1 |
NAME |
2 |
Coro - the only real threads in perl |
3 |
|
4 |
SYNOPSIS |
5 |
use Coro; |
6 |
|
7 |
async { |
8 |
# some asynchronous thread of execution |
9 |
print "2\n"; |
10 |
cede; # yield back to main |
11 |
print "4\n"; |
12 |
}; |
13 |
print "1\n"; |
14 |
cede; # yield to coro |
15 |
print "3\n"; |
16 |
cede; # and again |
17 |
|
18 |
# use locking |
19 |
my $lock = new Coro::Semaphore; |
20 |
my $locked; |
21 |
|
22 |
$lock->down; |
23 |
$locked = 1; |
24 |
$lock->up; |
25 |
|
26 |
DESCRIPTION |
27 |
For a tutorial-style introduction, please read the Coro::Intro manpage. |
28 |
This manpage mainly contains reference information. |
29 |
|
30 |
This module collection manages continuations in general, most often in |
31 |
the form of cooperative threads (also called coros, or simply "coro" in |
32 |
the documentation). They are similar to kernel threads but don't (in |
33 |
general) run in parallel at the same time even on SMP machines. The |
34 |
specific flavor of thread offered by this module also guarantees you |
35 |
that it will not switch between threads unless necessary, at |
36 |
easily-identified points in your program, so locking and parallel access |
37 |
are rarely an issue, making thread programming much safer and easier |
38 |
than using other thread models. |
39 |
|
40 |
Unlike the so-called "Perl threads" (which are not actually real threads |
41 |
but only the windows process emulation (see section of same name for |
42 |
more details) ported to UNIX, and as such act as processes), Coro |
43 |
provides a full shared address space, which makes communication between |
44 |
threads very easy. And coro threads are fast, too: disabling the Windows |
45 |
process emulation code in your perl and using Coro can easily result in |
46 |
a two to four times speed increase for your programs. A parallel matrix |
47 |
multiplication benchmark (very communication-intensive) runs over 300 |
48 |
times faster on a single core than perls pseudo-threads on a quad core |
49 |
using all four cores. |
50 |
|
51 |
Coro achieves that by supporting multiple running interpreters that |
52 |
share data, which is especially useful to code pseudo-parallel processes |
53 |
and for event-based programming, such as multiple HTTP-GET requests |
54 |
running concurrently. See Coro::AnyEvent to learn more on how to |
55 |
integrate Coro into an event-based environment. |
56 |
|
57 |
In this module, a thread is defined as "callchain + lexical variables + |
58 |
some package variables + C stack), that is, a thread has its own |
59 |
callchain, its own set of lexicals and its own set of perls most |
60 |
important global variables (see Coro::State for more configuration and |
61 |
background info). |
62 |
|
63 |
See also the "SEE ALSO" section at the end of this document - the Coro |
64 |
module family is quite large. |
65 |
|
66 |
CORO THREAD LIFE CYCLE |
67 |
During the long and exciting (or not) life of a coro thread, it goes |
68 |
through a number of states: |
69 |
|
70 |
1. Creation |
71 |
The first thing in the life of a coro thread is it's creation - |
72 |
obviously. The typical way to create a thread is to call the "async |
73 |
BLOCK" function: |
74 |
|
75 |
async { |
76 |
# thread code goes here |
77 |
}; |
78 |
|
79 |
You can also pass arguments, which are put in @_: |
80 |
|
81 |
async { |
82 |
print $_[1]; # prints 2 |
83 |
} 1, 2, 3; |
84 |
|
85 |
This creates a new coro thread and puts it into the ready queue, |
86 |
meaning it will run as soon as the CPU is free for it. |
87 |
|
88 |
"async" will return a Coro object - you can store this for future |
89 |
reference or ignore it - a thread that is running, ready to run or |
90 |
waiting for some event is alive on it's own. |
91 |
|
92 |
Another way to create a thread is to call the "new" constructor with |
93 |
a code-reference: |
94 |
|
95 |
new Coro sub { |
96 |
# thread code goes here |
97 |
}, @optional_arguments; |
98 |
|
99 |
This is quite similar to calling "async", but the important |
100 |
difference is that the new thread is not put into the ready queue, |
101 |
so the thread will not run until somebody puts it there. "async" is, |
102 |
therefore, identical to this sequence: |
103 |
|
104 |
my $coro = new Coro sub { |
105 |
# thread code goes here |
106 |
}; |
107 |
$coro->ready; |
108 |
return $coro; |
109 |
|
110 |
2. Startup |
111 |
When a new coro thread is created, only a copy of the code reference |
112 |
and the arguments are stored, no extra memory for stacks and so on |
113 |
is allocated, keeping the coro thread in a low-memory state. |
114 |
|
115 |
Only when it actually starts executing will all the resources be |
116 |
finally allocated. |
117 |
|
118 |
The optional arguments specified at coro creation are available in |
119 |
@_, similar to function calls. |
120 |
|
121 |
3. Running / Blocking |
122 |
A lot can happen after the coro thread has started running. Quite |
123 |
usually, it will not run to the end in one go (because you could use |
124 |
a function instead), but it will give up the CPU regularly because |
125 |
it waits for external events. |
126 |
|
127 |
As long as a coro thread runs, its Coro object is available in the |
128 |
global variable $Coro::current. |
129 |
|
130 |
The low-level way to give up the CPU is to call the scheduler, which |
131 |
selects a new coro thread to run: |
132 |
|
133 |
Coro::schedule; |
134 |
|
135 |
Since running threads are not in the ready queue, calling the |
136 |
scheduler without doing anything else will block the coro thread |
137 |
forever - you need to arrange either for the coro to put woken up |
138 |
(readied) by some other event or some other thread, or you can put |
139 |
it into the ready queue before scheduling: |
140 |
|
141 |
# this is exactly what Coro::cede does |
142 |
$Coro::current->ready; |
143 |
Coro::schedule; |
144 |
|
145 |
All the higher-level synchronisation methods (Coro::Semaphore, |
146 |
Coro::rouse_*...) are actually implemented via "->ready" and |
147 |
"Coro::schedule". |
148 |
|
149 |
While the coro thread is running it also might get assigned a |
150 |
C-level thread, or the C-level thread might be unassigned from it, |
151 |
as the Coro runtime wishes. A C-level thread needs to be assigned |
152 |
when your perl thread calls into some C-level function and that |
153 |
function in turn calls perl and perl then wants to switch |
154 |
coroutines. This happens most often when you run an event loop and |
155 |
block in the callback, or when perl itself calls some function such |
156 |
as "AUTOLOAD" or methods via the "tie" mechanism. |
157 |
|
158 |
4. Termination |
159 |
Many threads actually terminate after some time. There are a number |
160 |
of ways to terminate a coro thread, the simplest is returning from |
161 |
the top-level code reference: |
162 |
|
163 |
async { |
164 |
# after returning from here, the coro thread is terminated |
165 |
}; |
166 |
|
167 |
async { |
168 |
return if 0.5 < rand; # terminate a little earlier, maybe |
169 |
print "got a chance to print this\n"; |
170 |
# or here |
171 |
}; |
172 |
|
173 |
Any values returned from the coroutine can be recovered using |
174 |
"->join": |
175 |
|
176 |
my $coro = async { |
177 |
"hello, world\n" # return a string |
178 |
}; |
179 |
|
180 |
my $hello_world = $coro->join; |
181 |
|
182 |
print $hello_world; |
183 |
|
184 |
Another way to terminate is to call "Coro::terminate", which at any |
185 |
subroutine call nesting level: |
186 |
|
187 |
async { |
188 |
Coro::terminate "return value 1", "return value 2"; |
189 |
}; |
190 |
|
191 |
Yet another way is to "->cancel" (or "->safe_cancel") the coro |
192 |
thread from another thread: |
193 |
|
194 |
my $coro = async { |
195 |
exit 1; |
196 |
}; |
197 |
|
198 |
$coro->cancel; # also accepts values for ->join to retrieve |
199 |
|
200 |
Cancellation *can* be dangerous - it's a bit like calling "exit" |
201 |
without actually exiting, and might leave C libraries and XS modules |
202 |
in a weird state. Unlike other thread implementations, however, Coro |
203 |
is exceptionally safe with regards to cancellation, as perl will |
204 |
always be in a consistent state, and for those cases where you want |
205 |
to do truly marvellous things with your coro while it is being |
206 |
cancelled - that is, make sure all cleanup code is executed from the |
207 |
thread being cancelled - there is even a "->safe_cancel" method. |
208 |
|
209 |
So, cancelling a thread that runs in an XS event loop might not be |
210 |
the best idea, but any other combination that deals with perl only |
211 |
(cancelling when a thread is in a "tie" method or an "AUTOLOAD" for |
212 |
example) is safe. |
213 |
|
214 |
Last not least, a coro thread object that isn't referenced is |
215 |
"->cancel"'ed automatically - just like other objects in Perl. This |
216 |
is not such a common case, however - a running thread is referencedy |
217 |
by $Coro::current, a thread ready to run is referenced by the ready |
218 |
queue, a thread waiting on a lock or semaphore is referenced by |
219 |
being in some wait list and so on. But a thread that isn't in any of |
220 |
those queues gets cancelled: |
221 |
|
222 |
async { |
223 |
schedule; # cede to other coros, don't go into the ready queue |
224 |
}; |
225 |
|
226 |
cede; |
227 |
# now the async above is destroyed, as it is not referenced by anything. |
228 |
|
229 |
A slightly embellished example might make it clearer: |
230 |
|
231 |
async { |
232 |
my $guard = Guard::guard { print "destroyed\n" }; |
233 |
schedule while 1; |
234 |
}; |
235 |
|
236 |
cede; |
237 |
|
238 |
Superficially one might not expect any output - since the "async" |
239 |
implements an endless loop, the $guard will not be cleaned up. |
240 |
However, since the thread object returned by "async" is not stored |
241 |
anywhere, the thread is initially referenced because it is in the |
242 |
ready queue, when it runs it is referenced by $Coro::current, but |
243 |
when it calls "schedule", it gets "cancel"ed causing the guard |
244 |
object to be destroyed (see the next section), and printing it's |
245 |
message. |
246 |
|
247 |
If this seems a bit drastic, remember that this only happens when |
248 |
nothing references the thread anymore, which means there is no way |
249 |
to further execute it, ever. The only options at this point are |
250 |
leaking the thread, or cleaning it up, which brings us to... |
251 |
|
252 |
5. Cleanup |
253 |
Threads will allocate various resources. Most but not all will be |
254 |
returned when a thread terminates, during clean-up. |
255 |
|
256 |
Cleanup is quite similar to throwing an uncaught exception: perl |
257 |
will work it's way up through all subroutine calls and blocks. On |
258 |
it's way, it will release all "my" variables, undo all "local"'s and |
259 |
free any other resources truly local to the thread. |
260 |
|
261 |
So, a common way to free resources is to keep them referenced only |
262 |
by my variables: |
263 |
|
264 |
async { |
265 |
my $big_cache = new Cache ...; |
266 |
}; |
267 |
|
268 |
If there are no other references, then the $big_cache object will be |
269 |
freed when the thread terminates, regardless of how it does so. |
270 |
|
271 |
What it does "NOT" do is unlock any Coro::Semaphores or similar |
272 |
resources, but that's where the "guard" methods come in handy: |
273 |
|
274 |
my $sem = new Coro::Semaphore; |
275 |
|
276 |
async { |
277 |
my $lock_guard = $sem->guard; |
278 |
# if we return, or die or get cancelled, here, |
279 |
# then the semaphore will be "up"ed. |
280 |
}; |
281 |
|
282 |
The "Guard::guard" function comes in handy for any custom cleanup |
283 |
you might want to do (but you cannot switch to other coroutines from |
284 |
those code blocks): |
285 |
|
286 |
async { |
287 |
my $window = new Gtk2::Window "toplevel"; |
288 |
# The window will not be cleaned up automatically, even when $window |
289 |
# gets freed, so use a guard to ensure it's destruction |
290 |
# in case of an error: |
291 |
my $window_guard = Guard::guard { $window->destroy }; |
292 |
|
293 |
# we are safe here |
294 |
}; |
295 |
|
296 |
Last not least, "local" can often be handy, too, e.g. when |
297 |
temporarily replacing the coro thread description: |
298 |
|
299 |
sub myfunction { |
300 |
local $Coro::current->{desc} = "inside myfunction(@_)"; |
301 |
|
302 |
# if we return or die here, the description will be restored |
303 |
} |
304 |
|
305 |
6. Viva La Zombie Muerte |
306 |
Even after a thread has terminated and cleaned up its resources, the |
307 |
Coro object still is there and stores the return values of the |
308 |
thread. |
309 |
|
310 |
When there are no other references, it will simply be cleaned up and |
311 |
freed. |
312 |
|
313 |
If there areany references, the Coro object will stay around, and |
314 |
you can call "->join" as many times as you wish to retrieve the |
315 |
result values: |
316 |
|
317 |
async { |
318 |
print "hi\n"; |
319 |
1 |
320 |
}; |
321 |
|
322 |
# run the async above, and free everything before returning |
323 |
# from Coro::cede: |
324 |
Coro::cede; |
325 |
|
326 |
{ |
327 |
my $coro = async { |
328 |
print "hi\n"; |
329 |
1 |
330 |
}; |
331 |
|
332 |
# run the async above, and clean up, but do not free the coro |
333 |
# object: |
334 |
Coro::cede; |
335 |
|
336 |
# optionally retrieve the result values |
337 |
my @results = $coro->join; |
338 |
|
339 |
# now $coro goes out of scope, and presumably gets freed |
340 |
}; |
341 |
|
342 |
GLOBAL VARIABLES |
343 |
$Coro::main |
344 |
This variable stores the Coro object that represents the main |
345 |
program. While you can "ready" it and do most other things you can |
346 |
do to coro, it is mainly useful to compare again $Coro::current, to |
347 |
see whether you are running in the main program or not. |
348 |
|
349 |
$Coro::current |
350 |
The Coro object representing the current coro (the last coro that |
351 |
the Coro scheduler switched to). The initial value is $Coro::main |
352 |
(of course). |
353 |
|
354 |
This variable is strictly *read-only*. You can take copies of the |
355 |
value stored in it and use it as any other Coro object, but you must |
356 |
not otherwise modify the variable itself. |
357 |
|
358 |
$Coro::idle |
359 |
This variable is mainly useful to integrate Coro into event loops. |
360 |
It is usually better to rely on Coro::AnyEvent or Coro::EV, as this |
361 |
is pretty low-level functionality. |
362 |
|
363 |
This variable stores a Coro object that is put into the ready queue |
364 |
when there are no other ready threads (without invoking any ready |
365 |
hooks). |
366 |
|
367 |
The default implementation dies with "FATAL: deadlock detected.", |
368 |
followed by a thread listing, because the program has no other way |
369 |
to continue. |
370 |
|
371 |
This hook is overwritten by modules such as "Coro::EV" and |
372 |
"Coro::AnyEvent" to wait on an external event that hopefully wakes |
373 |
up a coro so the scheduler can run it. |
374 |
|
375 |
See Coro::EV or Coro::AnyEvent for examples of using this technique. |
376 |
|
377 |
SIMPLE CORO CREATION |
378 |
async { ... } [@args...] |
379 |
Create a new coro and return its Coro object (usually unused). The |
380 |
coro will be put into the ready queue, so it will start running |
381 |
automatically on the next scheduler run. |
382 |
|
383 |
The first argument is a codeblock/closure that should be executed in |
384 |
the coro. When it returns argument returns the coro is automatically |
385 |
terminated. |
386 |
|
387 |
The remaining arguments are passed as arguments to the closure. |
388 |
|
389 |
See the "Coro::State::new" constructor for info about the coro |
390 |
environment in which coro are executed. |
391 |
|
392 |
Calling "exit" in a coro will do the same as calling exit outside |
393 |
the coro. Likewise, when the coro dies, the program will exit, just |
394 |
as it would in the main program. |
395 |
|
396 |
If you do not want that, you can provide a default "die" handler, or |
397 |
simply avoid dieing (by use of "eval"). |
398 |
|
399 |
Example: Create a new coro that just prints its arguments. |
400 |
|
401 |
async { |
402 |
print "@_\n"; |
403 |
} 1,2,3,4; |
404 |
|
405 |
async_pool { ... } [@args...] |
406 |
Similar to "async", but uses a coro pool, so you should not call |
407 |
terminate or join on it (although you are allowed to), and you get a |
408 |
coro that might have executed other code already (which can be good |
409 |
or bad :). |
410 |
|
411 |
On the plus side, this function is about twice as fast as creating |
412 |
(and destroying) a completely new coro, so if you need a lot of |
413 |
generic coros in quick successsion, use "async_pool", not "async". |
414 |
|
415 |
The code block is executed in an "eval" context and a warning will |
416 |
be issued in case of an exception instead of terminating the |
417 |
program, as "async" does. As the coro is being reused, stuff like |
418 |
"on_destroy" will not work in the expected way, unless you call |
419 |
terminate or cancel, which somehow defeats the purpose of pooling |
420 |
(but is fine in the exceptional case). |
421 |
|
422 |
The priority will be reset to 0 after each run, tracing will be |
423 |
disabled, the description will be reset and the default output |
424 |
filehandle gets restored, so you can change all these. Otherwise the |
425 |
coro will be re-used "as-is": most notably if you change other |
426 |
per-coro global stuff such as $/ you *must needs* revert that |
427 |
change, which is most simply done by using local as in: "local $/". |
428 |
|
429 |
The idle pool size is limited to 8 idle coros (this can be adjusted |
430 |
by changing $Coro::POOL_SIZE), but there can be as many non-idle |
431 |
coros as required. |
432 |
|
433 |
If you are concerned about pooled coros growing a lot because a |
434 |
single "async_pool" used a lot of stackspace you can e.g. |
435 |
"async_pool { terminate }" once per second or so to slowly replenish |
436 |
the pool. In addition to that, when the stacks used by a handler |
437 |
grows larger than 32kb (adjustable via $Coro::POOL_RSS) it will also |
438 |
be destroyed. |
439 |
|
440 |
STATIC METHODS |
441 |
Static methods are actually functions that implicitly operate on the |
442 |
current coro. |
443 |
|
444 |
schedule |
445 |
Calls the scheduler. The scheduler will find the next coro that is |
446 |
to be run from the ready queue and switches to it. The next coro to |
447 |
be run is simply the one with the highest priority that is longest |
448 |
in its ready queue. If there is no coro ready, it will call the |
449 |
$Coro::idle hook. |
450 |
|
451 |
Please note that the current coro will *not* be put into the ready |
452 |
queue, so calling this function usually means you will never be |
453 |
called again unless something else (e.g. an event handler) calls |
454 |
"->ready", thus waking you up. |
455 |
|
456 |
This makes "schedule" *the* generic method to use to block the |
457 |
current coro and wait for events: first you remember the current |
458 |
coro in a variable, then arrange for some callback of yours to call |
459 |
"->ready" on that once some event happens, and last you call |
460 |
"schedule" to put yourself to sleep. Note that a lot of things can |
461 |
wake your coro up, so you need to check whether the event indeed |
462 |
happened, e.g. by storing the status in a variable. |
463 |
|
464 |
See HOW TO WAIT FOR A CALLBACK, below, for some ways to wait for |
465 |
callbacks. |
466 |
|
467 |
cede |
468 |
"Cede" to other coros. This function puts the current coro into the |
469 |
ready queue and calls "schedule", which has the effect of giving up |
470 |
the current "timeslice" to other coros of the same or higher |
471 |
priority. Once your coro gets its turn again it will automatically |
472 |
be resumed. |
473 |
|
474 |
This function is often called "yield" in other languages. |
475 |
|
476 |
Coro::cede_notself |
477 |
Works like cede, but is not exported by default and will cede to |
478 |
*any* coro, regardless of priority. This is useful sometimes to |
479 |
ensure progress is made. |
480 |
|
481 |
terminate [arg...] |
482 |
Terminates the current coro with the given status values (see |
483 |
cancel). The values will not be copied, but referenced directly. |
484 |
|
485 |
Coro::on_enter BLOCK, Coro::on_leave BLOCK |
486 |
These function install enter and leave winders in the current scope. |
487 |
The enter block will be executed when on_enter is called and |
488 |
whenever the current coro is re-entered by the scheduler, while the |
489 |
leave block is executed whenever the current coro is blocked by the |
490 |
scheduler, and also when the containing scope is exited (by whatever |
491 |
means, be it exit, die, last etc.). |
492 |
|
493 |
*Neither invoking the scheduler, nor exceptions, are allowed within |
494 |
those BLOCKs*. That means: do not even think about calling "die" |
495 |
without an eval, and do not even think of entering the scheduler in |
496 |
any way. |
497 |
|
498 |
Since both BLOCKs are tied to the current scope, they will |
499 |
automatically be removed when the current scope exits. |
500 |
|
501 |
These functions implement the same concept as "dynamic-wind" in |
502 |
scheme does, and are useful when you want to localise some resource |
503 |
to a specific coro. |
504 |
|
505 |
They slow down thread switching considerably for coros that use them |
506 |
(about 40% for a BLOCK with a single assignment, so thread switching |
507 |
is still reasonably fast if the handlers are fast). |
508 |
|
509 |
These functions are best understood by an example: The following |
510 |
function will change the current timezone to |
511 |
"Antarctica/South_Pole", which requires a call to "tzset", but by |
512 |
using "on_enter" and "on_leave", which remember/change the current |
513 |
timezone and restore the previous value, respectively, the timezone |
514 |
is only changed for the coro that installed those handlers. |
515 |
|
516 |
use POSIX qw(tzset); |
517 |
|
518 |
async { |
519 |
my $old_tz; # store outside TZ value here |
520 |
|
521 |
Coro::on_enter { |
522 |
$old_tz = $ENV{TZ}; # remember the old value |
523 |
|
524 |
$ENV{TZ} = "Antarctica/South_Pole"; |
525 |
tzset; # enable new value |
526 |
}; |
527 |
|
528 |
Coro::on_leave { |
529 |
$ENV{TZ} = $old_tz; |
530 |
tzset; # restore old value |
531 |
}; |
532 |
|
533 |
# at this place, the timezone is Antarctica/South_Pole, |
534 |
# without disturbing the TZ of any other coro. |
535 |
}; |
536 |
|
537 |
This can be used to localise about any resource (locale, uid, |
538 |
current working directory etc.) to a block, despite the existance of |
539 |
other coros. |
540 |
|
541 |
Another interesting example implements time-sliced multitasking |
542 |
using interval timers (this could obviously be optimised, but does |
543 |
the job): |
544 |
|
545 |
# "timeslice" the given block |
546 |
sub timeslice(&) { |
547 |
use Time::HiRes (); |
548 |
|
549 |
Coro::on_enter { |
550 |
# on entering the thread, we set an VTALRM handler to cede |
551 |
$SIG{VTALRM} = sub { cede }; |
552 |
# and then start the interval timer |
553 |
Time::HiRes::setitimer &Time::HiRes::ITIMER_VIRTUAL, 0.01, 0.01; |
554 |
}; |
555 |
Coro::on_leave { |
556 |
# on leaving the thread, we stop the interval timer again |
557 |
Time::HiRes::setitimer &Time::HiRes::ITIMER_VIRTUAL, 0, 0; |
558 |
}; |
559 |
|
560 |
&{+shift}; |
561 |
} |
562 |
|
563 |
# use like this: |
564 |
timeslice { |
565 |
# The following is an endless loop that would normally |
566 |
# monopolise the process. Since it runs in a timesliced |
567 |
# environment, it will regularly cede to other threads. |
568 |
while () { } |
569 |
}; |
570 |
|
571 |
killall |
572 |
Kills/terminates/cancels all coros except the currently running one. |
573 |
|
574 |
Note that while this will try to free some of the main interpreter |
575 |
resources if the calling coro isn't the main coro, but one cannot |
576 |
free all of them, so if a coro that is not the main coro calls this |
577 |
function, there will be some one-time resource leak. |
578 |
|
579 |
CORO OBJECT METHODS |
580 |
These are the methods you can call on coro objects (or to create them). |
581 |
|
582 |
new Coro \&sub [, @args...] |
583 |
Create a new coro and return it. When the sub returns, the coro |
584 |
automatically terminates as if "terminate" with the returned values |
585 |
were called. To make the coro run you must first put it into the |
586 |
ready queue by calling the ready method. |
587 |
|
588 |
See "async" and "Coro::State::new" for additional info about the |
589 |
coro environment. |
590 |
|
591 |
$success = $coro->ready |
592 |
Put the given coro into the end of its ready queue (there is one |
593 |
queue for each priority) and return true. If the coro is already in |
594 |
the ready queue, do nothing and return false. |
595 |
|
596 |
This ensures that the scheduler will resume this coro automatically |
597 |
once all the coro of higher priority and all coro of the same |
598 |
priority that were put into the ready queue earlier have been |
599 |
resumed. |
600 |
|
601 |
$coro->suspend |
602 |
Suspends the specified coro. A suspended coro works just like any |
603 |
other coro, except that the scheduler will not select a suspended |
604 |
coro for execution. |
605 |
|
606 |
Suspending a coro can be useful when you want to keep the coro from |
607 |
running, but you don't want to destroy it, or when you want to |
608 |
temporarily freeze a coro (e.g. for debugging) to resume it later. |
609 |
|
610 |
A scenario for the former would be to suspend all (other) coros |
611 |
after a fork and keep them alive, so their destructors aren't |
612 |
called, but new coros can be created. |
613 |
|
614 |
$coro->resume |
615 |
If the specified coro was suspended, it will be resumed. Note that |
616 |
when the coro was in the ready queue when it was suspended, it might |
617 |
have been unreadied by the scheduler, so an activation might have |
618 |
been lost. |
619 |
|
620 |
To avoid this, it is best to put a suspended coro into the ready |
621 |
queue unconditionally, as every synchronisation mechanism must |
622 |
protect itself against spurious wakeups, and the one in the Coro |
623 |
family certainly do that. |
624 |
|
625 |
$state->is_new |
626 |
Returns true iff this Coro object is "new", i.e. has never been run |
627 |
yet. Those states basically consist of only the code reference to |
628 |
call and the arguments, but consumes very little other resources. |
629 |
New states will automatically get assigned a perl interpreter when |
630 |
they are transfered to. |
631 |
|
632 |
$state->is_zombie |
633 |
Returns true iff the Coro object has been cancelled, i.e. it's |
634 |
resources freed because they were "cancel"'ed, "terminate"'d, |
635 |
"safe_cancel"'ed or simply went out of scope. |
636 |
|
637 |
The name "zombie" stems from UNIX culture, where a process that has |
638 |
exited and only stores and exit status and no other resources is |
639 |
called a "zombie". |
640 |
|
641 |
$is_ready = $coro->is_ready |
642 |
Returns true iff the Coro object is in the ready queue. Unless the |
643 |
Coro object gets destroyed, it will eventually be scheduled by the |
644 |
scheduler. |
645 |
|
646 |
$is_running = $coro->is_running |
647 |
Returns true iff the Coro object is currently running. Only one Coro |
648 |
object can ever be in the running state (but it currently is |
649 |
possible to have multiple running Coro::States). |
650 |
|
651 |
$is_suspended = $coro->is_suspended |
652 |
Returns true iff this Coro object has been suspended. Suspended |
653 |
Coros will not ever be scheduled. |
654 |
|
655 |
$coro->cancel (arg...) |
656 |
Terminates the given Coro thread and makes it return the given |
657 |
arguments as status (default: an empty list). Never returns if the |
658 |
Coro is the current Coro. |
659 |
|
660 |
This is a rather brutal way to free a coro, with some limitations - |
661 |
if the thread is inside a C callback that doesn't expect to be |
662 |
canceled, bad things can happen, or if the cancelled thread insists |
663 |
on running complicated cleanup handlers that rely on its thread |
664 |
context, things will not work. |
665 |
|
666 |
Any cleanup code being run (e.g. from "guard" blocks, destructors |
667 |
and so on) will be run without a thread context, and is not allowed |
668 |
to switch to other threads. A common mistake is to call "->cancel" |
669 |
from a destructor called by die'ing inside the thread to be |
670 |
cancelled for example. |
671 |
|
672 |
On the plus side, "->cancel" will always clean up the thread, no |
673 |
matter what. If your cleanup code is complex or you want to avoid |
674 |
cancelling a C-thread that doesn't know how to clean up itself, it |
675 |
can be better to "->throw" an exception, or use "->safe_cancel". |
676 |
|
677 |
The arguments to "->cancel" are not copied, but instead will be |
678 |
referenced directly (e.g. if you pass $var and after the call change |
679 |
that variable, then you might change the return values passed to |
680 |
e.g. "join", so don't do that). |
681 |
|
682 |
The resources of the Coro are usually freed (or destructed) before |
683 |
this call returns, but this can be delayed for an indefinite amount |
684 |
of time, as in some cases the manager thread has to run first to |
685 |
actually destruct the Coro object. |
686 |
|
687 |
$coro->safe_cancel ($arg...) |
688 |
Works mostly like "->cancel", but is inherently "safer", and |
689 |
consequently, can fail with an exception in cases the thread is not |
690 |
in a cancellable state. Essentially, "->safe_cancel" is a "->cancel" |
691 |
with extra checks before canceling. |
692 |
|
693 |
It works a bit like throwing an exception that cannot be caught - |
694 |
specifically, it will clean up the thread from within itself, so all |
695 |
cleanup handlers (e.g. "guard" blocks) are run with full thread |
696 |
context and can block if they wish. The downside is that there is no |
697 |
guarantee that the thread can be cancelled when you call this |
698 |
method, and therefore, it might fail. It is also considerably slower |
699 |
than "cancel" or "terminate". |
700 |
|
701 |
A thread is in a safe-cancellable state if it either hasn't been run |
702 |
yet, or it has no C context attached and is inside an SLF function. |
703 |
|
704 |
The latter two basically mean that the thread isn't currently inside |
705 |
a perl callback called from some C function (usually via some XS |
706 |
modules) and isn't currently executing inside some C function itself |
707 |
(via Coro's XS API). |
708 |
|
709 |
This call returns true when it could cancel the thread, or croaks |
710 |
with an error otherwise (i.e. it either returns true or doesn't |
711 |
return at all). |
712 |
|
713 |
Why the weird interface? Well, there are two common models on how |
714 |
and when to cancel things. In the first, you have the expectation |
715 |
that your coro thread can be cancelled when you want to cancel it - |
716 |
if the thread isn't cancellable, this would be a bug somewhere, so |
717 |
"->safe_cancel" croaks to notify of the bug. |
718 |
|
719 |
In the second model you sometimes want to ask nicely to cancel a |
720 |
thread, but if it's not a good time, well, then don't cancel. This |
721 |
can be done relatively easy like this: |
722 |
|
723 |
if (! eval { $coro->safe_cancel }) { |
724 |
warn "unable to cancel thread: $@"; |
725 |
} |
726 |
|
727 |
However, what you never should do is first try to cancel "safely" |
728 |
and if that fails, cancel the "hard" way with "->cancel". That makes |
729 |
no sense: either you rely on being able to execute cleanup code in |
730 |
your thread context, or you don't. If you do, then "->safe_cancel" |
731 |
is the only way, and if you don't, then "->cancel" is always faster |
732 |
and more direct. |
733 |
|
734 |
$coro->schedule_to |
735 |
Puts the current coro to sleep (like "Coro::schedule"), but instead |
736 |
of continuing with the next coro from the ready queue, always switch |
737 |
to the given coro object (regardless of priority etc.). The |
738 |
readyness state of that coro isn't changed. |
739 |
|
740 |
This is an advanced method for special cases - I'd love to hear |
741 |
about any uses for this one. |
742 |
|
743 |
$coro->cede_to |
744 |
Like "schedule_to", but puts the current coro into the ready queue. |
745 |
This has the effect of temporarily switching to the given coro, and |
746 |
continuing some time later. |
747 |
|
748 |
This is an advanced method for special cases - I'd love to hear |
749 |
about any uses for this one. |
750 |
|
751 |
$coro->throw ([$scalar]) |
752 |
If $throw is specified and defined, it will be thrown as an |
753 |
exception inside the coro at the next convenient point in time. |
754 |
Otherwise clears the exception object. |
755 |
|
756 |
Coro will check for the exception each time a schedule-like-function |
757 |
returns, i.e. after each "schedule", "cede", |
758 |
"Coro::Semaphore->down", "Coro::Handle->readable" and so on. Most of |
759 |
those functions (all that are part of Coro itself) detect this case |
760 |
and return early in case an exception is pending. |
761 |
|
762 |
The exception object will be thrown "as is" with the specified |
763 |
scalar in $@, i.e. if it is a string, no line number or newline will |
764 |
be appended (unlike with "die"). |
765 |
|
766 |
This can be used as a softer means than either "cancel" or |
767 |
"safe_cancel "to ask a coro to end itself, although there is no |
768 |
guarantee that the exception will lead to termination, and if the |
769 |
exception isn't caught it might well end the whole program. |
770 |
|
771 |
You might also think of "throw" as being the moral equivalent of |
772 |
"kill"ing a coro with a signal (in this case, a scalar). |
773 |
|
774 |
$coro->join |
775 |
Wait until the coro terminates and return any values given to the |
776 |
"terminate" or "cancel" functions. "join" can be called concurrently |
777 |
from multiple threads, and all will be resumed and given the status |
778 |
return once the $coro terminates. |
779 |
|
780 |
$coro->on_destroy (\&cb) |
781 |
Registers a callback that is called when this coro thread gets |
782 |
destroyed, that is, after it's resources have been freed but before |
783 |
it is joined. The callback gets passed the terminate/cancel |
784 |
arguments, if any, and *must not* die, under any circumstances. |
785 |
|
786 |
There can be any number of "on_destroy" callbacks per coro, and |
787 |
there is currently no way to remove a callback once added. |
788 |
|
789 |
$oldprio = $coro->prio ($newprio) |
790 |
Sets (or gets, if the argument is missing) the priority of the coro |
791 |
thread. Higher priority coro get run before lower priority coros. |
792 |
Priorities are small signed integers (currently -4 .. +3), that you |
793 |
can refer to using PRIO_xxx constants (use the import tag :prio to |
794 |
get then): |
795 |
|
796 |
PRIO_MAX > PRIO_HIGH > PRIO_NORMAL > PRIO_LOW > PRIO_IDLE > PRIO_MIN |
797 |
3 > 1 > 0 > -1 > -3 > -4 |
798 |
|
799 |
# set priority to HIGH |
800 |
current->prio (PRIO_HIGH); |
801 |
|
802 |
The idle coro thread ($Coro::idle) always has a lower priority than |
803 |
any existing coro. |
804 |
|
805 |
Changing the priority of the current coro will take effect |
806 |
immediately, but changing the priority of a coro in the ready queue |
807 |
(but not running) will only take effect after the next schedule (of |
808 |
that coro). This is a bug that will be fixed in some future version. |
809 |
|
810 |
$newprio = $coro->nice ($change) |
811 |
Similar to "prio", but subtract the given value from the priority |
812 |
(i.e. higher values mean lower priority, just as in UNIX's nice |
813 |
command). |
814 |
|
815 |
$olddesc = $coro->desc ($newdesc) |
816 |
Sets (or gets in case the argument is missing) the description for |
817 |
this coro thread. This is just a free-form string you can associate |
818 |
with a coro. |
819 |
|
820 |
This method simply sets the "$coro->{desc}" member to the given |
821 |
string. You can modify this member directly if you wish, and in |
822 |
fact, this is often preferred to indicate major processing states |
823 |
that can then be seen for example in a Coro::Debug session: |
824 |
|
825 |
sub my_long_function { |
826 |
local $Coro::current->{desc} = "now in my_long_function"; |
827 |
... |
828 |
$Coro::current->{desc} = "my_long_function: phase 1"; |
829 |
... |
830 |
$Coro::current->{desc} = "my_long_function: phase 2"; |
831 |
... |
832 |
} |
833 |
|
834 |
GLOBAL FUNCTIONS |
835 |
Coro::nready |
836 |
Returns the number of coro that are currently in the ready state, |
837 |
i.e. that can be switched to by calling "schedule" directory or |
838 |
indirectly. The value 0 means that the only runnable coro is the |
839 |
currently running one, so "cede" would have no effect, and |
840 |
"schedule" would cause a deadlock unless there is an idle handler |
841 |
that wakes up some coro. |
842 |
|
843 |
my $guard = Coro::guard { ... } |
844 |
This function still exists, but is deprecated. Please use the |
845 |
"Guard::guard" function instead. |
846 |
|
847 |
unblock_sub { ... } |
848 |
This utility function takes a BLOCK or code reference and "unblocks" |
849 |
it, returning a new coderef. Unblocking means that calling the new |
850 |
coderef will return immediately without blocking, returning nothing, |
851 |
while the original code ref will be called (with parameters) from |
852 |
within another coro. |
853 |
|
854 |
The reason this function exists is that many event libraries (such |
855 |
as the venerable Event module) are not thread-safe (a weaker form of |
856 |
reentrancy). This means you must not block within event callbacks, |
857 |
otherwise you might suffer from crashes or worse. The only event |
858 |
library currently known that is safe to use without "unblock_sub" is |
859 |
EV (but you might still run into deadlocks if all event loops are |
860 |
blocked). |
861 |
|
862 |
Coro will try to catch you when you block in the event loop ("FATAL: |
863 |
$Coro::idle blocked itself"), but this is just best effort and only |
864 |
works when you do not run your own event loop. |
865 |
|
866 |
This function allows your callbacks to block by executing them in |
867 |
another coro where it is safe to block. One example where blocking |
868 |
is handy is when you use the Coro::AIO functions to save results to |
869 |
disk, for example. |
870 |
|
871 |
In short: simply use "unblock_sub { ... }" instead of "sub { ... }" |
872 |
when creating event callbacks that want to block. |
873 |
|
874 |
If your handler does not plan to block (e.g. simply sends a message |
875 |
to another coro, or puts some other coro into the ready queue), |
876 |
there is no reason to use "unblock_sub". |
877 |
|
878 |
Note that you also need to use "unblock_sub" for any other callbacks |
879 |
that are indirectly executed by any C-based event loop. For example, |
880 |
when you use a module that uses AnyEvent (and you use |
881 |
Coro::AnyEvent) and it provides callbacks that are the result of |
882 |
some event callback, then you must not block either, or use |
883 |
"unblock_sub". |
884 |
|
885 |
$cb = rouse_cb |
886 |
Create and return a "rouse callback". That's a code reference that, |
887 |
when called, will remember a copy of its arguments and notify the |
888 |
owner coro of the callback. |
889 |
|
890 |
See the next function. |
891 |
|
892 |
@args = rouse_wait [$cb] |
893 |
Wait for the specified rouse callback (or the last one that was |
894 |
created in this coro). |
895 |
|
896 |
As soon as the callback is invoked (or when the callback was invoked |
897 |
before "rouse_wait"), it will return the arguments originally passed |
898 |
to the rouse callback. In scalar context, that means you get the |
899 |
*last* argument, just as if "rouse_wait" had a "return ($a1, $a2, |
900 |
$a3...)" statement at the end. |
901 |
|
902 |
See the section HOW TO WAIT FOR A CALLBACK for an actual usage |
903 |
example. |
904 |
|
905 |
HOW TO WAIT FOR A CALLBACK |
906 |
It is very common for a coro to wait for some callback to be called. |
907 |
This occurs naturally when you use coro in an otherwise event-based |
908 |
program, or when you use event-based libraries. |
909 |
|
910 |
These typically register a callback for some event, and call that |
911 |
callback when the event occured. In a coro, however, you typically want |
912 |
to just wait for the event, simplyifying things. |
913 |
|
914 |
For example "AnyEvent->child" registers a callback to be called when a |
915 |
specific child has exited: |
916 |
|
917 |
my $child_watcher = AnyEvent->child (pid => $pid, cb => sub { ... }); |
918 |
|
919 |
But from within a coro, you often just want to write this: |
920 |
|
921 |
my $status = wait_for_child $pid; |
922 |
|
923 |
Coro offers two functions specifically designed to make this easy, |
924 |
"rouse_cb" and "rouse_wait". |
925 |
|
926 |
The first function, "rouse_cb", generates and returns a callback that, |
927 |
when invoked, will save its arguments and notify the coro that created |
928 |
the callback. |
929 |
|
930 |
The second function, "rouse_wait", waits for the callback to be called |
931 |
(by calling "schedule" to go to sleep) and returns the arguments |
932 |
originally passed to the callback. |
933 |
|
934 |
Using these functions, it becomes easy to write the "wait_for_child" |
935 |
function mentioned above: |
936 |
|
937 |
sub wait_for_child($) { |
938 |
my ($pid) = @_; |
939 |
|
940 |
my $watcher = AnyEvent->child (pid => $pid, cb => rouse_cb); |
941 |
|
942 |
my ($rpid, $rstatus) = rouse_wait; |
943 |
$rstatus |
944 |
} |
945 |
|
946 |
In the case where "rouse_cb" and "rouse_wait" are not flexible enough, |
947 |
you can roll your own, using "schedule" and "ready": |
948 |
|
949 |
sub wait_for_child($) { |
950 |
my ($pid) = @_; |
951 |
|
952 |
# store the current coro in $current, |
953 |
# and provide result variables for the closure passed to ->child |
954 |
my $current = $Coro::current; |
955 |
my ($done, $rstatus); |
956 |
|
957 |
# pass a closure to ->child |
958 |
my $watcher = AnyEvent->child (pid => $pid, cb => sub { |
959 |
$rstatus = $_[1]; # remember rstatus |
960 |
$done = 1; # mark $rstatus as valid |
961 |
$current->ready; # wake up the waiting thread |
962 |
}); |
963 |
|
964 |
# wait until the closure has been called |
965 |
schedule while !$done; |
966 |
|
967 |
$rstatus |
968 |
} |
969 |
|
970 |
BUGS/LIMITATIONS |
971 |
fork with pthread backend |
972 |
When Coro is compiled using the pthread backend (which isn't |
973 |
recommended but required on many BSDs as their libcs are completely |
974 |
broken), then coro will not survive a fork. There is no known |
975 |
workaround except to fix your libc and use a saner backend. |
976 |
|
977 |
perl process emulation ("threads") |
978 |
This module is not perl-pseudo-thread-safe. You should only ever use |
979 |
this module from the first thread (this requirement might be removed |
980 |
in the future to allow per-thread schedulers, but Coro::State does |
981 |
not yet allow this). I recommend disabling thread support and using |
982 |
processes, as having the windows process emulation enabled under |
983 |
unix roughly halves perl performance, even when not used. |
984 |
|
985 |
Attempts to use threads created in another emulated process will |
986 |
crash ("cleanly", with a null pointer exception). |
987 |
|
988 |
coro switching is not signal safe |
989 |
You must not switch to another coro from within a signal handler |
990 |
(only relevant with %SIG - most event libraries provide safe |
991 |
signals), *unless* you are sure you are not interrupting a Coro |
992 |
function. |
993 |
|
994 |
That means you *MUST NOT* call any function that might "block" the |
995 |
current coro - "cede", "schedule" "Coro::Semaphore->down" or |
996 |
anything that calls those. Everything else, including calling |
997 |
"ready", works. |
998 |
|
999 |
WINDOWS PROCESS EMULATION |
1000 |
A great many people seem to be confused about ithreads (for example, |
1001 |
Chip Salzenberg called me unintelligent, incapable, stupid and gullible, |
1002 |
while in the same mail making rather confused statements about perl |
1003 |
ithreads (for example, that memory or files would be shared), showing |
1004 |
his lack of understanding of this area - if it is hard to understand for |
1005 |
Chip, it is probably not obvious to everybody). |
1006 |
|
1007 |
What follows is an ultra-condensed version of my talk about threads in |
1008 |
scripting languages given on the perl workshop 2009: |
1009 |
|
1010 |
The so-called "ithreads" were originally implemented for two reasons: |
1011 |
first, to (badly) emulate unix processes on native win32 perls, and |
1012 |
secondly, to replace the older, real thread model ("5.005-threads"). |
1013 |
|
1014 |
It does that by using threads instead of OS processes. The difference |
1015 |
between processes and threads is that threads share memory (and other |
1016 |
state, such as files) between threads within a single process, while |
1017 |
processes do not share anything (at least not semantically). That means |
1018 |
that modifications done by one thread are seen by others, while |
1019 |
modifications by one process are not seen by other processes. |
1020 |
|
1021 |
The "ithreads" work exactly like that: when creating a new ithreads |
1022 |
process, all state is copied (memory is copied physically, files and |
1023 |
code is copied logically). Afterwards, it isolates all modifications. On |
1024 |
UNIX, the same behaviour can be achieved by using operating system |
1025 |
processes, except that UNIX typically uses hardware built into the |
1026 |
system to do this efficiently, while the windows process emulation |
1027 |
emulates this hardware in software (rather efficiently, but of course it |
1028 |
is still much slower than dedicated hardware). |
1029 |
|
1030 |
As mentioned before, loading code, modifying code, modifying data |
1031 |
structures and so on is only visible in the ithreads process doing the |
1032 |
modification, not in other ithread processes within the same OS process. |
1033 |
|
1034 |
This is why "ithreads" do not implement threads for perl at all, only |
1035 |
processes. What makes it so bad is that on non-windows platforms, you |
1036 |
can actually take advantage of custom hardware for this purpose (as |
1037 |
evidenced by the forks module, which gives you the (i-) threads API, |
1038 |
just much faster). |
1039 |
|
1040 |
Sharing data is in the i-threads model is done by transfering data |
1041 |
structures between threads using copying semantics, which is very slow - |
1042 |
shared data simply does not exist. Benchmarks using i-threads which are |
1043 |
communication-intensive show extremely bad behaviour with i-threads (in |
1044 |
fact, so bad that Coro, which cannot take direct advantage of multiple |
1045 |
CPUs, is often orders of magnitude faster because it shares data using |
1046 |
real threads, refer to my talk for details). |
1047 |
|
1048 |
As summary, i-threads *use* threads to implement processes, while the |
1049 |
compatible forks module *uses* processes to emulate, uhm, processes. |
1050 |
I-threads slow down every perl program when enabled, and outside of |
1051 |
windows, serve no (or little) practical purpose, but disadvantages every |
1052 |
single-threaded Perl program. |
1053 |
|
1054 |
This is the reason that I try to avoid the name "ithreads", as it is |
1055 |
misleading as it implies that it implements some kind of thread model |
1056 |
for perl, and prefer the name "windows process emulation", which |
1057 |
describes the actual use and behaviour of it much better. |
1058 |
|
1059 |
SEE ALSO |
1060 |
Event-Loop integration: Coro::AnyEvent, Coro::EV, Coro::Event. |
1061 |
|
1062 |
Debugging: Coro::Debug. |
1063 |
|
1064 |
Support/Utility: Coro::Specific, Coro::Util. |
1065 |
|
1066 |
Locking and IPC: Coro::Signal, Coro::Channel, Coro::Semaphore, |
1067 |
Coro::SemaphoreSet, Coro::RWLock. |
1068 |
|
1069 |
I/O and Timers: Coro::Timer, Coro::Handle, Coro::Socket, Coro::AIO. |
1070 |
|
1071 |
Compatibility with other modules: Coro::LWP (but see also AnyEvent::HTTP |
1072 |
for a better-working alternative), Coro::BDB, Coro::Storable, |
1073 |
Coro::Select. |
1074 |
|
1075 |
XS API: Coro::MakeMaker. |
1076 |
|
1077 |
Low level Configuration, Thread Environment, Continuations: Coro::State. |
1078 |
|
1079 |
AUTHOR |
1080 |
Marc Lehmann <schmorp@schmorp.de> |
1081 |
http://home.schmorp.de/ |
1082 |
|