loaded before the process exits (i.e. before you fork in the first place). |
377 |
|
378 |
This means you cannot create a child watcher as the very first thing in an |
379 |
AnyEvent program, you I<have> to create at least one watcher before you |
380 |
C<fork> the child (alternatively, you can call C<AnyEvent::detect>). |
381 |
|
382 |
Example: fork a process and wait for it |
383 |
|
384 |
my $done = AnyEvent->condvar; |
385 |
|
386 |
my $pid = fork or exit 5; |
387 |
|
388 |
my $w = AnyEvent->child ( |
389 |
pid => $pid, |
390 |
cb => sub { |
391 |
my ($pid, $status) = @_; |
392 |
warn "pid $pid exited with status $status"; |
393 |
$done->send; |
394 |
}, |
395 |
); |
396 |
|
397 |
# do something else, then wait for process exit |
398 |
$done->recv; |
399 |
|
400 |
=head2 CONDITION VARIABLES |
401 |
|
402 |
If you are familiar with some event loops you will know that all of them |
403 |
require you to run some blocking "loop", "run" or similar function that |
404 |
will actively watch for new events and call your callbacks. |
405 |
|
406 |
AnyEvent is different, it expects somebody else to run the event loop and |
407 |
will only block when necessary (usually when told by the user). |
408 |
|
409 |
The instrument to do that is called a "condition variable", so called |
410 |
because they represent a condition that must become true. |
411 |
|
412 |
Condition variables can be created by calling the C<< AnyEvent->condvar |
413 |
>> method, usually without arguments. The only argument pair allowed is |
414 |
|
415 |
C<cb>, which specifies a callback to be called when the condition variable |
416 |
becomes true, with the condition variable as the first argument (but not |
417 |
the results). |
418 |
|
419 |
After creation, the condition variable is "false" until it becomes "true" |
420 |
by calling the C<send> method (or calling the condition variable as if it |
421 |
were a callback, read about the caveats in the description for the C<< |
422 |
->send >> method). |
423 |
|
424 |
Condition variables are similar to callbacks, except that you can |
425 |
optionally wait for them. They can also be called merge points - points |
426 |
in time where multiple outstanding events have been processed. And yet |
427 |
another way to call them is transactions - each condition variable can be |
428 |
used to represent a transaction, which finishes at some point and delivers |
429 |
a result. |
430 |
|
431 |
Condition variables are very useful to signal that something has finished, |
432 |
for example, if you write a module that does asynchronous http requests, |
433 |
then a condition variable would be the ideal candidate to signal the |
434 |
availability of results. The user can either act when the callback is |
435 |
called or can synchronously C<< ->recv >> for the results. |
436 |
|
437 |
You can also use them to simulate traditional event loops - for example, |
438 |
you can block your main program until an event occurs - for example, you |
439 |
could C<< ->recv >> in your main program until the user clicks the Quit |
440 |
button of your app, which would C<< ->send >> the "quit" event. |
441 |
|
442 |
Note that condition variables recurse into the event loop - if you have |
443 |
two pieces of code that call C<< ->recv >> in a round-robin fashion, you |
444 |
lose. Therefore, condition variables are good to export to your caller, but |
445 |
you should avoid making a blocking wait yourself, at least in callbacks, |
446 |
as this asks for trouble. |
447 |
|
448 |
Condition variables are represented by hash refs in perl, and the keys |
449 |
used by AnyEvent itself are all named C<_ae_XXX> to make subclassing |
450 |
easy (it is often useful to build your own transaction class on top of |
451 |
AnyEvent). To subclass, use C<AnyEvent::CondVar> as base class and call |
452 |
it's C<new> method in your own C<new> method. |
453 |
|
454 |
There are two "sides" to a condition variable - the "producer side" which |
455 |
eventually calls C<< -> send >>, and the "consumer side", which waits |
456 |
for the send to occur. |
457 |
|
458 |
Example: wait for a timer. |
459 |
|
460 |
# wait till the result is ready |
461 |
my $result_ready = AnyEvent->condvar; |
462 |
|
463 |
# do something such as adding a timer |
464 |
# or socket watcher the calls $result_ready->send |
465 |
# when the "result" is ready. |
466 |
# in this case, we simply use a timer: |
467 |
my $w = AnyEvent->timer ( |
468 |
after => 1, |
469 |
cb => sub { $result_ready->send }, |
470 |
); |
471 |
|
472 |
# this "blocks" (while handling events) till the callback |
473 |
# calls send |
474 |
$result_ready->recv; |
475 |
|
476 |
Example: wait for a timer, but take advantage of the fact that |
477 |
condition variables are also code references. |
478 |
|
479 |
my $done = AnyEvent->condvar; |
480 |
my $delay = AnyEvent->timer (after => 5, cb => $done); |
481 |
$done->recv; |
482 |
|
483 |
Example: Imagine an API that returns a condvar and doesn't support |
484 |
callbacks. This is how you make a synchronous call, for example from |
485 |
the main program: |
486 |
|
487 |
use AnyEvent::CouchDB; |
488 |
|
489 |
... |
490 |
|
491 |
my @info = $couchdb->info->recv; |
492 |
|
493 |
And this is how you would just ste a callback to be called whenever the |
494 |
results are available: |
495 |
|
496 |
$couchdb->info->cb (sub { |
497 |
my @info = $_[0]->recv; |
498 |
}); |
499 |
|
500 |
=head3 METHODS FOR PRODUCERS |
501 |
|
502 |
These methods should only be used by the producing side, i.e. the |
503 |
code/module that eventually sends the signal. Note that it is also |
504 |
the producer side which creates the condvar in most cases, but it isn't |
505 |
uncommon for the consumer to create it as well. |
506 |
|
507 |
=over 4 |
508 |
|
509 |
=item $cv->send (...) |
510 |
|
511 |
Flag the condition as ready - a running C<< ->recv >> and all further |
512 |
calls to C<recv> will (eventually) return after this method has been |
513 |
called. If nobody is waiting the send will be remembered. |
514 |
|
515 |
If a callback has been set on the condition variable, it is called |
516 |
immediately from within send. |
517 |
|
518 |
Any arguments passed to the C<send> call will be returned by all |
519 |
future C<< ->recv >> calls. |
520 |
|
521 |
Condition variables are overloaded so one can call them directly |
522 |
(as a code reference). Calling them directly is the same as calling |
523 |
C<send>. Note, however, that many C-based event loops do not handle |
524 |
overloading, so as tempting as it may be, passing a condition variable |
525 |
instead of a callback does not work. Both the pure perl and EV loops |
526 |
support overloading, however, as well as all functions that use perl to |
527 |
invoke a callback (as in L<AnyEvent::Socket> and L<AnyEvent::DNS> for |
528 |
example). |
529 |
|
530 |
=item $cv->croak ($error) |
531 |
|
532 |
Similar to send, but causes all call's to C<< ->recv >> to invoke |
533 |
C<Carp::croak> with the given error message/object/scalar. |
534 |
|
535 |
This can be used to signal any errors to the condition variable |
536 |
user/consumer. |
537 |
|
538 |
=item $cv->begin ([group callback]) |
539 |
|
540 |
=item $cv->end |
541 |
|
542 |
These two methods are EXPERIMENTAL and MIGHT CHANGE. |
543 |
|
544 |
These two methods can be used to combine many transactions/events into |
545 |
one. For example, a function that pings many hosts in parallel might want |
546 |
to use a condition variable for the whole process. |
547 |
|
548 |
Every call to C<< ->begin >> will increment a counter, and every call to |
549 |
C<< ->end >> will decrement it. If the counter reaches C<0> in C<< ->end |
550 |
>>, the (last) callback passed to C<begin> will be executed. That callback |
551 |
is I<supposed> to call C<< ->send >>, but that is not required. If no |
552 |
callback was set, C<send> will be called without any arguments. |
553 |
|
554 |
Let's clarify this with the ping example: |
555 |
|
556 |
my $cv = AnyEvent->condvar; |
557 |
|
558 |
my %result; |
559 |
$cv->begin (sub { $cv->send (\%result) }); |
560 |
|
561 |
for my $host (@list_of_hosts) { |
562 |
$cv->begin; |
563 |
ping_host_then_call_callback $host, sub { |
564 |
$result{$host} = ...; |
565 |
$cv->end; |
566 |
}; |
567 |
} |
568 |
|
569 |
$cv->end; |
570 |
|
571 |
This code fragment supposedly pings a number of hosts and calls |
572 |
C<send> after results for all then have have been gathered - in any |
573 |
order. To achieve this, the code issues a call to C<begin> when it starts |
574 |
each ping request and calls C<end> when it has received some result for |
575 |
it. Since C<begin> and C<end> only maintain a counter, the order in which |
576 |
results arrive is not relevant. |
577 |
|
578 |
There is an additional bracketing call to C<begin> and C<end> outside the |
579 |
loop, which serves two important purposes: first, it sets the callback |
580 |
to be called once the counter reaches C<0>, and second, it ensures that |
581 |
C<send> is called even when C<no> hosts are being pinged (the loop |
582 |
doesn't execute once). |
583 |
|
584 |
This is the general pattern when you "fan out" into multiple subrequests: |
585 |
use an outer C<begin>/C<end> pair to set the callback and ensure C<end> |
586 |
is called at least once, and then, for each subrequest you start, call |
587 |
C<begin> and for each subrequest you finish, call C<end>. |
588 |
|
589 |
=back |
590 |
|
591 |
=head3 METHODS FOR CONSUMERS |
592 |
|
593 |
These methods should only be used by the consuming side, i.e. the |
594 |
code awaits the condition. |
595 |
|
596 |
=over 4 |
597 |
|
598 |
=item $cv->recv |
599 |
|
600 |
Wait (blocking if necessary) until the C<< ->send >> or C<< ->croak |
601 |
>> methods have been called on c<$cv>, while servicing other watchers |
602 |
normally. |
603 |
|
604 |
You can only wait once on a condition - additional calls are valid but |
605 |
will return immediately. |
606 |
|
607 |
If an error condition has been set by calling C<< ->croak >>, then this |
608 |
function will call C<croak>. |
609 |
|
610 |
In list context, all parameters passed to C<send> will be returned, |
611 |
in scalar context only the first one will be returned. |
612 |
|
613 |
Not all event models support a blocking wait - some die in that case |
614 |
(programs might want to do that to stay interactive), so I<if you are |
615 |
using this from a module, never require a blocking wait>, but let the |
616 |
caller decide whether the call will block or not (for example, by coupling |
617 |
condition variables with some kind of request results and supporting |
618 |
callbacks so the caller knows that getting the result will not block, |
619 |
while still supporting blocking waits if the caller so desires). |
620 |
|
621 |
Another reason I<never> to C<< ->recv >> in a module is that you cannot |
622 |
sensibly have two C<< ->recv >>'s in parallel, as that would require |
623 |
multiple interpreters or coroutines/threads, none of which C<AnyEvent> |
624 |
can supply. |
625 |
|
626 |
The L<Coro> module, however, I<can> and I<does> supply coroutines and, in |
627 |
fact, L<Coro::AnyEvent> replaces AnyEvent's condvars by coroutine-safe |
628 |
versions and also integrates coroutines into AnyEvent, making blocking |
629 |
C<< ->recv >> calls perfectly safe as long as they are done from another |
630 |
coroutine (one that doesn't run the event loop). |
631 |
|
632 |
You can ensure that C<< -recv >> never blocks by setting a callback and |
633 |
only calling C<< ->recv >> from within that callback (or at a later |
634 |
time). This will work even when the event loop does not support blocking |
635 |
waits otherwise. |
636 |
|
637 |
=item $bool = $cv->ready |
638 |
|
639 |
Returns true when the condition is "true", i.e. whether C<send> or |
640 |
C<croak> have been called. |
641 |
|
642 |
=item $cb = $cv->cb ($cb->($cv)) |
643 |
|
644 |
This is a mutator function that returns the callback set and optionally |
645 |
replaces it before doing so. |
646 |
|
647 |
The callback will be called when the condition becomes "true", i.e. when |
648 |
C<send> or C<croak> are called, with the only argument being the condition |
649 |
variable itself. Calling C<recv> inside the callback or at any later time |
650 |
is guaranteed not to block. |
651 |
|
652 |
=back |
653 |
|
654 |
=head1 GLOBAL VARIABLES AND FUNCTIONS |
655 |
|
656 |
=over 4 |
657 |
|
658 |
=item $AnyEvent::MODEL |
659 |
|
660 |
Contains C<undef> until the first watcher is being created. Then it |
661 |
contains the event model that is being used, which is the name of the |
662 |
Perl class implementing the model. This class is usually one of the |
663 |
C<AnyEvent::Impl:xxx> modules, but can be any other class in the case |
664 |
AnyEvent has been extended at runtime (e.g. in I<rxvt-unicode>). |
665 |
|
666 |
The known classes so far are: |
667 |
|
668 |
AnyEvent::Impl::EV based on EV (an interface to libev, best choice). |
669 |
AnyEvent::Impl::Event based on Event, second best choice. |
670 |
AnyEvent::Impl::Perl pure-perl implementation, fast and portable. |
671 |
AnyEvent::Impl::Glib based on Glib, third-best choice. |
672 |
AnyEvent::Impl::Tk based on Tk, very bad choice. |
673 |
AnyEvent::Impl::Qt based on Qt, cannot be autoprobed (see its docs). |
674 |
AnyEvent::Impl::EventLib based on Event::Lib, leaks memory and worse. |
675 |
AnyEvent::Impl::POE based on POE, not generic enough for full support. |
676 |
|
677 |
There is no support for WxWidgets, as WxWidgets has no support for |
678 |
watching file handles. However, you can use WxWidgets through the |
679 |
POE Adaptor, as POE has a Wx backend that simply polls 20 times per |
680 |
second, which was considered to be too horrible to even consider for |
681 |
AnyEvent. Likewise, other POE backends can be used by AnyEvent by using |
682 |
it's adaptor. |
683 |
|
684 |
AnyEvent knows about L<Prima> and L<Wx> and will try to use L<POE> when |
685 |
autodetecting them. |
686 |
|
687 |
=item AnyEvent::detect |
688 |
|
689 |
Returns C<$AnyEvent::MODEL>, forcing autodetection of the event model |
690 |
if necessary. You should only call this function right before you would |
691 |
have created an AnyEvent watcher anyway, that is, as late as possible at |
692 |
runtime. |
693 |
|
694 |
=item $guard = AnyEvent::post_detect { BLOCK } |
695 |
|
696 |
Arranges for the code block to be executed as soon as the event model is |
697 |
autodetected (or immediately if this has already happened). |
698 |
|
699 |
If called in scalar or list context, then it creates and returns an object |
700 |
that automatically removes the callback again when it is destroyed. See |
701 |
L<Coro::BDB> for a case where this is useful. |
702 |
|
703 |
=item @AnyEvent::post_detect |
704 |
|
705 |
If there are any code references in this array (you can C<push> to it |
706 |
before or after loading AnyEvent), then they will called directly after |
707 |
the event loop has been chosen. |
708 |
|
709 |
You should check C<$AnyEvent::MODEL> before adding to this array, though: |
710 |
if it contains a true value then the event loop has already been detected, |
711 |
and the array will be ignored. |
712 |
|
713 |
Best use C<AnyEvent::post_detect { BLOCK }> instead. |
714 |
|
715 |
=back |
716 |
|
717 |
=head1 WHAT TO DO IN A MODULE |
718 |
|
719 |
As a module author, you should C<use AnyEvent> and call AnyEvent methods |
720 |
freely, but you should not load a specific event module or rely on it. |
721 |
|
722 |
Be careful when you create watchers in the module body - AnyEvent will |
723 |
decide which event module to use as soon as the first method is called, so |
724 |
by calling AnyEvent in your module body you force the user of your module |
725 |
to load the event module first. |
726 |
|
727 |
Never call C<< ->recv >> on a condition variable unless you I<know> that |
728 |
the C<< ->send >> method has been called on it already. This is |
729 |
because it will stall the whole program, and the whole point of using |
730 |
events is to stay interactive. |
731 |
|
732 |
It is fine, however, to call C<< ->recv >> when the user of your module |
733 |
requests it (i.e. if you create a http request object ad have a method |
734 |
called C<results> that returns the results, it should call C<< ->recv >> |
735 |
freely, as the user of your module knows what she is doing. always). |
736 |
|
737 |
=head1 WHAT TO DO IN THE MAIN PROGRAM |
738 |
|
739 |
There will always be a single main program - the only place that should |
740 |
dictate which event model to use. |
741 |
|
742 |
If it doesn't care, it can just "use AnyEvent" and use it itself, or not |
743 |
do anything special (it does not need to be event-based) and let AnyEvent |
744 |
decide which implementation to chose if some module relies on it. |
745 |
|
746 |
If the main program relies on a specific event model - for example, in |
747 |
Gtk2 programs you have to rely on the Glib module - you should load the |
748 |
event module before loading AnyEvent or any module that uses it: generally |
749 |
speaking, you should load it as early as possible. The reason is that |
750 |
modules might create watchers when they are loaded, and AnyEvent will |
751 |
decide on the event model to use as soon as it creates watchers, and it |
752 |
might chose the wrong one unless you load the correct one yourself. |
753 |
|
754 |
You can chose to use a pure-perl implementation by loading the |
755 |
C<AnyEvent::Impl::Perl> module, which gives you similar behaviour |
756 |
everywhere, but letting AnyEvent chose the model is generally better. |
757 |
|
758 |
=head2 MAINLOOP EMULATION |
759 |
|
760 |
Sometimes (often for short test scripts, or even standalone programs who |
761 |
only want to use AnyEvent), you do not want to run a specific event loop. |
762 |
|
763 |
In that case, you can use a condition variable like this: |
764 |
|
765 |
AnyEvent->condvar->recv; |
766 |
|
767 |
This has the effect of entering the event loop and looping forever. |
768 |
|
769 |
Note that usually your program has some exit condition, in which case |
770 |
it is better to use the "traditional" approach of storing a condition |
771 |
variable somewhere, waiting for it, and sending it when the program should |
772 |
exit cleanly. |
773 |
|
774 |
|
775 |
=head1 OTHER MODULES |
776 |
|
777 |
The following is a non-exhaustive list of additional modules that use |
778 |
AnyEvent and can therefore be mixed easily with other AnyEvent modules |
779 |
in the same program. Some of the modules come with AnyEvent, some are |
780 |
available via CPAN. |
781 |
|
782 |
=over 4 |
783 |
|
784 |
=item L<AnyEvent::Util> |
785 |
|
786 |
Contains various utility functions that replace often-used but blocking |
787 |
functions such as C<inet_aton> by event-/callback-based versions. |
788 |
|
789 |
=item L<AnyEvent::Socket> |
790 |
|
791 |
Provides various utility functions for (internet protocol) sockets, |
792 |
addresses and name resolution. Also functions to create non-blocking tcp |
793 |
connections or tcp servers, with IPv6 and SRV record support and more. |
794 |
|
795 |
=item L<AnyEvent::Handle> |
796 |
|
797 |
Provide read and write buffers, manages watchers for reads and writes, |
798 |
supports raw and formatted I/O, I/O queued and fully transparent and |
799 |
non-blocking SSL/TLS. |
800 |
|
801 |
=item L<AnyEvent::DNS> |
802 |
|
803 |
Provides rich asynchronous DNS resolver capabilities. |
804 |
|
805 |
=item L<AnyEvent::HTTP> |
806 |
|
807 |
A simple-to-use HTTP library that is capable of making a lot of concurrent |
808 |
HTTP requests. |
809 |
|
810 |
=item L<AnyEvent::HTTPD> |
811 |
|
812 |
Provides a simple web application server framework. |
813 |
|
814 |
=item L<AnyEvent::FastPing> |
815 |
|
816 |
The fastest ping in the west. |
817 |
|
818 |
=item L<AnyEvent::DBI> |
819 |
|
820 |
Executes L<DBI> requests asynchronously in a proxy process. |
821 |
|
822 |
=item L<AnyEvent::AIO> |
823 |
|
824 |
Truly asynchronous I/O, should be in the toolbox of every event |
825 |
programmer. AnyEvent::AIO transparently fuses L<IO::AIO> and AnyEvent |
826 |
together. |
827 |
|
828 |
=item L<AnyEvent::BDB> |
829 |
|
830 |
Truly asynchronous Berkeley DB access. AnyEvent::BDB transparently fuses |
831 |
L<BDB> and AnyEvent together. |
832 |
|
833 |
=item L<AnyEvent::GPSD> |
834 |
|
835 |
A non-blocking interface to gpsd, a daemon delivering GPS information. |
836 |
|
837 |
=item L<AnyEvent::IGS> |
838 |
|
839 |
A non-blocking interface to the Internet Go Server protocol (used by |
840 |
L<App::IGS>). |
841 |
|
842 |
=item L<AnyEvent::IRC> |
843 |
|
844 |
AnyEvent based IRC client module family (replacing the older Net::IRC3). |
845 |
|
846 |
=item L<Net::XMPP2> |
847 |
|
848 |
AnyEvent based XMPP (Jabber protocol) module family. |
849 |
|
850 |
=item L<Net::FCP> |
851 |
|
852 |
AnyEvent-based implementation of the Freenet Client Protocol, birthplace |
853 |
of AnyEvent. |
854 |
|
855 |
=item L<Event::ExecFlow> |
856 |
|
857 |
High level API for event-based execution flow control. |
858 |
|
859 |
=item L<Coro> |
860 |
|
861 |
Has special support for AnyEvent via L<Coro::AnyEvent>. |
862 |
|
863 |
=item L<IO::Lambda> |
864 |
|
865 |
The lambda approach to I/O - don't ask, look there. Can use AnyEvent. |
866 |
|
867 |
=back |
868 |
|
869 |
=cut |
870 |
|
871 |
package AnyEvent; |
872 |
|
873 |
no warnings; |
874 |
use strict qw(vars subs); |
875 |
|
876 |
use Carp; |
877 |
|
878 |
our $VERSION = 4.35; |
879 |
our $MODEL; |
880 |
|
881 |
our $AUTOLOAD; |
882 |
our @ISA; |
883 |
|
884 |
our @REGISTRY; |
885 |
|
886 |
our $WIN32; |
887 |
|
888 |
BEGIN { |
889 |
my $win32 = ! ! ($^O =~ /mswin32/i); |
890 |
eval "sub WIN32(){ $win32 }"; |
891 |
} |
892 |
|
893 |
our $verbose = $ENV{PERL_ANYEVENT_VERBOSE}*1; |
894 |
|
895 |
our %PROTOCOL; # (ipv4|ipv6) => (1|2), higher numbers are preferred |
896 |
|
897 |
{ |
898 |
my $idx; |
899 |
$PROTOCOL{$_} = ++$idx |
900 |
for reverse split /\s*,\s*/, |
901 |
$ENV{PERL_ANYEVENT_PROTOCOLS} || "ipv4,ipv6"; |
902 |
} |
903 |
|
904 |
my @models = ( |
905 |
[EV:: => AnyEvent::Impl::EV::], |
906 |
[Event:: => AnyEvent::Impl::Event::], |
907 |
[AnyEvent::Impl::Perl:: => AnyEvent::Impl::Perl::], |
908 |
# everything below here will not be autoprobed |
909 |
# as the pureperl backend should work everywhere |
910 |
# and is usually faster |
911 |
[Tk:: => AnyEvent::Impl::Tk::], # crashes with many handles |
912 |
[Glib:: => AnyEvent::Impl::Glib::], # becomes extremely slow with many watchers |
913 |
[Event::Lib:: => AnyEvent::Impl::EventLib::], # too buggy |
914 |
[Qt:: => AnyEvent::Impl::Qt::], # requires special main program |
915 |
[POE::Kernel:: => AnyEvent::Impl::POE::], # lasciate ogni speranza |
916 |
[Wx:: => AnyEvent::Impl::POE::], |
917 |
[Prima:: => AnyEvent::Impl::POE::], |
918 |
); |
919 |
|
920 |
our %method = map +($_ => 1), qw(io timer time now signal child condvar one_event DESTROY); |
921 |
|
922 |
our @post_detect; |
923 |
|
924 |
sub post_detect(&) { |
925 |
my ($cb) = @_; |
926 |
|
927 |
if ($MODEL) { |
928 |
$cb->(); |
929 |
|
930 |
1 |
931 |
} else { |
932 |
push @post_detect, $cb; |
933 |
|
934 |
defined wantarray |
935 |
? bless \$cb, "AnyEvent::Util::PostDetect" |
936 |
: () |
937 |
} |
938 |
} |
939 |
|
940 |
sub AnyEvent::Util::PostDetect::DESTROY { |
941 |
@post_detect = grep $_ != ${$_[0]}, @post_detect; |
942 |
} |
943 |
|
944 |
sub detect() { |
945 |
unless ($MODEL) { |
946 |
no strict 'refs'; |
947 |
local $SIG{__DIE__}; |
948 |
|
949 |
if ($ENV{PERL_ANYEVENT_MODEL} =~ /^([a-zA-Z]+)$/) { |
950 |
my $model = "AnyEvent::Impl::$1"; |
951 |
if (eval "require $model") { |
952 |
$MODEL = $model; |
953 |
warn "AnyEvent: loaded model '$model' (forced by \$PERL_ANYEVENT_MODEL), using it.\n" if $verbose > 1; |
954 |
} else { |
955 |
warn "AnyEvent: unable to load model '$model' (from \$PERL_ANYEVENT_MODEL):\n$@" if $verbose; |
956 |
} |
957 |
} |
958 |
|
959 |
# check for already loaded models |
960 |
unless ($MODEL) { |
961 |
for (@REGISTRY, @models) { |
962 |
my ($package, $model) = @$_; |
963 |
if (${"$package\::VERSION"} > 0) { |
964 |
if (eval "require $model") { |
965 |
$MODEL = $model; |
966 |
warn "AnyEvent: autodetected model '$model', using it.\n" if $verbose > 1; |
967 |
last; |
968 |
} |
969 |
} |
970 |
} |
971 |
|
972 |
unless ($MODEL) { |
973 |
# try to load a model |
974 |
|
975 |
for (@REGISTRY, @models) { |
976 |
my ($package, $model) = @$_; |
977 |
if (eval "require $package" |
978 |
and ${"$package\::VERSION"} > 0 |
979 |
and eval "require $model") { |
980 |
$MODEL = $model; |
981 |
warn "AnyEvent: autoprobed model '$model', using it.\n" if $verbose > 1; |
982 |
last; |
983 |
} |
984 |
} |
985 |
|
986 |
$MODEL |
987 |
or die "No event module selected for AnyEvent and autodetect failed. Install any one of these modules: EV, Event or Glib."; |
988 |
} |
989 |
} |
990 |
|
991 |
push @{"$MODEL\::ISA"}, "AnyEvent::Base"; |
992 |
|
993 |
unshift @ISA, $MODEL; |
994 |
|
995 |
require AnyEvent::Strict if $ENV{PERL_ANYEVENT_STRICT}; |
996 |
|
997 |
(shift @post_detect)->() while @post_detect; |
998 |
} |
999 |
|
1000 |
$MODEL |
1001 |
} |
1002 |
|
1003 |
sub AUTOLOAD { |
1004 |
(my $func = $AUTOLOAD) =~ s/.*://; |
1005 |
|
1006 |
$method{$func} |
1007 |
or croak "$func: not a valid method for AnyEvent objects"; |
1008 |
|
1009 |
detect unless $MODEL; |
1010 |
|
1011 |
my $class = shift; |
1012 |
$class->$func (@_); |
1013 |
} |
1014 |
|
1015 |
# utility function to dup a filehandle. this is used by many backends |
1016 |
# to support binding more than one watcher per filehandle (they usually |
1017 |
# allow only one watcher per fd, so we dup it to get a different one). |
1018 |
sub _dupfh($$$$) { |
1019 |
my ($poll, $fh, $r, $w) = @_; |
1020 |
|
1021 |
# cygwin requires the fh mode to be matching, unix doesn't |
1022 |
my ($rw, $mode) = $poll eq "r" ? ($r, "<") |
1023 |
: $poll eq "w" ? ($w, ">") |
1024 |
: Carp::croak "AnyEvent->io requires poll set to either 'r' or 'w'"; |
1025 |
|
1026 |
open my $fh2, "$mode&" . fileno $fh |
1027 |
or die "cannot dup() filehandle: $!"; |
1028 |
|
1029 |
# we assume CLOEXEC is already set by perl in all important cases |
1030 |
|
1031 |
($fh2, $rw) |
1032 |
} |
1033 |
|
1034 |
package AnyEvent::Base; |
1035 |
|
1036 |
# default implementation for now and time |
1037 |
|
1038 |
BEGIN { |
1039 |
if (eval "use Time::HiRes (); time (); 1") { |
1040 |
*_time = \&Time::HiRes::time; |
1041 |
# if (eval "use POSIX (); (POSIX::times())... |
1042 |
} else { |
1043 |
*_time = sub { time }; # epic fail |
1044 |
} |
1045 |
} |
1046 |
|
1047 |
sub time { _time } |
1048 |
sub now { _time } |
1049 |
|
1050 |
# default implementation for ->condvar |
1051 |
|
1052 |
sub condvar { |
1053 |
bless { @_ == 3 ? (_ae_cb => $_[2]) : () }, AnyEvent::CondVar:: |
1054 |
} |
1055 |
|
1056 |
# default implementation for ->signal |
1057 |
|
1058 |
our ($SIGPIPE_R, $SIGPIPE_W, %SIG_CB, %SIG_EV, $SIG_IO); |
1059 |
|
1060 |
sub _signal_exec { |
1061 |
sysread $SIGPIPE_R, my $dummy, 4; |
1062 |
|
1063 |
while (%SIG_EV) { |
1064 |
for (keys %SIG_EV) { |
1065 |
delete $SIG_EV{$_}; |
1066 |
$_->() for values %{ $SIG_CB{$_} || {} }; |
1067 |
} |
1068 |
} |
1069 |
} |
1070 |
|
1071 |
sub signal { |
1072 |
my (undef, %arg) = @_; |
1073 |
|
1074 |
unless ($SIGPIPE_R) { |
1075 |
if (AnyEvent::WIN32) { |
1076 |
($SIGPIPE_R, $SIGPIPE_W) = AnyEvent::Util::portable_pipe (); |
1077 |
AnyEvent::Util::fh_nonblocking ($SIGPIPE_R) if $SIGPIPE_R; |
1078 |
AnyEvent::Util::fh_nonblocking ($SIGPIPE_W) if $SIGPIPE_W; # just in case |
1079 |
} else { |
1080 |
pipe $SIGPIPE_R, $SIGPIPE_W; |
1081 |
require Fcntl; |
1082 |
fcntl $SIGPIPE_R, &Fcntl::F_SETFL, &Fcntl::O_NONBLOCK if $SIGPIPE_R; |
1083 |
fcntl $SIGPIPE_W, &Fcntl::F_SETFL, &Fcntl::O_NONBLOCK if $SIGPIPE_W; # just in case |
1084 |
} |
1085 |
|
1086 |
$SIGPIPE_R |
1087 |
or Carp::croak "AnyEvent: unable to create a signal reporting pipe: $!\n"; |
1088 |
|
1089 |
$SIG_IO = AnyEvent->io (fh => $SIGPIPE_R, poll => "r", cb => \&_signal_exec); |
1090 |
} |
1091 |
|
1092 |
my $signal = uc $arg{signal} |
1093 |
or Carp::croak "required option 'signal' is missing"; |
1094 |
|
1095 |
$SIG_CB{$signal}{$arg{cb}} = $arg{cb}; |
1096 |
$SIG{$signal} ||= sub { |
1097 |
syswrite $SIGPIPE_W, "\x00", 1 unless %SIG_EV; |
1098 |
undef $SIG_EV{$signal}; |
1099 |
}; |
1100 |
|
1101 |
bless [$signal, $arg{cb}], "AnyEvent::Base::Signal" |
1102 |
} |
1103 |
|
1104 |
sub AnyEvent::Base::Signal::DESTROY { |
1105 |
my ($signal, $cb) = @{$_[0]}; |
1106 |
|
1107 |
delete $SIG_CB{$signal}{$cb}; |
1108 |
|
1109 |
delete $SIG{$signal} unless keys %{ $SIG_CB{$signal} }; |
1110 |
} |
1111 |
|
1112 |
# default implementation for ->child |
1113 |
|
1114 |
our %PID_CB; |
1115 |
our $CHLD_W; |
1116 |
our $CHLD_DELAY_W; |
1117 |
our $PID_IDLE; |
1118 |
our $WNOHANG; |
1119 |
|
1120 |
sub _child_wait { |
1121 |
while (0 < (my $pid = waitpid -1, $WNOHANG)) { |
1122 |
$_->($pid, $?) for (values %{ $PID_CB{$pid} || {} }), |
1123 |
(values %{ $PID_CB{0} || {} }); |
1124 |
} |
1125 |
|
1126 |
undef $PID_IDLE; |
1127 |
} |
1128 |
|
1129 |
sub _sigchld { |
1130 |
# make sure we deliver these changes "synchronous" with the event loop. |
1131 |
$CHLD_DELAY_W ||= AnyEvent->timer (after => 0, cb => sub { |
1132 |
undef $CHLD_DELAY_W; |
1133 |
&_child_wait; |
1134 |
}); |
1135 |
} |
1136 |
|
1137 |
sub child { |
1138 |
my (undef, %arg) = @_; |
1139 |
|
1140 |
defined (my $pid = $arg{pid} + 0) |
1141 |
or Carp::croak "required option 'pid' is missing"; |
1142 |
|
1143 |
$PID_CB{$pid}{$arg{cb}} = $arg{cb}; |
1144 |
|
1145 |
unless ($WNOHANG) { |
1146 |
$WNOHANG = eval { local $SIG{__DIE__}; require POSIX; &POSIX::WNOHANG } || 1; |
1147 |
} |
1148 |
|
1149 |
unless ($CHLD_W) { |
1150 |
$CHLD_W = AnyEvent->signal (signal => 'CHLD', cb => \&_sigchld); |
1151 |
# child could be a zombie already, so make at least one round |
1152 |
&_sigchld; |
1153 |
} |
1154 |
|
1155 |
bless [$pid, $arg{cb}], "AnyEvent::Base::Child" |
1156 |
} |
1157 |
|
1158 |
sub AnyEvent::Base::Child::DESTROY { |
1159 |
my ($pid, $cb) = @{$_[0]}; |
1160 |
|
1161 |
delete $PID_CB{$pid}{$cb}; |
1162 |
delete $PID_CB{$pid} unless keys %{ $PID_CB{$pid} }; |
1163 |
|
1164 |
undef $CHLD_W unless keys %PID_CB; |
1165 |
} |
1166 |
|
1167 |
package AnyEvent::CondVar; |
1168 |
|
1169 |
our @ISA = AnyEvent::CondVar::Base::; |
1170 |
|
1171 |
package AnyEvent::CondVar::Base; |
1172 |
|
1173 |
use overload |
1174 |
'&{}' => sub { my $self = shift; sub { $self->send (@_) } }, |
1175 |
fallback => 1; |
1176 |
|
1177 |
sub _send { |
1178 |
# nop |
1179 |
} |
1180 |
|
1181 |
sub send { |
1182 |
my $cv = shift; |
1183 |
$cv->{_ae_sent} = [@_]; |
1184 |
(delete $cv->{_ae_cb})->($cv) if $cv->{_ae_cb}; |
1185 |
$cv->_send; |
1186 |
} |
1187 |
|
1188 |
sub croak { |
1189 |
$_[0]{_ae_croak} = $_[1]; |
1190 |
$_[0]->send; |
1191 |
} |
1192 |
|
1193 |
sub ready { |
1194 |
$_[0]{_ae_sent} |
1195 |
} |
1196 |
|
1197 |
sub _wait { |
1198 |
AnyEvent->one_event while !$_[0]{_ae_sent}; |
1199 |
} |
1200 |
|
1201 |
sub recv { |
1202 |
$_[0]->_wait; |
1203 |
|
1204 |
Carp::croak $_[0]{_ae_croak} if $_[0]{_ae_croak}; |
1205 |
wantarray ? @{ $_[0]{_ae_sent} } : $_[0]{_ae_sent}[0] |
1206 |
} |
1207 |
|
1208 |
sub cb { |
1209 |
$_[0]{_ae_cb} = $_[1] if @_ > 1; |
1210 |
$_[0]{_ae_cb} |
1211 |
} |
1212 |
|
1213 |
sub begin { |
1214 |
++$_[0]{_ae_counter}; |
1215 |
$_[0]{_ae_end_cb} = $_[1] if @_ > 1; |
1216 |
} |
1217 |
|
1218 |
sub end { |
1219 |
return if --$_[0]{_ae_counter}; |
1220 |
&{ $_[0]{_ae_end_cb} || sub { $_[0]->send } }; |
1221 |
} |
1222 |
|
1223 |
# undocumented/compatibility with pre-3.4 |
1224 |
*broadcast = \&send; |
1225 |
*wait = \&_wait; |
1226 |
|
1227 |
=head1 ERROR AND EXCEPTION HANDLING |
1228 |
|
1229 |
In general, AnyEvent does not do any error handling - it relies on the |
1230 |
caller to do that if required. The L<AnyEvent::Strict> module (see also |
1231 |
the C<PERL_ANYEVENT_STRICT> environment variable, below) provides strict |
1232 |
checking of all AnyEvent methods, however, which is highly useful during |
1233 |
development. |
1234 |
|
1235 |
As for exception handling (i.e. runtime errors and exceptions thrown while |
1236 |
executing a callback), this is not only highly event-loop specific, but |
1237 |
also not in any way wrapped by this module, as this is the job of the main |
1238 |
program. |
1239 |
|
1240 |
The pure perl event loop simply re-throws the exception (usually |
1241 |
within C<< condvar->recv >>), the L<Event> and L<EV> modules call C<< |
1242 |
$Event/EV::DIED->() >>, L<Glib> uses C<< install_exception_handler >> and |
1243 |
so on. |
1244 |
|
1245 |
=head1 ENVIRONMENT VARIABLES |
1246 |
|
1247 |
The following environment variables are used by this module or its |
1248 |
submodules: |
1249 |
|
1250 |
=over 4 |
1251 |
|
1252 |
=item C<PERL_ANYEVENT_VERBOSE> |
1253 |
|
1254 |
By default, AnyEvent will be completely silent except in fatal |
1255 |
conditions. You can set this environment variable to make AnyEvent more |
1256 |
talkative. |
1257 |
|
1258 |
When set to C<1> or higher, causes AnyEvent to warn about unexpected |
1259 |
conditions, such as not being able to load the event model specified by |
1260 |
C<PERL_ANYEVENT_MODEL>. |
1261 |
|
1262 |
When set to C<2> or higher, cause AnyEvent to report to STDERR which event |
1263 |
model it chooses. |
1264 |
|
1265 |
=item C<PERL_ANYEVENT_STRICT> |
1266 |
|
1267 |
AnyEvent does not do much argument checking by default, as thorough |
1268 |
argument checking is very costly. Setting this variable to a true value |
1269 |
will cause AnyEvent to load C<AnyEvent::Strict> and then to thoroughly |
1270 |
check the arguments passed to most method calls. If it finds any problems |
1271 |
it will croak. |
1272 |
|
1273 |
In other words, enables "strict" mode. |
1274 |
|
1275 |
Unlike C<use strict>, it is definitely recommended ot keep it off in |
1276 |
production. Keeping C<PERL_ANYEVENT_STRICT=1> in your environment while |
1277 |
developing programs can be very useful, however. |
1278 |
|
1279 |
=item C<PERL_ANYEVENT_MODEL> |
1280 |
|
1281 |
This can be used to specify the event model to be used by AnyEvent, before |
1282 |
auto detection and -probing kicks in. It must be a string consisting |
1283 |
entirely of ASCII letters. The string C<AnyEvent::Impl::> gets prepended |
1284 |
and the resulting module name is loaded and if the load was successful, |
1285 |
used as event model. If it fails to load AnyEvent will proceed with |
1286 |
auto detection and -probing. |
1287 |
|
1288 |
This functionality might change in future versions. |
1289 |
|
1290 |
For example, to force the pure perl model (L<AnyEvent::Impl::Perl>) you |
1291 |
could start your program like this: |
1292 |
|
1293 |
PERL_ANYEVENT_MODEL=Perl perl ... |
1294 |
|
1295 |
=item C<PERL_ANYEVENT_PROTOCOLS> |
1296 |
|
1297 |
Used by both L<AnyEvent::DNS> and L<AnyEvent::Socket> to determine preferences |
1298 |
for IPv4 or IPv6. The default is unspecified (and might change, or be the result |
1299 |
of auto probing). |
1300 |
|
1301 |
Must be set to a comma-separated list of protocols or address families, |
1302 |
current supported: C<ipv4> and C<ipv6>. Only protocols mentioned will be |
1303 |
used, and preference will be given to protocols mentioned earlier in the |
1304 |
list. |
1305 |
|
1306 |
This variable can effectively be used for denial-of-service attacks |
1307 |
against local programs (e.g. when setuid), although the impact is likely |
1308 |
small, as the program has to handle conenction and other failures anyways. |
1309 |
|
1310 |
Examples: C<PERL_ANYEVENT_PROTOCOLS=ipv4,ipv6> - prefer IPv4 over IPv6, |
1311 |
but support both and try to use both. C<PERL_ANYEVENT_PROTOCOLS=ipv4> |
1312 |
- only support IPv4, never try to resolve or contact IPv6 |
1313 |
addresses. C<PERL_ANYEVENT_PROTOCOLS=ipv6,ipv4> support either IPv4 or |
1314 |
IPv6, but prefer IPv6 over IPv4. |
1315 |
|
1316 |
=item C<PERL_ANYEVENT_EDNS0> |
1317 |
|
1318 |
Used by L<AnyEvent::DNS> to decide whether to use the EDNS0 extension |
1319 |
for DNS. This extension is generally useful to reduce DNS traffic, but |
1320 |
some (broken) firewalls drop such DNS packets, which is why it is off by |
1321 |
default. |
1322 |
|
1323 |
Setting this variable to C<1> will cause L<AnyEvent::DNS> to announce |
1324 |
EDNS0 in its DNS requests. |
1325 |
|
1326 |
=item C<PERL_ANYEVENT_MAX_FORKS> |
1327 |
|
1328 |
The maximum number of child processes that C<AnyEvent::Util::fork_call> |
1329 |
will create in parallel. |
1330 |
|
1331 |
=back |
1332 |
|
1333 |
=head1 SUPPLYING YOUR OWN EVENT MODEL INTERFACE |
1334 |
|
1335 |
This is an advanced topic that you do not normally need to use AnyEvent in |
1336 |
a module. This section is only of use to event loop authors who want to |
1337 |
provide AnyEvent compatibility. |
1338 |
|
1339 |
If you need to support another event library which isn't directly |
1340 |
supported by AnyEvent, you can supply your own interface to it by |
1341 |
pushing, before the first watcher gets created, the package name of |
1342 |
the event module and the package name of the interface to use onto |
1343 |
C<@AnyEvent::REGISTRY>. You can do that before and even without loading |
1344 |
AnyEvent, so it is reasonably cheap. |
1345 |
|
1346 |
Example: |
1347 |
|
1348 |
push @AnyEvent::REGISTRY, [urxvt => urxvt::anyevent::]; |
1349 |
|
1350 |
This tells AnyEvent to (literally) use the C<urxvt::anyevent::> |
1351 |
package/class when it finds the C<urxvt> package/module is already loaded. |
1352 |
|
1353 |
When AnyEvent is loaded and asked to find a suitable event model, it |
1354 |
will first check for the presence of urxvt by trying to C<use> the |
1355 |
C<urxvt::anyevent> module. |
1356 |
|
1357 |
The class should provide implementations for all watcher types. See |
1358 |
L<AnyEvent::Impl::EV> (source code), L<AnyEvent::Impl::Glib> (Source code) |
1359 |
and so on for actual examples. Use C<perldoc -m AnyEvent::Impl::Glib> to |
1360 |
see the sources. |
1361 |
|
1362 |
If you don't provide C<signal> and C<child> watchers than AnyEvent will |
1363 |
provide suitable (hopefully) replacements. |
1364 |
|
1365 |
The above example isn't fictitious, the I<rxvt-unicode> (a.k.a. urxvt) |
1366 |
terminal emulator uses the above line as-is. An interface isn't included |
1367 |
in AnyEvent because it doesn't make sense outside the embedded interpreter |
1368 |
inside I<rxvt-unicode>, and it is updated and maintained as part of the |
1369 |
I<rxvt-unicode> distribution. |
1370 |
|
1371 |
I<rxvt-unicode> also cheats a bit by not providing blocking access to |
1372 |
condition variables: code blocking while waiting for a condition will |
1373 |
C<die>. This still works with most modules/usages, and blocking calls must |
1374 |
not be done in an interactive application, so it makes sense. |
1375 |
|
1376 |
=head1 EXAMPLE PROGRAM |
1377 |
|
1378 |
The following program uses an I/O watcher to read data from STDIN, a timer |
1379 |
to display a message once per second, and a condition variable to quit the |
1380 |
program when the user enters quit: |
1381 |
|
1382 |
use AnyEvent; |
1383 |
|
1384 |
my $cv = AnyEvent->condvar; |
1385 |
|
1386 |
my $io_watcher = AnyEvent->io ( |
1387 |
fh => \*STDIN, |
1388 |
poll => 'r', |
1389 |
cb => sub { |
1390 |
warn "io event <$_[0]>\n"; # will always output <r> |
1391 |
chomp (my $input = <STDIN>); # read a line |
1392 |
warn "read: $input\n"; # output what has been read |
1393 |
$cv->send if $input =~ /^q/i; # quit program if /^q/i |
1394 |
}, |
1395 |
); |
1396 |
|
1397 |
my $time_watcher; # can only be used once |
1398 |
|
1399 |
sub new_timer { |
1400 |
$timer = AnyEvent->timer (after => 1, cb => sub { |
1401 |
warn "timeout\n"; # print 'timeout' about every second |
1402 |
&new_timer; # and restart the time |
1403 |
}); |
1404 |
} |
1405 |
|
1406 |
new_timer; # create first timer |
1407 |
|
1408 |
$cv->recv; # wait until user enters /^q/i |
1409 |
|
1410 |
=head1 REAL-WORLD EXAMPLE |
1411 |
|
1412 |
Consider the L<Net::FCP> module. It features (among others) the following |
1413 |
API calls, which are to freenet what HTTP GET requests are to http: |
1414 |
|
1415 |
my $data = $fcp->client_get ($url); # blocks |
1416 |
|
1417 |
my $transaction = $fcp->txn_client_get ($url); # does not block |
1418 |
$transaction->cb ( sub { ... } ); # set optional result callback |
1419 |
my $data = $transaction->result; # possibly blocks |
1420 |
|
1421 |
The C<client_get> method works like C<LWP::Simple::get>: it requests the |
1422 |
given URL and waits till the data has arrived. It is defined to be: |
1423 |
|
1424 |
sub client_get { $_[0]->txn_client_get ($_[1])->result } |
1425 |
|
1426 |
And in fact is automatically generated. This is the blocking API of |
1427 |
L<Net::FCP>, and it works as simple as in any other, similar, module. |
1428 |
|
1429 |
More complicated is C<txn_client_get>: It only creates a transaction |
1430 |
(completion, result, ...) object and initiates the transaction. |
1431 |
|
1432 |
my $txn = bless { }, Net::FCP::Txn::; |
1433 |
|
1434 |
It also creates a condition variable that is used to signal the completion |
1435 |
of the request: |
1436 |
|
1437 |
$txn->{finished} = AnyAvent->condvar; |
1438 |
|
1439 |
It then creates a socket in non-blocking mode. |
1440 |
|
1441 |
socket $txn->{fh}, ...; |
1442 |
fcntl $txn->{fh}, F_SETFL, O_NONBLOCK; |
1443 |
connect $txn->{fh}, ... |
1444 |
and !$!{EWOULDBLOCK} |
1445 |
and !$!{EINPROGRESS} |
1446 |
and Carp::croak "unable to connect: $!\n"; |
1447 |
|
1448 |
Then it creates a write-watcher which gets called whenever an error occurs |
1449 |
or the connection succeeds: |
1450 |
|
1451 |
$txn->{w} = AnyEvent->io (fh => $txn->{fh}, poll => 'w', cb => sub { $txn->fh_ready_w }); |
1452 |
|
1453 |
And returns this transaction object. The C<fh_ready_w> callback gets |
1454 |
called as soon as the event loop detects that the socket is ready for |
1455 |
writing. |
1456 |
|
1457 |
The C<fh_ready_w> method makes the socket blocking again, writes the |
1458 |
request data and replaces the watcher by a read watcher (waiting for reply |
1459 |
data). The actual code is more complicated, but that doesn't matter for |
1460 |
this example: |
1461 |
|
1462 |
fcntl $txn->{fh}, F_SETFL, 0; |
1463 |
syswrite $txn->{fh}, $txn->{request} |
1464 |
or die "connection or write error"; |
1465 |
$txn->{w} = AnyEvent->io (fh => $txn->{fh}, poll => 'r', cb => sub { $txn->fh_ready_r }); |
1466 |
|
1467 |
Again, C<fh_ready_r> waits till all data has arrived, and then stores the |
1468 |
result and signals any possible waiters that the request has finished: |
1469 |
|
1470 |
sysread $txn->{fh}, $txn->{buf}, length $txn->{$buf}; |
1471 |
|
1472 |
if (end-of-file or data complete) { |
1473 |
$txn->{result} = $txn->{buf}; |
1474 |
$txn->{finished}->send; |
1475 |
$txb->{cb}->($txn) of $txn->{cb}; # also call callback |
1476 |
} |
1477 |
|
1478 |
The C<result> method, finally, just waits for the finished signal (if the |
1479 |
request was already finished, it doesn't wait, of course, and returns the |
1480 |
data: |
1481 |
|
1482 |
$txn->{finished}->recv; |
1483 |
return $txn->{result}; |
1484 |
|
1485 |
The actual code goes further and collects all errors (C<die>s, exceptions) |
1486 |
that occurred during request processing. The C<result> method detects |
1487 |
whether an exception as thrown (it is stored inside the $txn object) |
1488 |
and just throws the exception, which means connection errors and other |
1489 |
problems get reported tot he code that tries to use the result, not in a |
1490 |
random callback. |
1491 |
|
1492 |
All of this enables the following usage styles: |
1493 |
|
1494 |
1. Blocking: |
1495 |
|
1496 |
my $data = $fcp->client_get ($url); |
1497 |
|
1498 |
2. Blocking, but running in parallel: |
1499 |
|
1500 |
my @datas = map $_->result, |
1501 |
map $fcp->txn_client_get ($_), |
1502 |
@urls; |
1503 |
|
1504 |
Both blocking examples work without the module user having to know |
1505 |
anything about events. |
1506 |
|
1507 |
3a. Event-based in a main program, using any supported event module: |
1508 |
|
1509 |
use EV; |
1510 |
|
1511 |
$fcp->txn_client_get ($url)->cb (sub { |
1512 |
my $txn = shift; |
1513 |
my $data = $txn->result; |
1514 |
... |
1515 |
}); |
1516 |
|
1517 |
EV::loop; |
1518 |
|
1519 |
3b. The module user could use AnyEvent, too: |
1520 |
|
1521 |
use AnyEvent; |
1522 |
|
1523 |
my $quit = AnyEvent->condvar; |
1524 |
|
1525 |
$fcp->txn_client_get ($url)->cb (sub { |
1526 |
... |
1527 |
$quit->send; |
1528 |
}); |
1529 |
|
1530 |
$quit->recv; |
1531 |
|
1532 |
|
1533 |
=head1 BENCHMARKS |
1534 |
|
1535 |
To give you an idea of the performance and overheads that AnyEvent adds |
1536 |
over the event loops themselves and to give you an impression of the speed |
1537 |
of various event loops I prepared some benchmarks. |
1538 |
|
1539 |
=head2 BENCHMARKING ANYEVENT OVERHEAD |
1540 |
|
1541 |
Here is a benchmark of various supported event models used natively and |
1542 |
through AnyEvent. The benchmark creates a lot of timers (with a zero |
1543 |
timeout) and I/O watchers (watching STDOUT, a pty, to become writable, |
1544 |
which it is), lets them fire exactly once and destroys them again. |
1545 |
|
1546 |
Source code for this benchmark is found as F<eg/bench> in the AnyEvent |
1547 |
distribution. |
1548 |
|
1549 |
=head3 Explanation of the columns |
1550 |
|
1551 |
I<watcher> is the number of event watchers created/destroyed. Since |
1552 |
different event models feature vastly different performances, each event |
1553 |
loop was given a number of watchers so that overall runtime is acceptable |
1554 |
and similar between tested event loop (and keep them from crashing): Glib |
1555 |
would probably take thousands of years if asked to process the same number |
1556 |
of watchers as EV in this benchmark. |
1557 |
|
1558 |
I<bytes> is the number of bytes (as measured by the resident set size, |
1559 |
RSS) consumed by each watcher. This method of measuring captures both C |
1560 |
and Perl-based overheads. |
1561 |
|
1562 |
I<create> is the time, in microseconds (millionths of seconds), that it |
1563 |
takes to create a single watcher. The callback is a closure shared between |
1564 |
all watchers, to avoid adding memory overhead. That means closure creation |
1565 |
and memory usage is not included in the figures. |
1566 |
|
1567 |
I<invoke> is the time, in microseconds, used to invoke a simple |
1568 |
callback. The callback simply counts down a Perl variable and after it was |
1569 |
invoked "watcher" times, it would C<< ->send >> a condvar once to |
1570 |
signal the end of this phase. |
1571 |
|
1572 |
I<destroy> is the time, in microseconds, that it takes to destroy a single |
1573 |
watcher. |
1574 |
|
1575 |
=head3 Results |
1576 |
|
1577 |
name watchers bytes create invoke destroy comment |
1578 |
EV/EV 400000 224 0.47 0.35 0.27 EV native interface |
1579 |
EV/Any 100000 224 2.88 0.34 0.27 EV + AnyEvent watchers |
1580 |
CoroEV/Any 100000 224 2.85 0.35 0.28 coroutines + Coro::Signal |
1581 |
Perl/Any 100000 452 4.13 0.73 0.95 pure perl implementation |
1582 |
Event/Event 16000 517 32.20 31.80 0.81 Event native interface |
1583 |
Event/Any 16000 590 35.85 31.55 1.06 Event + AnyEvent watchers |
1584 |
Glib/Any 16000 1357 102.33 12.31 51.00 quadratic behaviour |
1585 |
Tk/Any 2000 1860 27.20 66.31 14.00 SEGV with >> 2000 watchers |
1586 |
POE/Event 2000 6328 109.99 751.67 14.02 via POE::Loop::Event |
1587 |
POE/Select 2000 6027 94.54 809.13 579.80 via POE::Loop::Select |
1588 |
|
1589 |
=head3 Discussion |
1590 |
|
1591 |
The benchmark does I<not> measure scalability of the event loop very |
1592 |
well. For example, a select-based event loop (such as the pure perl one) |
1593 |
can never compete with an event loop that uses epoll when the number of |
1594 |
file descriptors grows high. In this benchmark, all events become ready at |
1595 |
the same time, so select/poll-based implementations get an unnatural speed |
1596 |
boost. |
1597 |
|
1598 |
Also, note that the number of watchers usually has a nonlinear effect on |
1599 |
overall speed, that is, creating twice as many watchers doesn't take twice |
1600 |
the time - usually it takes longer. This puts event loops tested with a |
1601 |
higher number of watchers at a disadvantage. |
1602 |
|
1603 |
To put the range of results into perspective, consider that on the |
1604 |
benchmark machine, handling an event takes roughly 1600 CPU cycles with |
1605 |
EV, 3100 CPU cycles with AnyEvent's pure perl loop and almost 3000000 CPU |
1606 |
cycles with POE. |
1607 |
|
1608 |
C<EV> is the sole leader regarding speed and memory use, which are both |
1609 |
maximal/minimal, respectively. Even when going through AnyEvent, it uses |
1610 |
far less memory than any other event loop and is still faster than Event |
1611 |
natively. |
1612 |
|
1613 |
The pure perl implementation is hit in a few sweet spots (both the |
1614 |
constant timeout and the use of a single fd hit optimisations in the perl |
1615 |
interpreter and the backend itself). Nevertheless this shows that it |
1616 |
adds very little overhead in itself. Like any select-based backend its |
1617 |
performance becomes really bad with lots of file descriptors (and few of |
1618 |
them active), of course, but this was not subject of this benchmark. |
1619 |
|
1620 |
The C<Event> module has a relatively high setup and callback invocation |
1621 |
cost, but overall scores in on the third place. |
1622 |
|
1623 |
C<Glib>'s memory usage is quite a bit higher, but it features a |
1624 |
faster callback invocation and overall ends up in the same class as |
1625 |
C<Event>. However, Glib scales extremely badly, doubling the number of |
1626 |
watchers increases the processing time by more than a factor of four, |
1627 |
making it completely unusable when using larger numbers of watchers |
1628 |
(note that only a single file descriptor was used in the benchmark, so |
1629 |
inefficiencies of C<poll> do not account for this). |
1630 |
|
1631 |
The C<Tk> adaptor works relatively well. The fact that it crashes with |
1632 |
more than 2000 watchers is a big setback, however, as correctness takes |
1633 |
precedence over speed. Nevertheless, its performance is surprising, as the |
1634 |
file descriptor is dup()ed for each watcher. This shows that the dup() |
1635 |
employed by some adaptors is not a big performance issue (it does incur a |
1636 |
hidden memory cost inside the kernel which is not reflected in the figures |
1637 |
above). |
1638 |
|
1639 |
C<POE>, regardless of underlying event loop (whether using its pure perl |
1640 |
select-based backend or the Event module, the POE-EV backend couldn't |
1641 |
be tested because it wasn't working) shows abysmal performance and |
1642 |
memory usage with AnyEvent: Watchers use almost 30 times as much memory |
1643 |
as EV watchers, and 10 times as much memory as Event (the high memory |
1644 |
requirements are caused by requiring a session for each watcher). Watcher |
1645 |
invocation speed is almost 900 times slower than with AnyEvent's pure perl |
1646 |
implementation. |
1647 |
|
1648 |
The design of the POE adaptor class in AnyEvent can not really account |
1649 |
for the performance issues, though, as session creation overhead is |
1650 |
small compared to execution of the state machine, which is coded pretty |
1651 |
optimally within L<AnyEvent::Impl::POE> (and while everybody agrees that |
1652 |
using multiple sessions is not a good approach, especially regarding |
1653 |
memory usage, even the author of POE could not come up with a faster |
1654 |
design). |
1655 |
|
1656 |
=head3 Summary |
1657 |
|
1658 |
=over 4 |
1659 |
|
1660 |
=item * Using EV through AnyEvent is faster than any other event loop |
1661 |
(even when used without AnyEvent), but most event loops have acceptable |
1662 |
performance with or without AnyEvent. |
1663 |
|
1664 |
=item * The overhead AnyEvent adds is usually much smaller than the overhead of |
1665 |
the actual event loop, only with extremely fast event loops such as EV |
1666 |
adds AnyEvent significant overhead. |
1667 |
|
1668 |
=item * You should avoid POE like the plague if you want performance or |
1669 |
reasonable memory usage. |
1670 |
|
1671 |
=back |
1672 |
|
1673 |
=head2 BENCHMARKING THE LARGE SERVER CASE |
1674 |
|
1675 |
This benchmark actually benchmarks the event loop itself. It works by |
1676 |
creating a number of "servers": each server consists of a socket pair, a |
1677 |
timeout watcher that gets reset on activity (but never fires), and an I/O |
1678 |
watcher waiting for input on one side of the socket. Each time the socket |
1679 |
watcher reads a byte it will write that byte to a random other "server". |
1680 |
|
1681 |
The effect is that there will be a lot of I/O watchers, only part of which |
1682 |
are active at any one point (so there is a constant number of active |
1683 |
fds for each loop iteration, but which fds these are is random). The |
1684 |
timeout is reset each time something is read because that reflects how |
1685 |
most timeouts work (and puts extra pressure on the event loops). |
1686 |
|
1687 |
In this benchmark, we use 10000 socket pairs (20000 sockets), of which 100 |
1688 |
(1%) are active. This mirrors the activity of large servers with many |
1689 |
connections, most of which are idle at any one point in time. |
1690 |
|
1691 |
Source code for this benchmark is found as F<eg/bench2> in the AnyEvent |
1692 |
distribution. |
1693 |
|
1694 |
=head3 Explanation of the columns |
1695 |
|
1696 |
I<sockets> is the number of sockets, and twice the number of "servers" (as |
1697 |
each server has a read and write socket end). |
1698 |
|
1699 |
I<create> is the time it takes to create a socket pair (which is |
1700 |
nontrivial) and two watchers: an I/O watcher and a timeout watcher. |
1701 |
|
1702 |
I<request>, the most important value, is the time it takes to handle a |
1703 |
single "request", that is, reading the token from the pipe and forwarding |
1704 |
it to another server. This includes deleting the old timeout and creating |
1705 |
a new one that moves the timeout into the future. |
1706 |
|
1707 |
=head3 Results |
1708 |
|
1709 |
name sockets create request |
1710 |
EV 20000 69.01 11.16 |
1711 |
Perl 20000 73.32 35.87 |
1712 |
Event 20000 212.62 257.32 |
1713 |
Glib 20000 651.16 1896.30 |
1714 |
POE 20000 349.67 12317.24 uses POE::Loop::Event |
1715 |
|
1716 |
=head3 Discussion |
1717 |
|
1718 |
This benchmark I<does> measure scalability and overall performance of the |
1719 |
particular event loop. |
1720 |
|
1721 |
EV is again fastest. Since it is using epoll on my system, the setup time |
1722 |
is relatively high, though. |
1723 |
|
1724 |
Perl surprisingly comes second. It is much faster than the C-based event |
1725 |
loops Event and Glib. |
1726 |
|
1727 |
Event suffers from high setup time as well (look at its code and you will |
1728 |
understand why). Callback invocation also has a high overhead compared to |
1729 |
the C<< $_->() for .. >>-style loop that the Perl event loop uses. Event |
1730 |
uses select or poll in basically all documented configurations. |
1731 |
|
1732 |
Glib is hit hard by its quadratic behaviour w.r.t. many watchers. It |
1733 |
clearly fails to perform with many filehandles or in busy servers. |
1734 |
|
1735 |
POE is still completely out of the picture, taking over 1000 times as long |
1736 |
as EV, and over 100 times as long as the Perl implementation, even though |
1737 |
it uses a C-based event loop in this case. |
1738 |
|
1739 |
=head3 Summary |
1740 |
|
1741 |
=over 4 |
1742 |
|
1743 |
=item * The pure perl implementation performs extremely well. |
1744 |
|
1745 |
=item * Avoid Glib or POE in large projects where performance matters. |
1746 |
|
1747 |
=back |
1748 |
|
1749 |
=head2 BENCHMARKING SMALL SERVERS |
1750 |
|
1751 |
While event loops should scale (and select-based ones do not...) even to |
1752 |
large servers, most programs we (or I :) actually write have only a few |
1753 |
I/O watchers. |
1754 |
|
1755 |
In this benchmark, I use the same benchmark program as in the large server |
1756 |
case, but it uses only eight "servers", of which three are active at any |
1757 |
one time. This should reflect performance for a small server relatively |
1758 |
well. |
1759 |
|
1760 |
The columns are identical to the previous table. |
1761 |
|
1762 |
=head3 Results |
1763 |
|
1764 |
name sockets create request |
1765 |
EV 16 20.00 6.54 |
1766 |
Perl 16 25.75 12.62 |
1767 |
Event 16 81.27 35.86 |
1768 |
Glib 16 32.63 15.48 |
1769 |
POE 16 261.87 276.28 uses POE::Loop::Event |
1770 |
|
1771 |
=head3 Discussion |
1772 |
|
1773 |
The benchmark tries to test the performance of a typical small |
1774 |
server. While knowing how various event loops perform is interesting, keep |
1775 |
in mind that their overhead in this case is usually not as important, due |
1776 |
to the small absolute number of watchers (that is, you need efficiency and |
1777 |
speed most when you have lots of watchers, not when you only have a few of |
1778 |
them). |
1779 |
|
1780 |
EV is again fastest. |
1781 |
|
1782 |
Perl again comes second. It is noticeably faster than the C-based event |
1783 |
loops Event and Glib, although the difference is too small to really |
1784 |
matter. |
1785 |
|
1786 |
POE also performs much better in this case, but is is still far behind the |
1787 |
others. |
1788 |
|
1789 |
=head3 Summary |
1790 |
|
1791 |
=over 4 |
1792 |
|
1793 |
=item * C-based event loops perform very well with small number of |
1794 |
watchers, as the management overhead dominates. |
1795 |
|
1796 |
=back |
1797 |
|
1798 |
|
1799 |
=head1 SIGNALS |
1800 |
|
1801 |
AnyEvent currently installs handlers for these signals: |
1802 |
|
1803 |
=over 4 |
1804 |
|
1805 |
=item SIGCHLD |
1806 |
|
1807 |
A handler for C<SIGCHLD> is installed by AnyEvent's child watcher |
1808 |
emulation for event loops that do not support them natively. Also, some |
1809 |
event loops install a similar handler. |
1810 |
|
1811 |
=item SIGPIPE |
1812 |
|
1813 |
A no-op handler is installed for C<SIGPIPE> when C<$SIG{PIPE}> is C<undef> |
1814 |
when AnyEvent gets loaded. |
1815 |
|
1816 |
The rationale for this is that AnyEvent users usually do not really depend |
1817 |
on SIGPIPE delivery (which is purely an optimisation for shell use, or |
1818 |
badly-written programs), but C<SIGPIPE> can cause spurious and rare |
1819 |
program exits as a lot of people do not expect C<SIGPIPE> when writing to |
1820 |
some random socket. |
1821 |
|
1822 |
The rationale for installing a no-op handler as opposed to ignoring it is |
1823 |
that this way, the handler will be restored to defaults on exec. |
1824 |
|
1825 |
Feel free to install your own handler, or reset it to defaults. |
1826 |
|
1827 |
=back |
1828 |
|
1829 |
=cut |
1830 |
|
1831 |
$SIG{PIPE} = sub { } |
1832 |
unless defined $SIG{PIPE}; |
1833 |
|
1834 |
|
1835 |
=head1 FORK |
1836 |
|
1837 |
Most event libraries are not fork-safe. The ones who are usually are |
1838 |
because they rely on inefficient but fork-safe C<select> or C<poll> |
1839 |
calls. Only L<EV> is fully fork-aware. |
1840 |
|
1841 |
If you have to fork, you must either do so I<before> creating your first |
1842 |
watcher OR you must not use AnyEvent at all in the child. |
1843 |
|
1844 |
|
1845 |
=head1 SECURITY CONSIDERATIONS |
1846 |
|
1847 |
AnyEvent can be forced to load any event model via |
1848 |
$ENV{PERL_ANYEVENT_MODEL}. While this cannot (to my knowledge) be used to |
1849 |
execute arbitrary code or directly gain access, it can easily be used to |
1850 |
make the program hang or malfunction in subtle ways, as AnyEvent watchers |
1851 |
will not be active when the program uses a different event model than |
1852 |
specified in the variable. |
1853 |
|
1854 |
You can make AnyEvent completely ignore this variable by deleting it |
1855 |
before the first watcher gets created, e.g. with a C<BEGIN> block: |
1856 |
|
1857 |
BEGIN { delete $ENV{PERL_ANYEVENT_MODEL} } |
1858 |
|
1859 |
use AnyEvent; |
1860 |
|
1861 |
Similar considerations apply to $ENV{PERL_ANYEVENT_VERBOSE}, as that can |
1862 |
be used to probe what backend is used and gain other information (which is |
1863 |
probably even less useful to an attacker than PERL_ANYEVENT_MODEL), and |
1864 |
$ENV{PERL_ANYEGENT_STRICT}. |
1865 |
|
1866 |
|
1867 |
=head1 BUGS |
1868 |
|
1869 |
Perl 5.8 has numerous memleaks that sometimes hit this module and are hard |
1870 |
to work around. If you suffer from memleaks, first upgrade to Perl 5.10 |
1871 |
and check wether the leaks still show up. (Perl 5.10.0 has other annoying |
1872 |
memleaks, such as leaking on C<map> and C<grep> but it is usually not as |
1873 |
pronounced). |
1874 |
|
1875 |
|
1876 |
=head1 SEE ALSO |
1877 |
|
1878 |
Utility functions: L<AnyEvent::Util>. |
1879 |
|
1880 |
Event modules: L<EV>, L<EV::Glib>, L<Glib::EV>, L<Event>, L<Glib::Event>, |
1881 |
L<Glib>, L<Tk>, L<Event::Lib>, L<Qt>, L<POE>. |
1882 |
|
1883 |
Implementations: L<AnyEvent::Impl::EV>, L<AnyEvent::Impl::Event>, |
1884 |
L<AnyEvent::Impl::Glib>, L<AnyEvent::Impl::Tk>, L<AnyEvent::Impl::Perl>, |
1885 |
L<AnyEvent::Impl::EventLib>, L<AnyEvent::Impl::Qt>, |
1886 |
L<AnyEvent::Impl::POE>. |
1887 |
|
1888 |
Non-blocking file handles, sockets, TCP clients and |
1889 |
servers: L<AnyEvent::Handle>, L<AnyEvent::Socket>. |
1890 |
|
1891 |
Asynchronous DNS: L<AnyEvent::DNS>. |
1892 |
|
1893 |
Coroutine support: L<Coro>, L<Coro::AnyEvent>, L<Coro::EV>, L<Coro::Event>, |
1894 |
|
1895 |
Nontrivial usage examples: L<Net::FCP>, L<Net::XMPP2>, L<AnyEvent::DNS>. |
1896 |
|
1897 |
|
1898 |
=head1 AUTHOR |
1899 |
|
1900 |
Marc Lehmann <schmorp@schmorp.de> |
1901 |
http://home.schmorp.de/ |
1902 |
|
1903 |
=cut |
1904 |
|
1905 |
1 |
1906 |
|