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