1 |
NAME |
2 |
AnyEvent - the DBI of event loop programming |
3 |
|
4 |
EV, Event, Glib, Tk, Perl, Event::Lib, Irssi, rxvt-unicode, IO::Async, |
5 |
Qt and POE are various supported event loops/environments. |
6 |
|
7 |
SYNOPSIS |
8 |
use AnyEvent; |
9 |
|
10 |
# if you prefer function calls, look at the AE manpage for |
11 |
# an alternative API. |
12 |
|
13 |
# file handle or descriptor readable |
14 |
my $w = AnyEvent->io (fh => $fh, poll => "r", cb => sub { ... }); |
15 |
|
16 |
# one-shot or repeating timers |
17 |
my $w = AnyEvent->timer (after => $seconds, cb => sub { ... }); |
18 |
my $w = AnyEvent->timer (after => $seconds, interval => $seconds, cb => ...); |
19 |
|
20 |
print AnyEvent->now; # prints current event loop time |
21 |
print AnyEvent->time; # think Time::HiRes::time or simply CORE::time. |
22 |
|
23 |
# POSIX signal |
24 |
my $w = AnyEvent->signal (signal => "TERM", cb => sub { ... }); |
25 |
|
26 |
# child process exit |
27 |
my $w = AnyEvent->child (pid => $pid, cb => sub { |
28 |
my ($pid, $status) = @_; |
29 |
... |
30 |
}); |
31 |
|
32 |
# called when event loop idle (if applicable) |
33 |
my $w = AnyEvent->idle (cb => sub { ... }); |
34 |
|
35 |
my $w = AnyEvent->condvar; # stores whether a condition was flagged |
36 |
$w->send; # wake up current and all future recv's |
37 |
$w->recv; # enters "main loop" till $condvar gets ->send |
38 |
# use a condvar in callback mode: |
39 |
$w->cb (sub { $_[0]->recv }); |
40 |
|
41 |
INTRODUCTION/TUTORIAL |
42 |
This manpage is mainly a reference manual. If you are interested in a |
43 |
tutorial or some gentle introduction, have a look at the AnyEvent::Intro |
44 |
manpage. |
45 |
|
46 |
SUPPORT |
47 |
An FAQ document is available as AnyEvent::FAQ. |
48 |
|
49 |
There also is a mailinglist for discussing all things AnyEvent, and an |
50 |
IRC channel, too. |
51 |
|
52 |
See the AnyEvent project page at the Schmorpforge Ta-Sa Software |
53 |
Repository, at <http://anyevent.schmorp.de>, for more info. |
54 |
|
55 |
WHY YOU SHOULD USE THIS MODULE (OR NOT) |
56 |
Glib, POE, IO::Async, Event... CPAN offers event models by the dozen |
57 |
nowadays. So what is different about AnyEvent? |
58 |
|
59 |
Executive Summary: AnyEvent is *compatible*, AnyEvent is *free of |
60 |
policy* and AnyEvent is *small and efficient*. |
61 |
|
62 |
First and foremost, *AnyEvent is not an event model* itself, it only |
63 |
interfaces to whatever event model the main program happens to use, in a |
64 |
pragmatic way. For event models and certain classes of immortals alike, |
65 |
the statement "there can only be one" is a bitter reality: In general, |
66 |
only one event loop can be active at the same time in a process. |
67 |
AnyEvent cannot change this, but it can hide the differences between |
68 |
those event loops. |
69 |
|
70 |
The goal of AnyEvent is to offer module authors the ability to do event |
71 |
programming (waiting for I/O or timer events) without subscribing to a |
72 |
religion, a way of living, and most importantly: without forcing your |
73 |
module users into the same thing by forcing them to use the same event |
74 |
model you use. |
75 |
|
76 |
For modules like POE or IO::Async (which is a total misnomer as it is |
77 |
actually doing all I/O *synchronously*...), using them in your module is |
78 |
like joining a cult: After you join, you are dependent on them and you |
79 |
cannot use anything else, as they are simply incompatible to everything |
80 |
that isn't them. What's worse, all the potential users of your module |
81 |
are *also* forced to use the same event loop you use. |
82 |
|
83 |
AnyEvent is different: AnyEvent + POE works fine. AnyEvent + Glib works |
84 |
fine. AnyEvent + Tk works fine etc. etc. but none of these work together |
85 |
with the rest: POE + EV? No go. Tk + Event? No go. Again: if your module |
86 |
uses one of those, every user of your module has to use it, too. But if |
87 |
your module uses AnyEvent, it works transparently with all event models |
88 |
it supports (including stuff like IO::Async, as long as those use one of |
89 |
the supported event loops. It is easy to add new event loops to |
90 |
AnyEvent, too, so it is future-proof). |
91 |
|
92 |
In addition to being free of having to use *the one and only true event |
93 |
model*, AnyEvent also is free of bloat and policy: with POE or similar |
94 |
modules, you get an enormous amount of code and strict rules you have to |
95 |
follow. AnyEvent, on the other hand, is lean and to the point, by only |
96 |
offering the functionality that is necessary, in as thin as a wrapper as |
97 |
technically possible. |
98 |
|
99 |
Of course, AnyEvent comes with a big (and fully optional!) toolbox of |
100 |
useful functionality, such as an asynchronous DNS resolver, 100% |
101 |
non-blocking connects (even with TLS/SSL, IPv6 and on broken platforms |
102 |
such as Windows) and lots of real-world knowledge and workarounds for |
103 |
platform bugs and differences. |
104 |
|
105 |
Now, if you *do want* lots of policy (this can arguably be somewhat |
106 |
useful) and you want to force your users to use the one and only event |
107 |
model, you should *not* use this module. |
108 |
|
109 |
DESCRIPTION |
110 |
AnyEvent provides a uniform interface to various event loops. This |
111 |
allows module authors to use event loop functionality without forcing |
112 |
module users to use a specific event loop implementation (since more |
113 |
than one event loop cannot coexist peacefully). |
114 |
|
115 |
The interface itself is vaguely similar, but not identical to the Event |
116 |
module. |
117 |
|
118 |
During the first call of any watcher-creation method, the module tries |
119 |
to detect the currently loaded event loop by probing whether one of the |
120 |
following modules is already loaded: EV, AnyEvent::Loop, Event, Glib, |
121 |
Tk, Event::Lib, Qt, POE. The first one found is used. If none are |
122 |
detected, the module tries to load the first four modules in the order |
123 |
given; but note that if EV is not available, the pure-perl |
124 |
AnyEvent::Loop should always work, so the other two are not normally |
125 |
tried. |
126 |
|
127 |
Because AnyEvent first checks for modules that are already loaded, |
128 |
loading an event model explicitly before first using AnyEvent will |
129 |
likely make that model the default. For example: |
130 |
|
131 |
use Tk; |
132 |
use AnyEvent; |
133 |
|
134 |
# .. AnyEvent will likely default to Tk |
135 |
|
136 |
The *likely* means that, if any module loads another event model and |
137 |
starts using it, all bets are off - this case should be very rare |
138 |
though, as very few modules hardcode event loops without announcing this |
139 |
very loudly. |
140 |
|
141 |
The pure-perl implementation of AnyEvent is called "AnyEvent::Loop". |
142 |
Like other event modules you can load it explicitly and enjoy the high |
143 |
availability of that event loop :) |
144 |
|
145 |
WATCHERS |
146 |
AnyEvent has the central concept of a *watcher*, which is an object that |
147 |
stores relevant data for each kind of event you are waiting for, such as |
148 |
the callback to call, the file handle to watch, etc. |
149 |
|
150 |
These watchers are normal Perl objects with normal Perl lifetime. After |
151 |
creating a watcher it will immediately "watch" for events and invoke the |
152 |
callback when the event occurs (of course, only when the event model is |
153 |
in control). |
154 |
|
155 |
Note that callbacks must not permanently change global variables |
156 |
potentially in use by the event loop (such as $_ or $[) and that |
157 |
callbacks must not "die". The former is good programming practice in |
158 |
Perl and the latter stems from the fact that exception handling differs |
159 |
widely between event loops. |
160 |
|
161 |
To disable a watcher you have to destroy it (e.g. by setting the |
162 |
variable you store it in to "undef" or otherwise deleting all references |
163 |
to it). |
164 |
|
165 |
All watchers are created by calling a method on the "AnyEvent" class. |
166 |
|
167 |
Many watchers either are used with "recursion" (repeating timers for |
168 |
example), or need to refer to their watcher object in other ways. |
169 |
|
170 |
One way to achieve that is this pattern: |
171 |
|
172 |
my $w; $w = AnyEvent->type (arg => value ..., cb => sub { |
173 |
# you can use $w here, for example to undef it |
174 |
undef $w; |
175 |
}); |
176 |
|
177 |
Note that "my $w; $w =" combination. This is necessary because in Perl, |
178 |
my variables are only visible after the statement in which they are |
179 |
declared. |
180 |
|
181 |
I/O WATCHERS |
182 |
$w = AnyEvent->io ( |
183 |
fh => <filehandle_or_fileno>, |
184 |
poll => <"r" or "w">, |
185 |
cb => <callback>, |
186 |
); |
187 |
|
188 |
You can create an I/O watcher by calling the "AnyEvent->io" method with |
189 |
the following mandatory key-value pairs as arguments: |
190 |
|
191 |
"fh" is the Perl *file handle* (or a naked file descriptor) to watch for |
192 |
events (AnyEvent might or might not keep a reference to this file |
193 |
handle). Note that only file handles pointing to things for which |
194 |
non-blocking operation makes sense are allowed. This includes sockets, |
195 |
most character devices, pipes, fifos and so on, but not for example |
196 |
files or block devices. |
197 |
|
198 |
"poll" must be a string that is either "r" or "w", which creates a |
199 |
watcher waiting for "r"eadable or "w"ritable events, respectively. |
200 |
|
201 |
"cb" is the callback to invoke each time the file handle becomes ready. |
202 |
|
203 |
Although the callback might get passed parameters, their value and |
204 |
presence is undefined and you cannot rely on them. Portable AnyEvent |
205 |
callbacks cannot use arguments passed to I/O watcher callbacks. |
206 |
|
207 |
The I/O watcher might use the underlying file descriptor or a copy of |
208 |
it. You must not close a file handle as long as any watcher is active on |
209 |
the underlying file descriptor. |
210 |
|
211 |
Some event loops issue spurious readiness notifications, so you should |
212 |
always use non-blocking calls when reading/writing from/to your file |
213 |
handles. |
214 |
|
215 |
Example: wait for readability of STDIN, then read a line and disable the |
216 |
watcher. |
217 |
|
218 |
my $w; $w = AnyEvent->io (fh => \*STDIN, poll => 'r', cb => sub { |
219 |
chomp (my $input = <STDIN>); |
220 |
warn "read: $input\n"; |
221 |
undef $w; |
222 |
}); |
223 |
|
224 |
TIME WATCHERS |
225 |
$w = AnyEvent->timer (after => <seconds>, cb => <callback>); |
226 |
|
227 |
$w = AnyEvent->timer ( |
228 |
after => <fractional_seconds>, |
229 |
interval => <fractional_seconds>, |
230 |
cb => <callback>, |
231 |
); |
232 |
|
233 |
You can create a time watcher by calling the "AnyEvent->timer" method |
234 |
with the following mandatory arguments: |
235 |
|
236 |
"after" specifies after how many seconds (fractional values are |
237 |
supported) the callback should be invoked. "cb" is the callback to |
238 |
invoke in that case. |
239 |
|
240 |
Although the callback might get passed parameters, their value and |
241 |
presence is undefined and you cannot rely on them. Portable AnyEvent |
242 |
callbacks cannot use arguments passed to time watcher callbacks. |
243 |
|
244 |
The callback will normally be invoked only once. If you specify another |
245 |
parameter, "interval", as a strictly positive number (> 0), then the |
246 |
callback will be invoked regularly at that interval (in fractional |
247 |
seconds) after the first invocation. If "interval" is specified with a |
248 |
false value, then it is treated as if it were not specified at all. |
249 |
|
250 |
The callback will be rescheduled before invoking the callback, but no |
251 |
attempt is made to avoid timer drift in most backends, so the interval |
252 |
is only approximate. |
253 |
|
254 |
Example: fire an event after 7.7 seconds. |
255 |
|
256 |
my $w = AnyEvent->timer (after => 7.7, cb => sub { |
257 |
warn "timeout\n"; |
258 |
}); |
259 |
|
260 |
# to cancel the timer: |
261 |
undef $w; |
262 |
|
263 |
Example 2: fire an event after 0.5 seconds, then roughly every second. |
264 |
|
265 |
my $w = AnyEvent->timer (after => 0.5, interval => 1, cb => sub { |
266 |
warn "timeout\n"; |
267 |
}; |
268 |
|
269 |
TIMING ISSUES |
270 |
There are two ways to handle timers: based on real time (relative, "fire |
271 |
in 10 seconds") and based on wallclock time (absolute, "fire at 12 |
272 |
o'clock"). |
273 |
|
274 |
While most event loops expect timers to specified in a relative way, |
275 |
they use absolute time internally. This makes a difference when your |
276 |
clock "jumps", for example, when ntp decides to set your clock backwards |
277 |
from the wrong date of 2014-01-01 to 2008-01-01, a watcher that is |
278 |
supposed to fire "after a second" might actually take six years to |
279 |
finally fire. |
280 |
|
281 |
AnyEvent cannot compensate for this. The only event loop that is |
282 |
conscious of these issues is EV, which offers both relative (ev_timer, |
283 |
based on true relative time) and absolute (ev_periodic, based on |
284 |
wallclock time) timers. |
285 |
|
286 |
AnyEvent always prefers relative timers, if available, matching the |
287 |
AnyEvent API. |
288 |
|
289 |
AnyEvent has two additional methods that return the "current time": |
290 |
|
291 |
AnyEvent->time |
292 |
This returns the "current wallclock time" as a fractional number of |
293 |
seconds since the Epoch (the same thing as "time" or |
294 |
"Time::HiRes::time" return, and the result is guaranteed to be |
295 |
compatible with those). |
296 |
|
297 |
It progresses independently of any event loop processing, i.e. each |
298 |
call will check the system clock, which usually gets updated |
299 |
frequently. |
300 |
|
301 |
AnyEvent->now |
302 |
This also returns the "current wallclock time", but unlike "time", |
303 |
above, this value might change only once per event loop iteration, |
304 |
depending on the event loop (most return the same time as "time", |
305 |
above). This is the time that AnyEvent's timers get scheduled |
306 |
against. |
307 |
|
308 |
*In almost all cases (in all cases if you don't care), this is the |
309 |
function to call when you want to know the current time.* |
310 |
|
311 |
This function is also often faster then "AnyEvent->time", and thus |
312 |
the preferred method if you want some timestamp (for example, |
313 |
AnyEvent::Handle uses this to update its activity timeouts). |
314 |
|
315 |
The rest of this section is only of relevance if you try to be very |
316 |
exact with your timing; you can skip it without a bad conscience. |
317 |
|
318 |
For a practical example of when these times differ, consider |
319 |
Event::Lib and EV and the following set-up: |
320 |
|
321 |
The event loop is running and has just invoked one of your callbacks |
322 |
at time=500 (assume no other callbacks delay processing). In your |
323 |
callback, you wait a second by executing "sleep 1" (blocking the |
324 |
process for a second) and then (at time=501) you create a relative |
325 |
timer that fires after three seconds. |
326 |
|
327 |
With Event::Lib, "AnyEvent->time" and "AnyEvent->now" will both |
328 |
return 501, because that is the current time, and the timer will be |
329 |
scheduled to fire at time=504 (501 + 3). |
330 |
|
331 |
With EV, "AnyEvent->time" returns 501 (as that is the current time), |
332 |
but "AnyEvent->now" returns 500, as that is the time the last event |
333 |
processing phase started. With EV, your timer gets scheduled to run |
334 |
at time=503 (500 + 3). |
335 |
|
336 |
In one sense, Event::Lib is more exact, as it uses the current time |
337 |
regardless of any delays introduced by event processing. However, |
338 |
most callbacks do not expect large delays in processing, so this |
339 |
causes a higher drift (and a lot more system calls to get the |
340 |
current time). |
341 |
|
342 |
In another sense, EV is more exact, as your timer will be scheduled |
343 |
at the same time, regardless of how long event processing actually |
344 |
took. |
345 |
|
346 |
In either case, if you care (and in most cases, you don't), then you |
347 |
can get whatever behaviour you want with any event loop, by taking |
348 |
the difference between "AnyEvent->time" and "AnyEvent->now" into |
349 |
account. |
350 |
|
351 |
AnyEvent->now_update |
352 |
Some event loops (such as EV or AnyEvent::Loop) cache the current |
353 |
time for each loop iteration (see the discussion of AnyEvent->now, |
354 |
above). |
355 |
|
356 |
When a callback runs for a long time (or when the process sleeps), |
357 |
then this "current" time will differ substantially from the real |
358 |
time, which might affect timers and time-outs. |
359 |
|
360 |
When this is the case, you can call this method, which will update |
361 |
the event loop's idea of "current time". |
362 |
|
363 |
A typical example would be a script in a web server (e.g. |
364 |
"mod_perl") - when mod_perl executes the script, then the event loop |
365 |
will have the wrong idea about the "current time" (being potentially |
366 |
far in the past, when the script ran the last time). In that case |
367 |
you should arrange a call to "AnyEvent->now_update" each time the |
368 |
web server process wakes up again (e.g. at the start of your script, |
369 |
or in a handler). |
370 |
|
371 |
Note that updating the time *might* cause some events to be handled. |
372 |
|
373 |
SIGNAL WATCHERS |
374 |
$w = AnyEvent->signal (signal => <uppercase_signal_name>, cb => <callback>); |
375 |
|
376 |
You can watch for signals using a signal watcher, "signal" is the signal |
377 |
*name* in uppercase and without any "SIG" prefix, "cb" is the Perl |
378 |
callback to be invoked whenever a signal occurs. |
379 |
|
380 |
Although the callback might get passed parameters, their value and |
381 |
presence is undefined and you cannot rely on them. Portable AnyEvent |
382 |
callbacks cannot use arguments passed to signal watcher callbacks. |
383 |
|
384 |
Multiple signal occurrences can be clumped together into one callback |
385 |
invocation, and callback invocation will be synchronous. Synchronous |
386 |
means that it might take a while until the signal gets handled by the |
387 |
process, but it is guaranteed not to interrupt any other callbacks. |
388 |
|
389 |
The main advantage of using these watchers is that you can share a |
390 |
signal between multiple watchers, and AnyEvent will ensure that signals |
391 |
will not interrupt your program at bad times. |
392 |
|
393 |
This watcher might use %SIG (depending on the event loop used), so |
394 |
programs overwriting those signals directly will likely not work |
395 |
correctly. |
396 |
|
397 |
Example: exit on SIGINT |
398 |
|
399 |
my $w = AnyEvent->signal (signal => "INT", cb => sub { exit 1 }); |
400 |
|
401 |
Restart Behaviour |
402 |
While restart behaviour is up to the event loop implementation, most |
403 |
will not restart syscalls (that includes Async::Interrupt and AnyEvent's |
404 |
pure perl implementation). |
405 |
|
406 |
Safe/Unsafe Signals |
407 |
Perl signals can be either "safe" (synchronous to opcode handling) or |
408 |
"unsafe" (asynchronous) - the former might get delayed indefinitely, the |
409 |
latter might corrupt your memory. |
410 |
|
411 |
AnyEvent signal handlers are, in addition, synchronous to the event |
412 |
loop, i.e. they will not interrupt your running perl program but will |
413 |
only be called as part of the normal event handling (just like timer, |
414 |
I/O etc. callbacks, too). |
415 |
|
416 |
Signal Races, Delays and Workarounds |
417 |
Many event loops (e.g. Glib, Tk, Qt, IO::Async) do not support attaching |
418 |
callbacks to signals in a generic way, which is a pity, as you cannot do |
419 |
race-free signal handling in perl, requiring C libraries for this. |
420 |
AnyEvent will try to do its best, which means in some cases, signals |
421 |
will be delayed. The maximum time a signal might be delayed is specified |
422 |
in $AnyEvent::MAX_SIGNAL_LATENCY (default: 10 seconds). This variable |
423 |
can be changed only before the first signal watcher is created, and |
424 |
should be left alone otherwise. This variable determines how often |
425 |
AnyEvent polls for signals (in case a wake-up was missed). Higher values |
426 |
will cause fewer spurious wake-ups, which is better for power and CPU |
427 |
saving. |
428 |
|
429 |
All these problems can be avoided by installing the optional |
430 |
Async::Interrupt module, which works with most event loops. It will not |
431 |
work with inherently broken event loops such as Event or Event::Lib (and |
432 |
not with POE currently, as POE does its own workaround with one-second |
433 |
latency). For those, you just have to suffer the delays. |
434 |
|
435 |
CHILD PROCESS WATCHERS |
436 |
$w = AnyEvent->child (pid => <process id>, cb => <callback>); |
437 |
|
438 |
You can also watch for a child process exit and catch its exit status. |
439 |
|
440 |
The child process is specified by the "pid" argument (on some backends, |
441 |
using 0 watches for any child process exit, on others this will croak). |
442 |
The watcher will be triggered only when the child process has finished |
443 |
and an exit status is available, not on any trace events |
444 |
(stopped/continued). |
445 |
|
446 |
The callback will be called with the pid and exit status (as returned by |
447 |
waitpid), so unlike other watcher types, you *can* rely on child watcher |
448 |
callback arguments. |
449 |
|
450 |
This watcher type works by installing a signal handler for "SIGCHLD", |
451 |
and since it cannot be shared, nothing else should use SIGCHLD or reap |
452 |
random child processes (waiting for specific child processes, e.g. |
453 |
inside "system", is just fine). |
454 |
|
455 |
There is a slight catch to child watchers, however: you usually start |
456 |
them *after* the child process was created, and this means the process |
457 |
could have exited already (and no SIGCHLD will be sent anymore). |
458 |
|
459 |
Not all event models handle this correctly (neither POE nor IO::Async |
460 |
do, see their AnyEvent::Impl manpages for details), but even for event |
461 |
models that *do* handle this correctly, they usually need to be loaded |
462 |
before the process exits (i.e. before you fork in the first place). |
463 |
AnyEvent's pure perl event loop handles all cases correctly regardless |
464 |
of when you start the watcher. |
465 |
|
466 |
This means you cannot create a child watcher as the very first thing in |
467 |
an AnyEvent program, you *have* to create at least one watcher before |
468 |
you "fork" the child (alternatively, you can call "AnyEvent::detect"). |
469 |
|
470 |
As most event loops do not support waiting for child events, they will |
471 |
be emulated by AnyEvent in most cases, in which case the latency and |
472 |
race problems mentioned in the description of signal watchers apply. |
473 |
|
474 |
Example: fork a process and wait for it |
475 |
|
476 |
my $done = AnyEvent->condvar; |
477 |
|
478 |
my $pid = fork or exit 5; |
479 |
|
480 |
my $w = AnyEvent->child ( |
481 |
pid => $pid, |
482 |
cb => sub { |
483 |
my ($pid, $status) = @_; |
484 |
warn "pid $pid exited with status $status"; |
485 |
$done->send; |
486 |
}, |
487 |
); |
488 |
|
489 |
# do something else, then wait for process exit |
490 |
$done->recv; |
491 |
|
492 |
IDLE WATCHERS |
493 |
$w = AnyEvent->idle (cb => <callback>); |
494 |
|
495 |
This will repeatedly invoke the callback after the process becomes idle, |
496 |
until either the watcher is destroyed or new events have been detected. |
497 |
|
498 |
Idle watchers are useful when there is a need to do something, but it is |
499 |
not so important (or wise) to do it instantly. The callback will be |
500 |
invoked only when there is "nothing better to do", which is usually |
501 |
defined as "all outstanding events have been handled and no new events |
502 |
have been detected". That means that idle watchers ideally get invoked |
503 |
when the event loop has just polled for new events but none have been |
504 |
detected. Instead of blocking to wait for more events, the idle watchers |
505 |
will be invoked. |
506 |
|
507 |
Unfortunately, most event loops do not really support idle watchers |
508 |
(only EV, Event and Glib do it in a usable fashion) - for the rest, |
509 |
AnyEvent will simply call the callback "from time to time". |
510 |
|
511 |
Example: read lines from STDIN, but only process them when the program |
512 |
is otherwise idle: |
513 |
|
514 |
my @lines; # read data |
515 |
my $idle_w; |
516 |
my $io_w = AnyEvent->io (fh => \*STDIN, poll => 'r', cb => sub { |
517 |
push @lines, scalar <STDIN>; |
518 |
|
519 |
# start an idle watcher, if not already done |
520 |
$idle_w ||= AnyEvent->idle (cb => sub { |
521 |
# handle only one line, when there are lines left |
522 |
if (my $line = shift @lines) { |
523 |
print "handled when idle: $line"; |
524 |
} else { |
525 |
# otherwise disable the idle watcher again |
526 |
undef $idle_w; |
527 |
} |
528 |
}); |
529 |
}); |
530 |
|
531 |
CONDITION VARIABLES |
532 |
$cv = AnyEvent->condvar; |
533 |
|
534 |
$cv->send (<list>); |
535 |
my @res = $cv->recv; |
536 |
|
537 |
If you are familiar with some event loops you will know that all of them |
538 |
require you to run some blocking "loop", "run" or similar function that |
539 |
will actively watch for new events and call your callbacks. |
540 |
|
541 |
AnyEvent is slightly different: it expects somebody else to run the |
542 |
event loop and will only block when necessary (usually when told by the |
543 |
user). |
544 |
|
545 |
The tool to do that is called a "condition variable", so called because |
546 |
they represent a condition that must become true. |
547 |
|
548 |
Now is probably a good time to look at the examples further below. |
549 |
|
550 |
Condition variables can be created by calling the "AnyEvent->condvar" |
551 |
method, usually without arguments. The only argument pair allowed is |
552 |
"cb", which specifies a callback to be called when the condition |
553 |
variable becomes true, with the condition variable as the first argument |
554 |
(but not the results). |
555 |
|
556 |
After creation, the condition variable is "false" until it becomes |
557 |
"true" by calling the "send" method (or calling the condition variable |
558 |
as if it were a callback, read about the caveats in the description for |
559 |
the "->send" method). |
560 |
|
561 |
Since condition variables are the most complex part of the AnyEvent API, |
562 |
here are some different mental models of what they are - pick the ones |
563 |
you can connect to: |
564 |
|
565 |
* Condition variables are like callbacks - you can call them (and pass |
566 |
them instead of callbacks). Unlike callbacks however, you can also |
567 |
wait for them to be called. |
568 |
|
569 |
* Condition variables are signals - one side can emit or send them, |
570 |
the other side can wait for them, or install a handler that is |
571 |
called when the signal fires. |
572 |
|
573 |
* Condition variables are like "Merge Points" - points in your program |
574 |
where you merge multiple independent results/control flows into one. |
575 |
|
576 |
* Condition variables represent a transaction - functions that start |
577 |
some kind of transaction can return them, leaving the caller the |
578 |
choice between waiting in a blocking fashion, or setting a callback. |
579 |
|
580 |
* Condition variables represent future values, or promises to deliver |
581 |
some result, long before the result is available. |
582 |
|
583 |
Condition variables are very useful to signal that something has |
584 |
finished, for example, if you write a module that does asynchronous http |
585 |
requests, then a condition variable would be the ideal candidate to |
586 |
signal the availability of results. The user can either act when the |
587 |
callback is called or can synchronously "->recv" for the results. |
588 |
|
589 |
You can also use them to simulate traditional event loops - for example, |
590 |
you can block your main program until an event occurs - for example, you |
591 |
could "->recv" in your main program until the user clicks the Quit |
592 |
button of your app, which would "->send" the "quit" event. |
593 |
|
594 |
Note that condition variables recurse into the event loop - if you have |
595 |
two pieces of code that call "->recv" in a round-robin fashion, you |
596 |
lose. Therefore, condition variables are good to export to your caller, |
597 |
but you should avoid making a blocking wait yourself, at least in |
598 |
callbacks, as this asks for trouble. |
599 |
|
600 |
Condition variables are represented by hash refs in perl, and the keys |
601 |
used by AnyEvent itself are all named "_ae_XXX" to make subclassing easy |
602 |
(it is often useful to build your own transaction class on top of |
603 |
AnyEvent). To subclass, use "AnyEvent::CondVar" as base class and call |
604 |
its "new" method in your own "new" method. |
605 |
|
606 |
There are two "sides" to a condition variable - the "producer side" |
607 |
which eventually calls "-> send", and the "consumer side", which waits |
608 |
for the send to occur. |
609 |
|
610 |
Example: wait for a timer. |
611 |
|
612 |
# condition: "wait till the timer is fired" |
613 |
my $timer_fired = AnyEvent->condvar; |
614 |
|
615 |
# create the timer - we could wait for, say |
616 |
# a handle becomign ready, or even an |
617 |
# AnyEvent::HTTP request to finish, but |
618 |
# in this case, we simply use a timer: |
619 |
my $w = AnyEvent->timer ( |
620 |
after => 1, |
621 |
cb => sub { $timer_fired->send }, |
622 |
); |
623 |
|
624 |
# this "blocks" (while handling events) till the callback |
625 |
# calls ->send |
626 |
$timer_fired->recv; |
627 |
|
628 |
Example: wait for a timer, but take advantage of the fact that condition |
629 |
variables are also callable directly. |
630 |
|
631 |
my $done = AnyEvent->condvar; |
632 |
my $delay = AnyEvent->timer (after => 5, cb => $done); |
633 |
$done->recv; |
634 |
|
635 |
Example: Imagine an API that returns a condvar and doesn't support |
636 |
callbacks. This is how you make a synchronous call, for example from the |
637 |
main program: |
638 |
|
639 |
use AnyEvent::CouchDB; |
640 |
|
641 |
... |
642 |
|
643 |
my @info = $couchdb->info->recv; |
644 |
|
645 |
And this is how you would just set a callback to be called whenever the |
646 |
results are available: |
647 |
|
648 |
$couchdb->info->cb (sub { |
649 |
my @info = $_[0]->recv; |
650 |
}); |
651 |
|
652 |
METHODS FOR PRODUCERS |
653 |
These methods should only be used by the producing side, i.e. the |
654 |
code/module that eventually sends the signal. Note that it is also the |
655 |
producer side which creates the condvar in most cases, but it isn't |
656 |
uncommon for the consumer to create it as well. |
657 |
|
658 |
$cv->send (...) |
659 |
Flag the condition as ready - a running "->recv" and all further |
660 |
calls to "recv" will (eventually) return after this method has been |
661 |
called. If nobody is waiting the send will be remembered. |
662 |
|
663 |
If a callback has been set on the condition variable, it is called |
664 |
immediately from within send. |
665 |
|
666 |
Any arguments passed to the "send" call will be returned by all |
667 |
future "->recv" calls. |
668 |
|
669 |
Condition variables are overloaded so one can call them directly (as |
670 |
if they were a code reference). Calling them directly is the same as |
671 |
calling "send". |
672 |
|
673 |
$cv->croak ($error) |
674 |
Similar to send, but causes all calls to "->recv" to invoke |
675 |
"Carp::croak" with the given error message/object/scalar. |
676 |
|
677 |
This can be used to signal any errors to the condition variable |
678 |
user/consumer. Doing it this way instead of calling "croak" directly |
679 |
delays the error detection, but has the overwhelming advantage that |
680 |
it diagnoses the error at the place where the result is expected, |
681 |
and not deep in some event callback with no connection to the actual |
682 |
code causing the problem. |
683 |
|
684 |
$cv->begin ([group callback]) |
685 |
$cv->end |
686 |
These two methods can be used to combine many transactions/events |
687 |
into one. For example, a function that pings many hosts in parallel |
688 |
might want to use a condition variable for the whole process. |
689 |
|
690 |
Every call to "->begin" will increment a counter, and every call to |
691 |
"->end" will decrement it. If the counter reaches 0 in "->end", the |
692 |
(last) callback passed to "begin" will be executed, passing the |
693 |
condvar as first argument. That callback is *supposed* to call |
694 |
"->send", but that is not required. If no group callback was set, |
695 |
"send" will be called without any arguments. |
696 |
|
697 |
You can think of "$cv->send" giving you an OR condition (one call |
698 |
sends), while "$cv->begin" and "$cv->end" giving you an AND |
699 |
condition (all "begin" calls must be "end"'ed before the condvar |
700 |
sends). |
701 |
|
702 |
Let's start with a simple example: you have two I/O watchers (for |
703 |
example, STDOUT and STDERR for a program), and you want to wait for |
704 |
both streams to close before activating a condvar: |
705 |
|
706 |
my $cv = AnyEvent->condvar; |
707 |
|
708 |
$cv->begin; # first watcher |
709 |
my $w1 = AnyEvent->io (fh => $fh1, cb => sub { |
710 |
defined sysread $fh1, my $buf, 4096 |
711 |
or $cv->end; |
712 |
}); |
713 |
|
714 |
$cv->begin; # second watcher |
715 |
my $w2 = AnyEvent->io (fh => $fh2, cb => sub { |
716 |
defined sysread $fh2, my $buf, 4096 |
717 |
or $cv->end; |
718 |
}); |
719 |
|
720 |
$cv->recv; |
721 |
|
722 |
This works because for every event source (EOF on file handle), |
723 |
there is one call to "begin", so the condvar waits for all calls to |
724 |
"end" before sending. |
725 |
|
726 |
The ping example mentioned above is slightly more complicated, as |
727 |
the there are results to be passwd back, and the number of tasks |
728 |
that are begun can potentially be zero: |
729 |
|
730 |
my $cv = AnyEvent->condvar; |
731 |
|
732 |
my %result; |
733 |
$cv->begin (sub { shift->send (\%result) }); |
734 |
|
735 |
for my $host (@list_of_hosts) { |
736 |
$cv->begin; |
737 |
ping_host_then_call_callback $host, sub { |
738 |
$result{$host} = ...; |
739 |
$cv->end; |
740 |
}; |
741 |
} |
742 |
|
743 |
$cv->end; |
744 |
|
745 |
This code fragment supposedly pings a number of hosts and calls |
746 |
"send" after results for all then have have been gathered - in any |
747 |
order. To achieve this, the code issues a call to "begin" when it |
748 |
starts each ping request and calls "end" when it has received some |
749 |
result for it. Since "begin" and "end" only maintain a counter, the |
750 |
order in which results arrive is not relevant. |
751 |
|
752 |
There is an additional bracketing call to "begin" and "end" outside |
753 |
the loop, which serves two important purposes: first, it sets the |
754 |
callback to be called once the counter reaches 0, and second, it |
755 |
ensures that "send" is called even when "no" hosts are being pinged |
756 |
(the loop doesn't execute once). |
757 |
|
758 |
This is the general pattern when you "fan out" into multiple (but |
759 |
potentially zero) subrequests: use an outer "begin"/"end" pair to |
760 |
set the callback and ensure "end" is called at least once, and then, |
761 |
for each subrequest you start, call "begin" and for each subrequest |
762 |
you finish, call "end". |
763 |
|
764 |
METHODS FOR CONSUMERS |
765 |
These methods should only be used by the consuming side, i.e. the code |
766 |
awaits the condition. |
767 |
|
768 |
$cv->recv |
769 |
Wait (blocking if necessary) until the "->send" or "->croak" methods |
770 |
have been called on $cv, while servicing other watchers normally. |
771 |
|
772 |
You can only wait once on a condition - additional calls are valid |
773 |
but will return immediately. |
774 |
|
775 |
If an error condition has been set by calling "->croak", then this |
776 |
function will call "croak". |
777 |
|
778 |
In list context, all parameters passed to "send" will be returned, |
779 |
in scalar context only the first one will be returned. |
780 |
|
781 |
Note that doing a blocking wait in a callback is not supported by |
782 |
any event loop, that is, recursive invocation of a blocking "->recv" |
783 |
is not allowed, and the "recv" call will "croak" if such a condition |
784 |
is detected. This condition can be slightly loosened by using |
785 |
Coro::AnyEvent, which allows you to do a blocking "->recv" from any |
786 |
thread that doesn't run the event loop itself. |
787 |
|
788 |
Not all event models support a blocking wait - some die in that case |
789 |
(programs might want to do that to stay interactive), so *if you are |
790 |
using this from a module, never require a blocking wait*. Instead, |
791 |
let the caller decide whether the call will block or not (for |
792 |
example, by coupling condition variables with some kind of request |
793 |
results and supporting callbacks so the caller knows that getting |
794 |
the result will not block, while still supporting blocking waits if |
795 |
the caller so desires). |
796 |
|
797 |
You can ensure that "->recv" never blocks by setting a callback and |
798 |
only calling "->recv" from within that callback (or at a later |
799 |
time). This will work even when the event loop does not support |
800 |
blocking waits otherwise. |
801 |
|
802 |
$bool = $cv->ready |
803 |
Returns true when the condition is "true", i.e. whether "send" or |
804 |
"croak" have been called. |
805 |
|
806 |
$cb = $cv->cb ($cb->($cv)) |
807 |
This is a mutator function that returns the callback set and |
808 |
optionally replaces it before doing so. |
809 |
|
810 |
The callback will be called when the condition becomes "true", i.e. |
811 |
when "send" or "croak" are called, with the only argument being the |
812 |
condition variable itself. If the condition is already true, the |
813 |
callback is called immediately when it is set. Calling "recv" inside |
814 |
the callback or at any later time is guaranteed not to block. |
815 |
|
816 |
SUPPORTED EVENT LOOPS/BACKENDS |
817 |
The available backend classes are (every class has its own manpage): |
818 |
|
819 |
Backends that are autoprobed when no other event loop can be found. |
820 |
EV is the preferred backend when no other event loop seems to be in |
821 |
use. If EV is not installed, then AnyEvent will fall back to its own |
822 |
pure-perl implementation, which is available everywhere as it comes |
823 |
with AnyEvent itself. |
824 |
|
825 |
AnyEvent::Impl::EV based on EV (interface to libev, best choice). |
826 |
AnyEvent::Impl::Perl pure-perl AnyEvent::Loop, fast and portable. |
827 |
|
828 |
Backends that are transparently being picked up when they are used. |
829 |
These will be used if they are already loaded when the first watcher |
830 |
is created, in which case it is assumed that the application is |
831 |
using them. This means that AnyEvent will automatically pick the |
832 |
right backend when the main program loads an event module before |
833 |
anything starts to create watchers. Nothing special needs to be done |
834 |
by the main program. |
835 |
|
836 |
AnyEvent::Impl::Event based on Event, very stable, few glitches. |
837 |
AnyEvent::Impl::Glib based on Glib, slow but very stable. |
838 |
AnyEvent::Impl::Tk based on Tk, very broken. |
839 |
AnyEvent::Impl::EventLib based on Event::Lib, leaks memory and worse. |
840 |
AnyEvent::Impl::POE based on POE, very slow, some limitations. |
841 |
AnyEvent::Impl::Irssi used when running within irssi. |
842 |
AnyEvent::Impl::IOAsync based on IO::Async. |
843 |
AnyEvent::Impl::Cocoa based on Cocoa::EventLoop. |
844 |
AnyEvent::Impl::FLTK2 based on FLTK (fltk 2 binding). |
845 |
|
846 |
Backends with special needs. |
847 |
Qt requires the Qt::Application to be instantiated first, but will |
848 |
otherwise be picked up automatically. As long as the main program |
849 |
instantiates the application before any AnyEvent watchers are |
850 |
created, everything should just work. |
851 |
|
852 |
AnyEvent::Impl::Qt based on Qt. |
853 |
|
854 |
Event loops that are indirectly supported via other backends. |
855 |
Some event loops can be supported via other modules: |
856 |
|
857 |
There is no direct support for WxWidgets (Wx) or Prima. |
858 |
|
859 |
WxWidgets has no support for watching file handles. However, you can |
860 |
use WxWidgets through the POE adaptor, as POE has a Wx backend that |
861 |
simply polls 20 times per second, which was considered to be too |
862 |
horrible to even consider for AnyEvent. |
863 |
|
864 |
Prima is not supported as nobody seems to be using it, but it has a |
865 |
POE backend, so it can be supported through POE. |
866 |
|
867 |
AnyEvent knows about both Prima and Wx, however, and will try to |
868 |
load POE when detecting them, in the hope that POE will pick them |
869 |
up, in which case everything will be automatic. |
870 |
|
871 |
GLOBAL VARIABLES AND FUNCTIONS |
872 |
These are not normally required to use AnyEvent, but can be useful to |
873 |
write AnyEvent extension modules. |
874 |
|
875 |
$AnyEvent::MODEL |
876 |
Contains "undef" until the first watcher is being created, before |
877 |
the backend has been autodetected. |
878 |
|
879 |
Afterwards it contains the event model that is being used, which is |
880 |
the name of the Perl class implementing the model. This class is |
881 |
usually one of the "AnyEvent::Impl::xxx" modules, but can be any |
882 |
other class in the case AnyEvent has been extended at runtime (e.g. |
883 |
in *rxvt-unicode* it will be "urxvt::anyevent"). |
884 |
|
885 |
AnyEvent::detect |
886 |
Returns $AnyEvent::MODEL, forcing autodetection of the event model |
887 |
if necessary. You should only call this function right before you |
888 |
would have created an AnyEvent watcher anyway, that is, as late as |
889 |
possible at runtime, and not e.g. during initialisation of your |
890 |
module. |
891 |
|
892 |
The effect of calling this function is as if a watcher had been |
893 |
created (specifically, actions that happen "when the first watcher |
894 |
is created" happen when calling detetc as well). |
895 |
|
896 |
If you need to do some initialisation before AnyEvent watchers are |
897 |
created, use "post_detect". |
898 |
|
899 |
$guard = AnyEvent::post_detect { BLOCK } |
900 |
Arranges for the code block to be executed as soon as the event |
901 |
model is autodetected (or immediately if that has already happened). |
902 |
|
903 |
The block will be executed *after* the actual backend has been |
904 |
detected ($AnyEvent::MODEL is set), but *before* any watchers have |
905 |
been created, so it is possible to e.g. patch @AnyEvent::ISA or do |
906 |
other initialisations - see the sources of AnyEvent::Strict or |
907 |
AnyEvent::AIO to see how this is used. |
908 |
|
909 |
The most common usage is to create some global watchers, without |
910 |
forcing event module detection too early, for example, AnyEvent::AIO |
911 |
creates and installs the global IO::AIO watcher in a "post_detect" |
912 |
block to avoid autodetecting the event module at load time. |
913 |
|
914 |
If called in scalar or list context, then it creates and returns an |
915 |
object that automatically removes the callback again when it is |
916 |
destroyed (or "undef" when the hook was immediately executed). See |
917 |
AnyEvent::AIO for a case where this is useful. |
918 |
|
919 |
Example: Create a watcher for the IO::AIO module and store it in |
920 |
$WATCHER, but do so only do so after the event loop is initialised. |
921 |
|
922 |
our WATCHER; |
923 |
|
924 |
my $guard = AnyEvent::post_detect { |
925 |
$WATCHER = AnyEvent->io (fh => IO::AIO::poll_fileno, poll => 'r', cb => \&IO::AIO::poll_cb); |
926 |
}; |
927 |
|
928 |
# the ||= is important in case post_detect immediately runs the block, |
929 |
# as to not clobber the newly-created watcher. assigning both watcher and |
930 |
# post_detect guard to the same variable has the advantage of users being |
931 |
# able to just C<undef $WATCHER> if the watcher causes them grief. |
932 |
|
933 |
$WATCHER ||= $guard; |
934 |
|
935 |
@AnyEvent::post_detect |
936 |
If there are any code references in this array (you can "push" to it |
937 |
before or after loading AnyEvent), then they will be called directly |
938 |
after the event loop has been chosen. |
939 |
|
940 |
You should check $AnyEvent::MODEL before adding to this array, |
941 |
though: if it is defined then the event loop has already been |
942 |
detected, and the array will be ignored. |
943 |
|
944 |
Best use "AnyEvent::post_detect { BLOCK }" when your application |
945 |
allows it, as it takes care of these details. |
946 |
|
947 |
This variable is mainly useful for modules that can do something |
948 |
useful when AnyEvent is used and thus want to know when it is |
949 |
initialised, but do not need to even load it by default. This array |
950 |
provides the means to hook into AnyEvent passively, without loading |
951 |
it. |
952 |
|
953 |
Example: To load Coro::AnyEvent whenever Coro and AnyEvent are used |
954 |
together, you could put this into Coro (this is the actual code used |
955 |
by Coro to accomplish this): |
956 |
|
957 |
if (defined $AnyEvent::MODEL) { |
958 |
# AnyEvent already initialised, so load Coro::AnyEvent |
959 |
require Coro::AnyEvent; |
960 |
} else { |
961 |
# AnyEvent not yet initialised, so make sure to load Coro::AnyEvent |
962 |
# as soon as it is |
963 |
push @AnyEvent::post_detect, sub { require Coro::AnyEvent }; |
964 |
} |
965 |
|
966 |
AnyEvent::postpone { BLOCK } |
967 |
Arranges for the block to be executed as soon as possible, but not |
968 |
before the call itself returns. In practise, the block will be |
969 |
executed just before the event loop polls for new events, or shortly |
970 |
afterwards. |
971 |
|
972 |
This function never returns anything (to make the "return postpone { |
973 |
... }" idiom more useful. |
974 |
|
975 |
To understand the usefulness of this function, consider a function |
976 |
that asynchronously does something for you and returns some |
977 |
transaction object or guard to let you cancel the operation. For |
978 |
example, "AnyEvent::Socket::tcp_connect": |
979 |
|
980 |
# start a conenction attempt unless one is active |
981 |
$self->{connect_guard} ||= AnyEvent::Socket::tcp_connect "www.example.net", 80, sub { |
982 |
delete $self->{connect_guard}; |
983 |
... |
984 |
}; |
985 |
|
986 |
Imagine that this function could instantly call the callback, for |
987 |
example, because it detects an obvious error such as a negative port |
988 |
number. Invoking the callback before the function returns causes |
989 |
problems however: the callback will be called and will try to delete |
990 |
the guard object. But since the function hasn't returned yet, there |
991 |
is nothing to delete. When the function eventually returns it will |
992 |
assign the guard object to "$self->{connect_guard}", where it will |
993 |
likely never be deleted, so the program thinks it is still trying to |
994 |
connect. |
995 |
|
996 |
This is where "AnyEvent::postpone" should be used. Instead of |
997 |
calling the callback directly on error: |
998 |
|
999 |
$cb->(undef), return # signal error to callback, BAD! |
1000 |
if $some_error_condition; |
1001 |
|
1002 |
It should use "postpone": |
1003 |
|
1004 |
AnyEvent::postpone { $cb->(undef) }, return # signal error to callback, later |
1005 |
if $some_error_condition; |
1006 |
|
1007 |
WHAT TO DO IN A MODULE |
1008 |
As a module author, you should "use AnyEvent" and call AnyEvent methods |
1009 |
freely, but you should not load a specific event module or rely on it. |
1010 |
|
1011 |
Be careful when you create watchers in the module body - AnyEvent will |
1012 |
decide which event module to use as soon as the first method is called, |
1013 |
so by calling AnyEvent in your module body you force the user of your |
1014 |
module to load the event module first. |
1015 |
|
1016 |
Never call "->recv" on a condition variable unless you *know* that the |
1017 |
"->send" method has been called on it already. This is because it will |
1018 |
stall the whole program, and the whole point of using events is to stay |
1019 |
interactive. |
1020 |
|
1021 |
It is fine, however, to call "->recv" when the user of your module |
1022 |
requests it (i.e. if you create a http request object ad have a method |
1023 |
called "results" that returns the results, it may call "->recv" freely, |
1024 |
as the user of your module knows what she is doing. Always). |
1025 |
|
1026 |
WHAT TO DO IN THE MAIN PROGRAM |
1027 |
There will always be a single main program - the only place that should |
1028 |
dictate which event model to use. |
1029 |
|
1030 |
If the program is not event-based, it need not do anything special, even |
1031 |
when it depends on a module that uses an AnyEvent. If the program itself |
1032 |
uses AnyEvent, but does not care which event loop is used, all it needs |
1033 |
to do is "use AnyEvent". In either case, AnyEvent will choose the best |
1034 |
available loop implementation. |
1035 |
|
1036 |
If the main program relies on a specific event model - for example, in |
1037 |
Gtk2 programs you have to rely on the Glib module - you should load the |
1038 |
event module before loading AnyEvent or any module that uses it: |
1039 |
generally speaking, you should load it as early as possible. The reason |
1040 |
is that modules might create watchers when they are loaded, and AnyEvent |
1041 |
will decide on the event model to use as soon as it creates watchers, |
1042 |
and it might choose the wrong one unless you load the correct one |
1043 |
yourself. |
1044 |
|
1045 |
You can chose to use a pure-perl implementation by loading the |
1046 |
"AnyEvent::Loop" module, which gives you similar behaviour everywhere, |
1047 |
but letting AnyEvent chose the model is generally better. |
1048 |
|
1049 |
MAINLOOP EMULATION |
1050 |
Sometimes (often for short test scripts, or even standalone programs who |
1051 |
only want to use AnyEvent), you do not want to run a specific event |
1052 |
loop. |
1053 |
|
1054 |
In that case, you can use a condition variable like this: |
1055 |
|
1056 |
AnyEvent->condvar->recv; |
1057 |
|
1058 |
This has the effect of entering the event loop and looping forever. |
1059 |
|
1060 |
Note that usually your program has some exit condition, in which case it |
1061 |
is better to use the "traditional" approach of storing a condition |
1062 |
variable somewhere, waiting for it, and sending it when the program |
1063 |
should exit cleanly. |
1064 |
|
1065 |
OTHER MODULES |
1066 |
The following is a non-exhaustive list of additional modules that use |
1067 |
AnyEvent as a client and can therefore be mixed easily with other |
1068 |
AnyEvent modules and other event loops in the same program. Some of the |
1069 |
modules come as part of AnyEvent, the others are available via CPAN. |
1070 |
|
1071 |
AnyEvent::Util |
1072 |
Contains various utility functions that replace often-used blocking |
1073 |
functions such as "inet_aton" with event/callback-based versions. |
1074 |
|
1075 |
AnyEvent::Socket |
1076 |
Provides various utility functions for (internet protocol) sockets, |
1077 |
addresses and name resolution. Also functions to create non-blocking |
1078 |
tcp connections or tcp servers, with IPv6 and SRV record support and |
1079 |
more. |
1080 |
|
1081 |
AnyEvent::Handle |
1082 |
Provide read and write buffers, manages watchers for reads and |
1083 |
writes, supports raw and formatted I/O, I/O queued and fully |
1084 |
transparent and non-blocking SSL/TLS (via AnyEvent::TLS). |
1085 |
|
1086 |
AnyEvent::DNS |
1087 |
Provides rich asynchronous DNS resolver capabilities. |
1088 |
|
1089 |
AnyEvent::HTTP, AnyEvent::IRC, AnyEvent::XMPP, AnyEvent::GPSD, |
1090 |
AnyEvent::IGS, AnyEvent::FCP |
1091 |
Implement event-based interfaces to the protocols of the same name |
1092 |
(for the curious, IGS is the International Go Server and FCP is the |
1093 |
Freenet Client Protocol). |
1094 |
|
1095 |
AnyEvent::Handle::UDP |
1096 |
Here be danger! |
1097 |
|
1098 |
As Pauli would put it, "Not only is it not right, it's not even |
1099 |
wrong!" - there are so many things wrong with AnyEvent::Handle::UDP, |
1100 |
most notably its use of a stream-based API with a protocol that |
1101 |
isn't streamable, that the only way to improve it is to delete it. |
1102 |
|
1103 |
It features data corruption (but typically only under load) and |
1104 |
general confusion. On top, the author is not only clueless about UDP |
1105 |
but also fact-resistant - some gems of his understanding: "connect |
1106 |
doesn't work with UDP", "UDP packets are not IP packets", "UDP only |
1107 |
has datagrams, not packets", "I don't need to implement proper error |
1108 |
checking as UDP doesn't support error checking" and so on - he |
1109 |
doesn't even understand what's wrong with his module when it is |
1110 |
explained to him. |
1111 |
|
1112 |
AnyEvent::DBI |
1113 |
Executes DBI requests asynchronously in a proxy process for you, |
1114 |
notifying you in an event-based way when the operation is finished. |
1115 |
|
1116 |
AnyEvent::AIO |
1117 |
Truly asynchronous (as opposed to non-blocking) I/O, should be in |
1118 |
the toolbox of every event programmer. AnyEvent::AIO transparently |
1119 |
fuses IO::AIO and AnyEvent together, giving AnyEvent access to |
1120 |
event-based file I/O, and much more. |
1121 |
|
1122 |
AnyEvent::HTTPD |
1123 |
A simple embedded webserver. |
1124 |
|
1125 |
AnyEvent::FastPing |
1126 |
The fastest ping in the west. |
1127 |
|
1128 |
Coro |
1129 |
Has special support for AnyEvent via Coro::AnyEvent. |
1130 |
|
1131 |
SIMPLIFIED AE API |
1132 |
Starting with version 5.0, AnyEvent officially supports a second, much |
1133 |
simpler, API that is designed to reduce the calling, typing and memory |
1134 |
overhead by using function call syntax and a fixed number of parameters. |
1135 |
|
1136 |
See the AE manpage for details. |
1137 |
|
1138 |
ERROR AND EXCEPTION HANDLING |
1139 |
In general, AnyEvent does not do any error handling - it relies on the |
1140 |
caller to do that if required. The AnyEvent::Strict module (see also the |
1141 |
"PERL_ANYEVENT_STRICT" environment variable, below) provides strict |
1142 |
checking of all AnyEvent methods, however, which is highly useful during |
1143 |
development. |
1144 |
|
1145 |
As for exception handling (i.e. runtime errors and exceptions thrown |
1146 |
while executing a callback), this is not only highly event-loop |
1147 |
specific, but also not in any way wrapped by this module, as this is the |
1148 |
job of the main program. |
1149 |
|
1150 |
The pure perl event loop simply re-throws the exception (usually within |
1151 |
"condvar->recv"), the Event and EV modules call "$Event/EV::DIED->()", |
1152 |
Glib uses "install_exception_handler" and so on. |
1153 |
|
1154 |
ENVIRONMENT VARIABLES |
1155 |
The following environment variables are used by this module or its |
1156 |
submodules. |
1157 |
|
1158 |
Note that AnyEvent will remove *all* environment variables starting with |
1159 |
"PERL_ANYEVENT_" from %ENV when it is loaded while taint mode is |
1160 |
enabled. |
1161 |
|
1162 |
"PERL_ANYEVENT_VERBOSE" |
1163 |
By default, AnyEvent will be completely silent except in fatal |
1164 |
conditions. You can set this environment variable to make AnyEvent |
1165 |
more talkative. |
1166 |
|
1167 |
When set to 1 or higher, causes AnyEvent to warn about unexpected |
1168 |
conditions, such as not being able to load the event model specified |
1169 |
by "PERL_ANYEVENT_MODEL". |
1170 |
|
1171 |
When set to 2 or higher, cause AnyEvent to report to STDERR which |
1172 |
event model it chooses. |
1173 |
|
1174 |
When set to 8 or higher, then AnyEvent will report extra information |
1175 |
on which optional modules it loads and how it implements certain |
1176 |
features. |
1177 |
|
1178 |
"PERL_ANYEVENT_STRICT" |
1179 |
AnyEvent does not do much argument checking by default, as thorough |
1180 |
argument checking is very costly. Setting this variable to a true |
1181 |
value will cause AnyEvent to load "AnyEvent::Strict" and then to |
1182 |
thoroughly check the arguments passed to most method calls. If it |
1183 |
finds any problems, it will croak. |
1184 |
|
1185 |
In other words, enables "strict" mode. |
1186 |
|
1187 |
Unlike "use strict" (or its modern cousin, "use common::sense", it |
1188 |
is definitely recommended to keep it off in production. Keeping |
1189 |
"PERL_ANYEVENT_STRICT=1" in your environment while developing |
1190 |
programs can be very useful, however. |
1191 |
|
1192 |
"PERL_ANYEVENT_DEBUG_SHELL" |
1193 |
If this env variable is set, then its contents will be interpreted |
1194 |
by "AnyEvent::Socket::parse_hostport" (after replacing every |
1195 |
occurance of $$ by the process pid) and an "AnyEvent::Debug::shell" |
1196 |
is bound on that port. The shell object is saved in |
1197 |
$AnyEvent::Debug::SHELL. |
1198 |
|
1199 |
This takes place when the first watcher is created. |
1200 |
|
1201 |
For example, to bind a debug shell on a unix domain socket in |
1202 |
/tmp/debug<pid>.sock, you could use this: |
1203 |
|
1204 |
PERL_ANYEVENT_DEBUG_SHELL=unix/:/tmp/debug\$\$.sock perlprog |
1205 |
|
1206 |
Note that creating sockets in /tmp is very unsafe on multiuser |
1207 |
systems. |
1208 |
|
1209 |
"PERL_ANYEVENT_DEBUG_WRAP" |
1210 |
Can be set to 0, 1 or 2 and enables wrapping of all watchers for |
1211 |
debugging purposes. See "AnyEvent::Debug::wrap" for details. |
1212 |
|
1213 |
"PERL_ANYEVENT_MODEL" |
1214 |
This can be used to specify the event model to be used by AnyEvent, |
1215 |
before auto detection and -probing kicks in. |
1216 |
|
1217 |
It normally is a string consisting entirely of ASCII letters (e.g. |
1218 |
"EV" or "IOAsync"). The string "AnyEvent::Impl::" gets prepended and |
1219 |
the resulting module name is loaded and - if the load was successful |
1220 |
- used as event model backend. If it fails to load then AnyEvent |
1221 |
will proceed with auto detection and -probing. |
1222 |
|
1223 |
If the string ends with "::" instead (e.g. "AnyEvent::Impl::EV::") |
1224 |
then nothing gets prepended and the module name is used as-is (hint: |
1225 |
"::" at the end of a string designates a module name and quotes it |
1226 |
appropriately). |
1227 |
|
1228 |
For example, to force the pure perl model (AnyEvent::Loop::Perl) you |
1229 |
could start your program like this: |
1230 |
|
1231 |
PERL_ANYEVENT_MODEL=Perl perl ... |
1232 |
|
1233 |
"PERL_ANYEVENT_PROTOCOLS" |
1234 |
Used by both AnyEvent::DNS and AnyEvent::Socket to determine |
1235 |
preferences for IPv4 or IPv6. The default is unspecified (and might |
1236 |
change, or be the result of auto probing). |
1237 |
|
1238 |
Must be set to a comma-separated list of protocols or address |
1239 |
families, current supported: "ipv4" and "ipv6". Only protocols |
1240 |
mentioned will be used, and preference will be given to protocols |
1241 |
mentioned earlier in the list. |
1242 |
|
1243 |
This variable can effectively be used for denial-of-service attacks |
1244 |
against local programs (e.g. when setuid), although the impact is |
1245 |
likely small, as the program has to handle conenction and other |
1246 |
failures anyways. |
1247 |
|
1248 |
Examples: "PERL_ANYEVENT_PROTOCOLS=ipv4,ipv6" - prefer IPv4 over |
1249 |
IPv6, but support both and try to use both. |
1250 |
"PERL_ANYEVENT_PROTOCOLS=ipv4" - only support IPv4, never try to |
1251 |
resolve or contact IPv6 addresses. |
1252 |
"PERL_ANYEVENT_PROTOCOLS=ipv6,ipv4" support either IPv4 or IPv6, but |
1253 |
prefer IPv6 over IPv4. |
1254 |
|
1255 |
"PERL_ANYEVENT_EDNS0" |
1256 |
Used by AnyEvent::DNS to decide whether to use the EDNS0 extension |
1257 |
for DNS. This extension is generally useful to reduce DNS traffic, |
1258 |
but some (broken) firewalls drop such DNS packets, which is why it |
1259 |
is off by default. |
1260 |
|
1261 |
Setting this variable to 1 will cause AnyEvent::DNS to announce |
1262 |
EDNS0 in its DNS requests. |
1263 |
|
1264 |
"PERL_ANYEVENT_MAX_FORKS" |
1265 |
The maximum number of child processes that |
1266 |
"AnyEvent::Util::fork_call" will create in parallel. |
1267 |
|
1268 |
"PERL_ANYEVENT_MAX_OUTSTANDING_DNS" |
1269 |
The default value for the "max_outstanding" parameter for the |
1270 |
default DNS resolver - this is the maximum number of parallel DNS |
1271 |
requests that are sent to the DNS server. |
1272 |
|
1273 |
"PERL_ANYEVENT_RESOLV_CONF" |
1274 |
The file to use instead of /etc/resolv.conf (or OS-specific |
1275 |
configuration) in the default resolver. When set to the empty |
1276 |
string, no default config will be used. |
1277 |
|
1278 |
"PERL_ANYEVENT_CA_FILE", "PERL_ANYEVENT_CA_PATH". |
1279 |
When neither "ca_file" nor "ca_path" was specified during |
1280 |
AnyEvent::TLS context creation, and either of these environment |
1281 |
variables exist, they will be used to specify CA certificate |
1282 |
locations instead of a system-dependent default. |
1283 |
|
1284 |
"PERL_ANYEVENT_AVOID_GUARD" and "PERL_ANYEVENT_AVOID_ASYNC_INTERRUPT" |
1285 |
When these are set to 1, then the respective modules are not loaded. |
1286 |
Mostly good for testing AnyEvent itself. |
1287 |
|
1288 |
SUPPLYING YOUR OWN EVENT MODEL INTERFACE |
1289 |
This is an advanced topic that you do not normally need to use AnyEvent |
1290 |
in a module. This section is only of use to event loop authors who want |
1291 |
to provide AnyEvent compatibility. |
1292 |
|
1293 |
If you need to support another event library which isn't directly |
1294 |
supported by AnyEvent, you can supply your own interface to it by |
1295 |
pushing, before the first watcher gets created, the package name of the |
1296 |
event module and the package name of the interface to use onto |
1297 |
@AnyEvent::REGISTRY. You can do that before and even without loading |
1298 |
AnyEvent, so it is reasonably cheap. |
1299 |
|
1300 |
Example: |
1301 |
|
1302 |
push @AnyEvent::REGISTRY, [urxvt => urxvt::anyevent::]; |
1303 |
|
1304 |
This tells AnyEvent to (literally) use the "urxvt::anyevent::" |
1305 |
package/class when it finds the "urxvt" package/module is already |
1306 |
loaded. |
1307 |
|
1308 |
When AnyEvent is loaded and asked to find a suitable event model, it |
1309 |
will first check for the presence of urxvt by trying to "use" the |
1310 |
"urxvt::anyevent" module. |
1311 |
|
1312 |
The class should provide implementations for all watcher types. See |
1313 |
AnyEvent::Impl::EV (source code), AnyEvent::Impl::Glib (Source code) and |
1314 |
so on for actual examples. Use "perldoc -m AnyEvent::Impl::Glib" to see |
1315 |
the sources. |
1316 |
|
1317 |
If you don't provide "signal" and "child" watchers than AnyEvent will |
1318 |
provide suitable (hopefully) replacements. |
1319 |
|
1320 |
The above example isn't fictitious, the *rxvt-unicode* (a.k.a. urxvt) |
1321 |
terminal emulator uses the above line as-is. An interface isn't included |
1322 |
in AnyEvent because it doesn't make sense outside the embedded |
1323 |
interpreter inside *rxvt-unicode*, and it is updated and maintained as |
1324 |
part of the *rxvt-unicode* distribution. |
1325 |
|
1326 |
*rxvt-unicode* also cheats a bit by not providing blocking access to |
1327 |
condition variables: code blocking while waiting for a condition will |
1328 |
"die". This still works with most modules/usages, and blocking calls |
1329 |
must not be done in an interactive application, so it makes sense. |
1330 |
|
1331 |
EXAMPLE PROGRAM |
1332 |
The following program uses an I/O watcher to read data from STDIN, a |
1333 |
timer to display a message once per second, and a condition variable to |
1334 |
quit the program when the user enters quit: |
1335 |
|
1336 |
use AnyEvent; |
1337 |
|
1338 |
my $cv = AnyEvent->condvar; |
1339 |
|
1340 |
my $io_watcher = AnyEvent->io ( |
1341 |
fh => \*STDIN, |
1342 |
poll => 'r', |
1343 |
cb => sub { |
1344 |
warn "io event <$_[0]>\n"; # will always output <r> |
1345 |
chomp (my $input = <STDIN>); # read a line |
1346 |
warn "read: $input\n"; # output what has been read |
1347 |
$cv->send if $input =~ /^q/i; # quit program if /^q/i |
1348 |
}, |
1349 |
); |
1350 |
|
1351 |
my $time_watcher = AnyEvent->timer (after => 1, interval => 1, cb => sub { |
1352 |
warn "timeout\n"; # print 'timeout' at most every second |
1353 |
}); |
1354 |
|
1355 |
$cv->recv; # wait until user enters /^q/i |
1356 |
|
1357 |
REAL-WORLD EXAMPLE |
1358 |
Consider the Net::FCP module. It features (among others) the following |
1359 |
API calls, which are to freenet what HTTP GET requests are to http: |
1360 |
|
1361 |
my $data = $fcp->client_get ($url); # blocks |
1362 |
|
1363 |
my $transaction = $fcp->txn_client_get ($url); # does not block |
1364 |
$transaction->cb ( sub { ... } ); # set optional result callback |
1365 |
my $data = $transaction->result; # possibly blocks |
1366 |
|
1367 |
The "client_get" method works like "LWP::Simple::get": it requests the |
1368 |
given URL and waits till the data has arrived. It is defined to be: |
1369 |
|
1370 |
sub client_get { $_[0]->txn_client_get ($_[1])->result } |
1371 |
|
1372 |
And in fact is automatically generated. This is the blocking API of |
1373 |
Net::FCP, and it works as simple as in any other, similar, module. |
1374 |
|
1375 |
More complicated is "txn_client_get": It only creates a transaction |
1376 |
(completion, result, ...) object and initiates the transaction. |
1377 |
|
1378 |
my $txn = bless { }, Net::FCP::Txn::; |
1379 |
|
1380 |
It also creates a condition variable that is used to signal the |
1381 |
completion of the request: |
1382 |
|
1383 |
$txn->{finished} = AnyAvent->condvar; |
1384 |
|
1385 |
It then creates a socket in non-blocking mode. |
1386 |
|
1387 |
socket $txn->{fh}, ...; |
1388 |
fcntl $txn->{fh}, F_SETFL, O_NONBLOCK; |
1389 |
connect $txn->{fh}, ... |
1390 |
and !$!{EWOULDBLOCK} |
1391 |
and !$!{EINPROGRESS} |
1392 |
and Carp::croak "unable to connect: $!\n"; |
1393 |
|
1394 |
Then it creates a write-watcher which gets called whenever an error |
1395 |
occurs or the connection succeeds: |
1396 |
|
1397 |
$txn->{w} = AnyEvent->io (fh => $txn->{fh}, poll => 'w', cb => sub { $txn->fh_ready_w }); |
1398 |
|
1399 |
And returns this transaction object. The "fh_ready_w" callback gets |
1400 |
called as soon as the event loop detects that the socket is ready for |
1401 |
writing. |
1402 |
|
1403 |
The "fh_ready_w" method makes the socket blocking again, writes the |
1404 |
request data and replaces the watcher by a read watcher (waiting for |
1405 |
reply data). The actual code is more complicated, but that doesn't |
1406 |
matter for this example: |
1407 |
|
1408 |
fcntl $txn->{fh}, F_SETFL, 0; |
1409 |
syswrite $txn->{fh}, $txn->{request} |
1410 |
or die "connection or write error"; |
1411 |
$txn->{w} = AnyEvent->io (fh => $txn->{fh}, poll => 'r', cb => sub { $txn->fh_ready_r }); |
1412 |
|
1413 |
Again, "fh_ready_r" waits till all data has arrived, and then stores the |
1414 |
result and signals any possible waiters that the request has finished: |
1415 |
|
1416 |
sysread $txn->{fh}, $txn->{buf}, length $txn->{$buf}; |
1417 |
|
1418 |
if (end-of-file or data complete) { |
1419 |
$txn->{result} = $txn->{buf}; |
1420 |
$txn->{finished}->send; |
1421 |
$txb->{cb}->($txn) of $txn->{cb}; # also call callback |
1422 |
} |
1423 |
|
1424 |
The "result" method, finally, just waits for the finished signal (if the |
1425 |
request was already finished, it doesn't wait, of course, and returns |
1426 |
the data: |
1427 |
|
1428 |
$txn->{finished}->recv; |
1429 |
return $txn->{result}; |
1430 |
|
1431 |
The actual code goes further and collects all errors ("die"s, |
1432 |
exceptions) that occurred during request processing. The "result" method |
1433 |
detects whether an exception as thrown (it is stored inside the $txn |
1434 |
object) and just throws the exception, which means connection errors and |
1435 |
other problems get reported to the code that tries to use the result, |
1436 |
not in a random callback. |
1437 |
|
1438 |
All of this enables the following usage styles: |
1439 |
|
1440 |
1. Blocking: |
1441 |
|
1442 |
my $data = $fcp->client_get ($url); |
1443 |
|
1444 |
2. Blocking, but running in parallel: |
1445 |
|
1446 |
my @datas = map $_->result, |
1447 |
map $fcp->txn_client_get ($_), |
1448 |
@urls; |
1449 |
|
1450 |
Both blocking examples work without the module user having to know |
1451 |
anything about events. |
1452 |
|
1453 |
3a. Event-based in a main program, using any supported event module: |
1454 |
|
1455 |
use EV; |
1456 |
|
1457 |
$fcp->txn_client_get ($url)->cb (sub { |
1458 |
my $txn = shift; |
1459 |
my $data = $txn->result; |
1460 |
... |
1461 |
}); |
1462 |
|
1463 |
EV::loop; |
1464 |
|
1465 |
3b. The module user could use AnyEvent, too: |
1466 |
|
1467 |
use AnyEvent; |
1468 |
|
1469 |
my $quit = AnyEvent->condvar; |
1470 |
|
1471 |
$fcp->txn_client_get ($url)->cb (sub { |
1472 |
... |
1473 |
$quit->send; |
1474 |
}); |
1475 |
|
1476 |
$quit->recv; |
1477 |
|
1478 |
BENCHMARKS |
1479 |
To give you an idea of the performance and overheads that AnyEvent adds |
1480 |
over the event loops themselves and to give you an impression of the |
1481 |
speed of various event loops I prepared some benchmarks. |
1482 |
|
1483 |
BENCHMARKING ANYEVENT OVERHEAD |
1484 |
Here is a benchmark of various supported event models used natively and |
1485 |
through AnyEvent. The benchmark creates a lot of timers (with a zero |
1486 |
timeout) and I/O watchers (watching STDOUT, a pty, to become writable, |
1487 |
which it is), lets them fire exactly once and destroys them again. |
1488 |
|
1489 |
Source code for this benchmark is found as eg/bench in the AnyEvent |
1490 |
distribution. It uses the AE interface, which makes a real difference |
1491 |
for the EV and Perl backends only. |
1492 |
|
1493 |
Explanation of the columns |
1494 |
*watcher* is the number of event watchers created/destroyed. Since |
1495 |
different event models feature vastly different performances, each event |
1496 |
loop was given a number of watchers so that overall runtime is |
1497 |
acceptable and similar between tested event loop (and keep them from |
1498 |
crashing): Glib would probably take thousands of years if asked to |
1499 |
process the same number of watchers as EV in this benchmark. |
1500 |
|
1501 |
*bytes* is the number of bytes (as measured by the resident set size, |
1502 |
RSS) consumed by each watcher. This method of measuring captures both C |
1503 |
and Perl-based overheads. |
1504 |
|
1505 |
*create* is the time, in microseconds (millionths of seconds), that it |
1506 |
takes to create a single watcher. The callback is a closure shared |
1507 |
between all watchers, to avoid adding memory overhead. That means |
1508 |
closure creation and memory usage is not included in the figures. |
1509 |
|
1510 |
*invoke* is the time, in microseconds, used to invoke a simple callback. |
1511 |
The callback simply counts down a Perl variable and after it was invoked |
1512 |
"watcher" times, it would "->send" a condvar once to signal the end of |
1513 |
this phase. |
1514 |
|
1515 |
*destroy* is the time, in microseconds, that it takes to destroy a |
1516 |
single watcher. |
1517 |
|
1518 |
Results |
1519 |
name watchers bytes create invoke destroy comment |
1520 |
EV/EV 100000 223 0.47 0.43 0.27 EV native interface |
1521 |
EV/Any 100000 223 0.48 0.42 0.26 EV + AnyEvent watchers |
1522 |
Coro::EV/Any 100000 223 0.47 0.42 0.26 coroutines + Coro::Signal |
1523 |
Perl/Any 100000 431 2.70 0.74 0.92 pure perl implementation |
1524 |
Event/Event 16000 516 31.16 31.84 0.82 Event native interface |
1525 |
Event/Any 16000 1203 42.61 34.79 1.80 Event + AnyEvent watchers |
1526 |
IOAsync/Any 16000 1911 41.92 27.45 16.81 via IO::Async::Loop::IO_Poll |
1527 |
IOAsync/Any 16000 1726 40.69 26.37 15.25 via IO::Async::Loop::Epoll |
1528 |
Glib/Any 16000 1118 89.00 12.57 51.17 quadratic behaviour |
1529 |
Tk/Any 2000 1346 20.96 10.75 8.00 SEGV with >> 2000 watchers |
1530 |
POE/Any 2000 6951 108.97 795.32 14.24 via POE::Loop::Event |
1531 |
POE/Any 2000 6648 94.79 774.40 575.51 via POE::Loop::Select |
1532 |
|
1533 |
Discussion |
1534 |
The benchmark does *not* measure scalability of the event loop very |
1535 |
well. For example, a select-based event loop (such as the pure perl one) |
1536 |
can never compete with an event loop that uses epoll when the number of |
1537 |
file descriptors grows high. In this benchmark, all events become ready |
1538 |
at the same time, so select/poll-based implementations get an unnatural |
1539 |
speed boost. |
1540 |
|
1541 |
Also, note that the number of watchers usually has a nonlinear effect on |
1542 |
overall speed, that is, creating twice as many watchers doesn't take |
1543 |
twice the time - usually it takes longer. This puts event loops tested |
1544 |
with a higher number of watchers at a disadvantage. |
1545 |
|
1546 |
To put the range of results into perspective, consider that on the |
1547 |
benchmark machine, handling an event takes roughly 1600 CPU cycles with |
1548 |
EV, 3100 CPU cycles with AnyEvent's pure perl loop and almost 3000000 |
1549 |
CPU cycles with POE. |
1550 |
|
1551 |
"EV" is the sole leader regarding speed and memory use, which are both |
1552 |
maximal/minimal, respectively. When using the AE API there is zero |
1553 |
overhead (when going through the AnyEvent API create is about 5-6 times |
1554 |
slower, with other times being equal, so still uses far less memory than |
1555 |
any other event loop and is still faster than Event natively). |
1556 |
|
1557 |
The pure perl implementation is hit in a few sweet spots (both the |
1558 |
constant timeout and the use of a single fd hit optimisations in the |
1559 |
perl interpreter and the backend itself). Nevertheless this shows that |
1560 |
it adds very little overhead in itself. Like any select-based backend |
1561 |
its performance becomes really bad with lots of file descriptors (and |
1562 |
few of them active), of course, but this was not subject of this |
1563 |
benchmark. |
1564 |
|
1565 |
The "Event" module has a relatively high setup and callback invocation |
1566 |
cost, but overall scores in on the third place. |
1567 |
|
1568 |
"IO::Async" performs admirably well, about on par with "Event", even |
1569 |
when using its pure perl backend. |
1570 |
|
1571 |
"Glib"'s memory usage is quite a bit higher, but it features a faster |
1572 |
callback invocation and overall ends up in the same class as "Event". |
1573 |
However, Glib scales extremely badly, doubling the number of watchers |
1574 |
increases the processing time by more than a factor of four, making it |
1575 |
completely unusable when using larger numbers of watchers (note that |
1576 |
only a single file descriptor was used in the benchmark, so |
1577 |
inefficiencies of "poll" do not account for this). |
1578 |
|
1579 |
The "Tk" adaptor works relatively well. The fact that it crashes with |
1580 |
more than 2000 watchers is a big setback, however, as correctness takes |
1581 |
precedence over speed. Nevertheless, its performance is surprising, as |
1582 |
the file descriptor is dup()ed for each watcher. This shows that the |
1583 |
dup() employed by some adaptors is not a big performance issue (it does |
1584 |
incur a hidden memory cost inside the kernel which is not reflected in |
1585 |
the figures above). |
1586 |
|
1587 |
"POE", regardless of underlying event loop (whether using its pure perl |
1588 |
select-based backend or the Event module, the POE-EV backend couldn't be |
1589 |
tested because it wasn't working) shows abysmal performance and memory |
1590 |
usage with AnyEvent: Watchers use almost 30 times as much memory as EV |
1591 |
watchers, and 10 times as much memory as Event (the high memory |
1592 |
requirements are caused by requiring a session for each watcher). |
1593 |
Watcher invocation speed is almost 900 times slower than with AnyEvent's |
1594 |
pure perl implementation. |
1595 |
|
1596 |
The design of the POE adaptor class in AnyEvent can not really account |
1597 |
for the performance issues, though, as session creation overhead is |
1598 |
small compared to execution of the state machine, which is coded pretty |
1599 |
optimally within AnyEvent::Impl::POE (and while everybody agrees that |
1600 |
using multiple sessions is not a good approach, especially regarding |
1601 |
memory usage, even the author of POE could not come up with a faster |
1602 |
design). |
1603 |
|
1604 |
Summary |
1605 |
* Using EV through AnyEvent is faster than any other event loop (even |
1606 |
when used without AnyEvent), but most event loops have acceptable |
1607 |
performance with or without AnyEvent. |
1608 |
|
1609 |
* The overhead AnyEvent adds is usually much smaller than the overhead |
1610 |
of the actual event loop, only with extremely fast event loops such |
1611 |
as EV adds AnyEvent significant overhead. |
1612 |
|
1613 |
* You should avoid POE like the plague if you want performance or |
1614 |
reasonable memory usage. |
1615 |
|
1616 |
BENCHMARKING THE LARGE SERVER CASE |
1617 |
This benchmark actually benchmarks the event loop itself. It works by |
1618 |
creating a number of "servers": each server consists of a socket pair, a |
1619 |
timeout watcher that gets reset on activity (but never fires), and an |
1620 |
I/O watcher waiting for input on one side of the socket. Each time the |
1621 |
socket watcher reads a byte it will write that byte to a random other |
1622 |
"server". |
1623 |
|
1624 |
The effect is that there will be a lot of I/O watchers, only part of |
1625 |
which are active at any one point (so there is a constant number of |
1626 |
active fds for each loop iteration, but which fds these are is random). |
1627 |
The timeout is reset each time something is read because that reflects |
1628 |
how most timeouts work (and puts extra pressure on the event loops). |
1629 |
|
1630 |
In this benchmark, we use 10000 socket pairs (20000 sockets), of which |
1631 |
100 (1%) are active. This mirrors the activity of large servers with |
1632 |
many connections, most of which are idle at any one point in time. |
1633 |
|
1634 |
Source code for this benchmark is found as eg/bench2 in the AnyEvent |
1635 |
distribution. It uses the AE interface, which makes a real difference |
1636 |
for the EV and Perl backends only. |
1637 |
|
1638 |
Explanation of the columns |
1639 |
*sockets* is the number of sockets, and twice the number of "servers" |
1640 |
(as each server has a read and write socket end). |
1641 |
|
1642 |
*create* is the time it takes to create a socket pair (which is |
1643 |
nontrivial) and two watchers: an I/O watcher and a timeout watcher. |
1644 |
|
1645 |
*request*, the most important value, is the time it takes to handle a |
1646 |
single "request", that is, reading the token from the pipe and |
1647 |
forwarding it to another server. This includes deleting the old timeout |
1648 |
and creating a new one that moves the timeout into the future. |
1649 |
|
1650 |
Results |
1651 |
name sockets create request |
1652 |
EV 20000 62.66 7.99 |
1653 |
Perl 20000 68.32 32.64 |
1654 |
IOAsync 20000 174.06 101.15 epoll |
1655 |
IOAsync 20000 174.67 610.84 poll |
1656 |
Event 20000 202.69 242.91 |
1657 |
Glib 20000 557.01 1689.52 |
1658 |
POE 20000 341.54 12086.32 uses POE::Loop::Event |
1659 |
|
1660 |
Discussion |
1661 |
This benchmark *does* measure scalability and overall performance of the |
1662 |
particular event loop. |
1663 |
|
1664 |
EV is again fastest. Since it is using epoll on my system, the setup |
1665 |
time is relatively high, though. |
1666 |
|
1667 |
Perl surprisingly comes second. It is much faster than the C-based event |
1668 |
loops Event and Glib. |
1669 |
|
1670 |
IO::Async performs very well when using its epoll backend, and still |
1671 |
quite good compared to Glib when using its pure perl backend. |
1672 |
|
1673 |
Event suffers from high setup time as well (look at its code and you |
1674 |
will understand why). Callback invocation also has a high overhead |
1675 |
compared to the "$_->() for .."-style loop that the Perl event loop |
1676 |
uses. Event uses select or poll in basically all documented |
1677 |
configurations. |
1678 |
|
1679 |
Glib is hit hard by its quadratic behaviour w.r.t. many watchers. It |
1680 |
clearly fails to perform with many filehandles or in busy servers. |
1681 |
|
1682 |
POE is still completely out of the picture, taking over 1000 times as |
1683 |
long as EV, and over 100 times as long as the Perl implementation, even |
1684 |
though it uses a C-based event loop in this case. |
1685 |
|
1686 |
Summary |
1687 |
* The pure perl implementation performs extremely well. |
1688 |
|
1689 |
* Avoid Glib or POE in large projects where performance matters. |
1690 |
|
1691 |
BENCHMARKING SMALL SERVERS |
1692 |
While event loops should scale (and select-based ones do not...) even to |
1693 |
large servers, most programs we (or I :) actually write have only a few |
1694 |
I/O watchers. |
1695 |
|
1696 |
In this benchmark, I use the same benchmark program as in the large |
1697 |
server case, but it uses only eight "servers", of which three are active |
1698 |
at any one time. This should reflect performance for a small server |
1699 |
relatively well. |
1700 |
|
1701 |
The columns are identical to the previous table. |
1702 |
|
1703 |
Results |
1704 |
name sockets create request |
1705 |
EV 16 20.00 6.54 |
1706 |
Perl 16 25.75 12.62 |
1707 |
Event 16 81.27 35.86 |
1708 |
Glib 16 32.63 15.48 |
1709 |
POE 16 261.87 276.28 uses POE::Loop::Event |
1710 |
|
1711 |
Discussion |
1712 |
The benchmark tries to test the performance of a typical small server. |
1713 |
While knowing how various event loops perform is interesting, keep in |
1714 |
mind that their overhead in this case is usually not as important, due |
1715 |
to the small absolute number of watchers (that is, you need efficiency |
1716 |
and speed most when you have lots of watchers, not when you only have a |
1717 |
few of them). |
1718 |
|
1719 |
EV is again fastest. |
1720 |
|
1721 |
Perl again comes second. It is noticeably faster than the C-based event |
1722 |
loops Event and Glib, although the difference is too small to really |
1723 |
matter. |
1724 |
|
1725 |
POE also performs much better in this case, but is is still far behind |
1726 |
the others. |
1727 |
|
1728 |
Summary |
1729 |
* C-based event loops perform very well with small number of watchers, |
1730 |
as the management overhead dominates. |
1731 |
|
1732 |
THE IO::Lambda BENCHMARK |
1733 |
Recently I was told about the benchmark in the IO::Lambda manpage, which |
1734 |
could be misinterpreted to make AnyEvent look bad. In fact, the |
1735 |
benchmark simply compares IO::Lambda with POE, and IO::Lambda looks |
1736 |
better (which shouldn't come as a surprise to anybody). As such, the |
1737 |
benchmark is fine, and mostly shows that the AnyEvent backend from |
1738 |
IO::Lambda isn't very optimal. But how would AnyEvent compare when used |
1739 |
without the extra baggage? To explore this, I wrote the equivalent |
1740 |
benchmark for AnyEvent. |
1741 |
|
1742 |
The benchmark itself creates an echo-server, and then, for 500 times, |
1743 |
connects to the echo server, sends a line, waits for the reply, and then |
1744 |
creates the next connection. This is a rather bad benchmark, as it |
1745 |
doesn't test the efficiency of the framework or much non-blocking I/O, |
1746 |
but it is a benchmark nevertheless. |
1747 |
|
1748 |
name runtime |
1749 |
Lambda/select 0.330 sec |
1750 |
+ optimized 0.122 sec |
1751 |
Lambda/AnyEvent 0.327 sec |
1752 |
+ optimized 0.138 sec |
1753 |
Raw sockets/select 0.077 sec |
1754 |
POE/select, components 0.662 sec |
1755 |
POE/select, raw sockets 0.226 sec |
1756 |
POE/select, optimized 0.404 sec |
1757 |
|
1758 |
AnyEvent/select/nb 0.085 sec |
1759 |
AnyEvent/EV/nb 0.068 sec |
1760 |
+state machine 0.134 sec |
1761 |
|
1762 |
The benchmark is also a bit unfair (my fault): the IO::Lambda/POE |
1763 |
benchmarks actually make blocking connects and use 100% blocking I/O, |
1764 |
defeating the purpose of an event-based solution. All of the newly |
1765 |
written AnyEvent benchmarks use 100% non-blocking connects (using |
1766 |
AnyEvent::Socket::tcp_connect and the asynchronous pure perl DNS |
1767 |
resolver), so AnyEvent is at a disadvantage here, as non-blocking |
1768 |
connects generally require a lot more bookkeeping and event handling |
1769 |
than blocking connects (which involve a single syscall only). |
1770 |
|
1771 |
The last AnyEvent benchmark additionally uses AnyEvent::Handle, which |
1772 |
offers similar expressive power as POE and IO::Lambda, using |
1773 |
conventional Perl syntax. This means that both the echo server and the |
1774 |
client are 100% non-blocking, further placing it at a disadvantage. |
1775 |
|
1776 |
As you can see, the AnyEvent + EV combination even beats the |
1777 |
hand-optimised "raw sockets benchmark", while AnyEvent + its pure perl |
1778 |
backend easily beats IO::Lambda and POE. |
1779 |
|
1780 |
And even the 100% non-blocking version written using the high-level (and |
1781 |
slow :) AnyEvent::Handle abstraction beats both POE and IO::Lambda |
1782 |
higher level ("unoptimised") abstractions by a large margin, even though |
1783 |
it does all of DNS, tcp-connect and socket I/O in a non-blocking way. |
1784 |
|
1785 |
The two AnyEvent benchmarks programs can be found as eg/ae0.pl and |
1786 |
eg/ae2.pl in the AnyEvent distribution, the remaining benchmarks are |
1787 |
part of the IO::Lambda distribution and were used without any changes. |
1788 |
|
1789 |
SIGNALS |
1790 |
AnyEvent currently installs handlers for these signals: |
1791 |
|
1792 |
SIGCHLD |
1793 |
A handler for "SIGCHLD" is installed by AnyEvent's child watcher |
1794 |
emulation for event loops that do not support them natively. Also, |
1795 |
some event loops install a similar handler. |
1796 |
|
1797 |
Additionally, when AnyEvent is loaded and SIGCHLD is set to IGNORE, |
1798 |
then AnyEvent will reset it to default, to avoid losing child exit |
1799 |
statuses. |
1800 |
|
1801 |
SIGPIPE |
1802 |
A no-op handler is installed for "SIGPIPE" when $SIG{PIPE} is |
1803 |
"undef" when AnyEvent gets loaded. |
1804 |
|
1805 |
The rationale for this is that AnyEvent users usually do not really |
1806 |
depend on SIGPIPE delivery (which is purely an optimisation for |
1807 |
shell use, or badly-written programs), but "SIGPIPE" can cause |
1808 |
spurious and rare program exits as a lot of people do not expect |
1809 |
"SIGPIPE" when writing to some random socket. |
1810 |
|
1811 |
The rationale for installing a no-op handler as opposed to ignoring |
1812 |
it is that this way, the handler will be restored to defaults on |
1813 |
exec. |
1814 |
|
1815 |
Feel free to install your own handler, or reset it to defaults. |
1816 |
|
1817 |
RECOMMENDED/OPTIONAL MODULES |
1818 |
One of AnyEvent's main goals is to be 100% Pure-Perl(tm): only perl (and |
1819 |
its built-in modules) are required to use it. |
1820 |
|
1821 |
That does not mean that AnyEvent won't take advantage of some additional |
1822 |
modules if they are installed. |
1823 |
|
1824 |
This section explains which additional modules will be used, and how |
1825 |
they affect AnyEvent's operation. |
1826 |
|
1827 |
Async::Interrupt |
1828 |
This slightly arcane module is used to implement fast signal |
1829 |
handling: To my knowledge, there is no way to do completely |
1830 |
race-free and quick signal handling in pure perl. To ensure that |
1831 |
signals still get delivered, AnyEvent will start an interval timer |
1832 |
to wake up perl (and catch the signals) with some delay (default is |
1833 |
10 seconds, look for $AnyEvent::MAX_SIGNAL_LATENCY). |
1834 |
|
1835 |
If this module is available, then it will be used to implement |
1836 |
signal catching, which means that signals will not be delayed, and |
1837 |
the event loop will not be interrupted regularly, which is more |
1838 |
efficient (and good for battery life on laptops). |
1839 |
|
1840 |
This affects not just the pure-perl event loop, but also other event |
1841 |
loops that have no signal handling on their own (e.g. Glib, Tk, Qt). |
1842 |
|
1843 |
Some event loops (POE, Event, Event::Lib) offer signal watchers |
1844 |
natively, and either employ their own workarounds (POE) or use |
1845 |
AnyEvent's workaround (using $AnyEvent::MAX_SIGNAL_LATENCY). |
1846 |
Installing Async::Interrupt does nothing for those backends. |
1847 |
|
1848 |
EV This module isn't really "optional", as it is simply one of the |
1849 |
backend event loops that AnyEvent can use. However, it is simply the |
1850 |
best event loop available in terms of features, speed and stability: |
1851 |
It supports the AnyEvent API optimally, implements all the watcher |
1852 |
types in XS, does automatic timer adjustments even when no monotonic |
1853 |
clock is available, can take avdantage of advanced kernel interfaces |
1854 |
such as "epoll" and "kqueue", and is the fastest backend *by far*. |
1855 |
You can even embed Glib/Gtk2 in it (or vice versa, see EV::Glib and |
1856 |
Glib::EV). |
1857 |
|
1858 |
If you only use backends that rely on another event loop (e.g. |
1859 |
"Tk"), then this module will do nothing for you. |
1860 |
|
1861 |
Guard |
1862 |
The guard module, when used, will be used to implement |
1863 |
"AnyEvent::Util::guard". This speeds up guards considerably (and |
1864 |
uses a lot less memory), but otherwise doesn't affect guard |
1865 |
operation much. It is purely used for performance. |
1866 |
|
1867 |
JSON and JSON::XS |
1868 |
One of these modules is required when you want to read or write JSON |
1869 |
data via AnyEvent::Handle. JSON is also written in pure-perl, but |
1870 |
can take advantage of the ultra-high-speed JSON::XS module when it |
1871 |
is installed. |
1872 |
|
1873 |
Net::SSLeay |
1874 |
Implementing TLS/SSL in Perl is certainly interesting, but not very |
1875 |
worthwhile: If this module is installed, then AnyEvent::Handle (with |
1876 |
the help of AnyEvent::TLS), gains the ability to do TLS/SSL. |
1877 |
|
1878 |
Time::HiRes |
1879 |
This module is part of perl since release 5.008. It will be used |
1880 |
when the chosen event library does not come with a timing source of |
1881 |
its own. The pure-perl event loop (AnyEvent::Loop) will additionally |
1882 |
load it to try to use a monotonic clock for timing stability. |
1883 |
|
1884 |
FORK |
1885 |
Most event libraries are not fork-safe. The ones who are usually are |
1886 |
because they rely on inefficient but fork-safe "select" or "poll" calls |
1887 |
- higher performance APIs such as BSD's kqueue or the dreaded Linux |
1888 |
epoll are usually badly thought-out hacks that are incompatible with |
1889 |
fork in one way or another. Only EV is fully fork-aware and ensures that |
1890 |
you continue event-processing in both parent and child (or both, if you |
1891 |
know what you are doing). |
1892 |
|
1893 |
This means that, in general, you cannot fork and do event processing in |
1894 |
the child if the event library was initialised before the fork (which |
1895 |
usually happens when the first AnyEvent watcher is created, or the |
1896 |
library is loaded). |
1897 |
|
1898 |
If you have to fork, you must either do so *before* creating your first |
1899 |
watcher OR you must not use AnyEvent at all in the child OR you must do |
1900 |
something completely out of the scope of AnyEvent. |
1901 |
|
1902 |
The problem of doing event processing in the parent *and* the child is |
1903 |
much more complicated: even for backends that *are* fork-aware or |
1904 |
fork-safe, their behaviour is not usually what you want: fork clones all |
1905 |
watchers, that means all timers, I/O watchers etc. are active in both |
1906 |
parent and child, which is almost never what you want. USing "exec" to |
1907 |
start worker children from some kind of manage rprocess is usually |
1908 |
preferred, because it is much easier and cleaner, at the expense of |
1909 |
having to have another binary. |
1910 |
|
1911 |
SECURITY CONSIDERATIONS |
1912 |
AnyEvent can be forced to load any event model via |
1913 |
$ENV{PERL_ANYEVENT_MODEL}. While this cannot (to my knowledge) be used |
1914 |
to execute arbitrary code or directly gain access, it can easily be used |
1915 |
to make the program hang or malfunction in subtle ways, as AnyEvent |
1916 |
watchers will not be active when the program uses a different event |
1917 |
model than specified in the variable. |
1918 |
|
1919 |
You can make AnyEvent completely ignore this variable by deleting it |
1920 |
before the first watcher gets created, e.g. with a "BEGIN" block: |
1921 |
|
1922 |
BEGIN { delete $ENV{PERL_ANYEVENT_MODEL} } |
1923 |
|
1924 |
use AnyEvent; |
1925 |
|
1926 |
Similar considerations apply to $ENV{PERL_ANYEVENT_VERBOSE}, as that can |
1927 |
be used to probe what backend is used and gain other information (which |
1928 |
is probably even less useful to an attacker than PERL_ANYEVENT_MODEL), |
1929 |
and $ENV{PERL_ANYEVENT_STRICT}. |
1930 |
|
1931 |
Note that AnyEvent will remove *all* environment variables starting with |
1932 |
"PERL_ANYEVENT_" from %ENV when it is loaded while taint mode is |
1933 |
enabled. |
1934 |
|
1935 |
BUGS |
1936 |
Perl 5.8 has numerous memleaks that sometimes hit this module and are |
1937 |
hard to work around. If you suffer from memleaks, first upgrade to Perl |
1938 |
5.10 and check wether the leaks still show up. (Perl 5.10.0 has other |
1939 |
annoying memleaks, such as leaking on "map" and "grep" but it is usually |
1940 |
not as pronounced). |
1941 |
|
1942 |
SEE ALSO |
1943 |
Tutorial/Introduction: AnyEvent::Intro. |
1944 |
|
1945 |
FAQ: AnyEvent::FAQ. |
1946 |
|
1947 |
Utility functions: AnyEvent::Util. |
1948 |
|
1949 |
Event modules: AnyEvent::Loop, EV, EV::Glib, Glib::EV, Event, |
1950 |
Glib::Event, Glib, Tk, Event::Lib, Qt, POE. |
1951 |
|
1952 |
Implementations: AnyEvent::Impl::EV, AnyEvent::Impl::Event, |
1953 |
AnyEvent::Impl::Glib, AnyEvent::Impl::Tk, AnyEvent::Impl::Perl, |
1954 |
AnyEvent::Impl::EventLib, AnyEvent::Impl::Qt, AnyEvent::Impl::POE, |
1955 |
AnyEvent::Impl::IOAsync, Anyevent::Impl::Irssi. |
1956 |
|
1957 |
Non-blocking file handles, sockets, TCP clients and servers: |
1958 |
AnyEvent::Handle, AnyEvent::Socket, AnyEvent::TLS. |
1959 |
|
1960 |
Asynchronous DNS: AnyEvent::DNS. |
1961 |
|
1962 |
Thread support: Coro, Coro::AnyEvent, Coro::EV, Coro::Event. |
1963 |
|
1964 |
Nontrivial usage examples: AnyEvent::GPSD, AnyEvent::IRC, |
1965 |
AnyEvent::HTTP. |
1966 |
|
1967 |
AUTHOR |
1968 |
Marc Lehmann <schmorp@schmorp.de> |
1969 |
http://home.schmorp.de/ |
1970 |
|