… | |
… | |
6 | |
6 | |
7 | use AnyEvent; |
7 | use AnyEvent; |
8 | use AnyEvent::Fork::Pool; |
8 | use AnyEvent::Fork::Pool; |
9 | # use AnyEvent::Fork is not needed |
9 | # use AnyEvent::Fork is not needed |
10 | |
10 | |
11 | # all parameters with default values |
11 | # all possible parameters shown, with default values |
12 | my $pool = AnyEvent::Fork |
12 | my $pool = AnyEvent::Fork |
13 | ->new |
13 | ->new |
14 | ->require ("MyWorker") |
14 | ->require ("MyWorker") |
15 | ->AnyEvent::Fork::Pool::run ( |
15 | ->AnyEvent::Fork::Pool::run ( |
16 | "MyWorker::run", # the worker function |
16 | "MyWorker::run", # the worker function |
17 | |
17 | |
18 | # pool management |
18 | # pool management |
19 | max => 4, # absolute maximum # of processes |
19 | max => 4, # absolute maximum # of processes |
20 | idle => 2, # minimum # of idle processes |
20 | idle => 0, # minimum # of idle processes |
21 | load => 2, # queue at most this number of jobs per process |
21 | load => 2, # queue at most this number of jobs per process |
22 | start => 0.1, # wait this many seconds before starting a new process |
22 | start => 0.1, # wait this many seconds before starting a new process |
23 | stop => 1, # wait this many seconds before stopping an idle process |
23 | stop => 10, # wait this many seconds before stopping an idle process |
24 | on_destroy => (my $finish = AE::cv), # called when object is destroyed |
24 | on_destroy => (my $finish = AE::cv), # called when object is destroyed |
25 | |
25 | |
26 | # parameters passed to AnyEvent::Fork::RPC |
26 | # parameters passed to AnyEvent::Fork::RPC |
27 | async => 0, |
27 | async => 0, |
28 | on_error => sub { die "FATAL: $_[0]\n" }, |
28 | on_error => sub { die "FATAL: $_[0]\n" }, |
… | |
… | |
48 | pool of processes that handles jobs. |
48 | pool of processes that handles jobs. |
49 | |
49 | |
50 | Understanding of L<AnyEvent::Fork> is helpful but not critical to be able |
50 | Understanding of L<AnyEvent::Fork> is helpful but not critical to be able |
51 | to use this module, but a thorough understanding of L<AnyEvent::Fork::RPC> |
51 | to use this module, but a thorough understanding of L<AnyEvent::Fork::RPC> |
52 | is, as it defines the actual API that needs to be implemented in the |
52 | is, as it defines the actual API that needs to be implemented in the |
53 | children. |
53 | worker processes. |
54 | |
54 | |
55 | =head1 EXAMPLES |
55 | =head1 EXAMPLES |
56 | |
56 | |
57 | =head1 PARENT USAGE |
57 | =head1 PARENT USAGE |
|
|
58 | |
|
|
59 | To create a pool, you first have to create a L<AnyEvent::Fork> object - |
|
|
60 | this object becomes your template process. Whenever a new worker process |
|
|
61 | is needed, it is forked from this template process. Then you need to |
|
|
62 | "hand off" this template process to the C<AnyEvent::Fork::Pool> module by |
|
|
63 | calling its run method on it: |
|
|
64 | |
|
|
65 | my $template = AnyEvent::Fork |
|
|
66 | ->new |
|
|
67 | ->require ("SomeModule", "MyWorkerModule"); |
|
|
68 | |
|
|
69 | my $pool = $template->AnyEvent::Fork::Pool::run ("MyWorkerModule::myfunction"); |
|
|
70 | |
|
|
71 | The pool "object" is not a regular Perl object, but a code reference that |
|
|
72 | you can call and that works roughly like calling the worker function |
|
|
73 | directly, except that it returns nothing but instead you need to specify a |
|
|
74 | callback to be invoked once results are in: |
|
|
75 | |
|
|
76 | $pool->(1, 2, 3, sub { warn "myfunction(1,2,3) returned @_" }); |
58 | |
77 | |
59 | =over 4 |
78 | =over 4 |
60 | |
79 | |
61 | =cut |
80 | =cut |
62 | |
81 | |
… | |
… | |
81 | |
100 | |
82 | our $VERSION = 0.1; |
101 | our $VERSION = 0.1; |
83 | |
102 | |
84 | =item my $pool = AnyEvent::Fork::Pool::run $fork, $function, [key => value...] |
103 | =item my $pool = AnyEvent::Fork::Pool::run $fork, $function, [key => value...] |
85 | |
104 | |
86 | The traditional way to call it. But it is way cooler to call it in the |
105 | The traditional way to call the pool creation function. But it is way |
87 | following way: |
106 | cooler to call it in the following way: |
88 | |
107 | |
89 | =item my $pool = $fork->AnyEvent::Fork::Pool::run ($function, [key => value...]) |
108 | =item my $pool = $fork->AnyEvent::Fork::Pool::run ($function, [key => value...]) |
90 | |
109 | |
91 | Creates a new pool object with the specified C<$function> as function |
110 | Creates a new pool object with the specified C<$function> as function |
92 | (name) to call for each request. The pool uses the C<$fork> object as the |
111 | (name) to call for each request. The pool uses the C<$fork> object as the |
… | |
… | |
105 | |
124 | |
106 | The pool consists of a certain number of worker processes. These options |
125 | The pool consists of a certain number of worker processes. These options |
107 | decide how many of these processes exist and when they are started and |
126 | decide how many of these processes exist and when they are started and |
108 | stopped. |
127 | stopped. |
109 | |
128 | |
|
|
129 | The worker pool is dynamically resized, according to (perceived :) |
|
|
130 | load. The minimum size is given by the C<idle> parameter and the maximum |
|
|
131 | size is given by the C<max> parameter. A new worker is started every |
|
|
132 | C<start> seconds at most, and an idle worker is stopped at most every |
|
|
133 | C<stop> second. |
|
|
134 | |
|
|
135 | You can specify the amount of jobs sent to a worker concurrently using the |
|
|
136 | C<load> parameter. |
|
|
137 | |
110 | =over 4 |
138 | =over 4 |
111 | |
139 | |
112 | =item idle => $count (default: 0) |
140 | =item idle => $count (default: 0) |
113 | |
141 | |
114 | The minimum amount of idle processes in the pool - when there are fewer |
142 | The minimum amount of idle processes in the pool - when there are fewer |
115 | than this many idle workers, C<AnyEvent::Fork::Pool> will try to start new |
143 | than this many idle workers, C<AnyEvent::Fork::Pool> will try to start new |
116 | ones, subject to C<max> and C<start>. |
144 | ones, subject to the limits set by C<max> and C<start>. |
117 | |
145 | |
118 | This is also the initial/minimum amount of workers in the pool. The |
146 | This is also the initial amount of workers in the pool. The default of |
119 | default of zero means that the pool starts empty and can shrink back to |
147 | zero means that the pool starts empty and can shrink back to zero workers |
120 | zero workers over time. |
148 | over time. |
121 | |
149 | |
122 | =item max => $count (default: 4) |
150 | =item max => $count (default: 4) |
123 | |
151 | |
124 | The maximum number of processes in the pool, in addition to the template |
152 | The maximum number of processes in the pool, in addition to the template |
125 | process. C<AnyEvent::Fork::Pool> will never create more than this number |
153 | process. C<AnyEvent::Fork::Pool> will never have more than this number of |
126 | of worker processes, although there can be more temporarily when a worker |
154 | worker processes, although there can be more temporarily when a worker is |
127 | is shut down and hasn't exited yet. |
155 | shut down and hasn't exited yet. |
128 | |
156 | |
129 | =item load => $count (default: 2) |
157 | =item load => $count (default: 2) |
130 | |
158 | |
131 | The maximum number of concurrent jobs sent to a single worker |
159 | The maximum number of concurrent jobs sent to a single worker process. |
132 | process. Worker processes that handle this number of jobs already are |
|
|
133 | called "busy". |
|
|
134 | |
160 | |
135 | Jobs that cannot be sent to a worker immediately (because all workers are |
161 | Jobs that cannot be sent to a worker immediately (because all workers are |
136 | busy) will be queued until a worker is available. |
162 | busy) will be queued until a worker is available. |
137 | |
163 | |
|
|
164 | Setting this low improves latency. For example, at C<1>, every job that |
|
|
165 | is sent to a worker is sent to a completely idle worker that doesn't run |
|
|
166 | any other jobs. The downside is that throughput is reduced - a worker that |
|
|
167 | finishes a job needs to wait for a new job from the parent. |
|
|
168 | |
|
|
169 | The default of C<2> is usually a good compromise. |
|
|
170 | |
138 | =item start => $seconds (default: 0.1) |
171 | =item start => $seconds (default: 0.1) |
139 | |
172 | |
140 | When a job is queued and all workers are busy, a timer is started. If the |
173 | When there are fewer than C<idle> workers (or all workers are completely |
141 | timer elapses and there are still jobs that cannot be queued to a worker, |
174 | busy), then a timer is started. If the timer elapses and there are still |
142 | a new worker is started. |
175 | jobs that cannot be queued to a worker, a new worker is started. |
143 | |
176 | |
144 | This configurs the time that all workers must be busy before a new worker |
177 | This sets the minimum time that all workers must be busy before a new |
145 | is started. Or, put differently, the minimum delay betwene starting new |
178 | worker is started. Or, put differently, the minimum delay between starting |
146 | workers. |
179 | new workers. |
147 | |
180 | |
148 | The delay is zero by default, which means new workers will be started |
181 | The delay is small by default, which means new workers will be started |
149 | without delay. |
182 | relatively quickly. A delay of C<0> is possible, and ensures that the pool |
|
|
183 | will grow as quickly as possible under load. |
150 | |
184 | |
|
|
185 | Non-zero values are useful to avoid "exploding" a pool because a lot of |
|
|
186 | jobs are queued in an instant. |
|
|
187 | |
|
|
188 | Higher values are often useful to improve efficiency at the cost of |
|
|
189 | latency - when fewer processes can do the job over time, starting more and |
|
|
190 | more is not necessarily going to help. |
|
|
191 | |
151 | =item stop => $seconds (default: 1) |
192 | =item stop => $seconds (default: 10) |
152 | |
193 | |
153 | When a worker has no jobs to execute it becomes idle. An idle worker that |
194 | When a worker has no jobs to execute it becomes idle. An idle worker that |
154 | hasn't executed a job within this amount of time will be stopped, unless |
195 | hasn't executed a job within this amount of time will be stopped, unless |
155 | the other parameters say otherwise. |
196 | the other parameters say otherwise. |
156 | |
197 | |
|
|
198 | Setting this to a very high value means that workers stay around longer, |
|
|
199 | even when they have nothing to do, which can be good as they don't have to |
|
|
200 | be started on the netx load spike again. |
|
|
201 | |
|
|
202 | Setting this to a lower value can be useful to avoid memory or simply |
|
|
203 | process table wastage. |
|
|
204 | |
|
|
205 | Usually, setting this to a time longer than the time between load spikes |
|
|
206 | is best - if you expect a lot of requests every minute and little work |
|
|
207 | in between, setting this to longer than a minute avoids having to stop |
|
|
208 | and start workers. On the other hand, you have to ask yourself if letting |
|
|
209 | workers run idle is a good use of your resources. Try to find a good |
|
|
210 | balance between resource usage of your workers and the time to start new |
|
|
211 | workers - the processes created by L<AnyEvent::Fork> itself is fats at |
|
|
212 | creating workers while not using much memory for them, so most of the |
|
|
213 | overhead is likely from your own code. |
|
|
214 | |
157 | =item on_destroy => $callback->() (default: none) |
215 | =item on_destroy => $callback->() (default: none) |
158 | |
216 | |
159 | When a pool object goes out of scope, it will still handle all outstanding |
217 | When a pool object goes out of scope, the outstanding requests are still |
160 | jobs. After that, it will destroy all workers (and also the template |
218 | handled till completion. Only after handling all jobs will the workers |
161 | process if it isn't referenced otherwise). |
219 | be destroyed (and also the template process if it isn't referenced |
|
|
220 | otherwise). |
|
|
221 | |
|
|
222 | To find out when a pool I<really> has finished its work, you can set this |
|
|
223 | callback, which will be called when the pool has been destroyed. |
162 | |
224 | |
163 | =back |
225 | =back |
164 | |
226 | |
165 | =item Template Process |
227 | =item AnyEvent::Fork::RPC Parameters |
166 | |
228 | |
167 | The worker processes are all forked from a single template |
229 | These parameters are all passed more or less directly to |
168 | process. Ideally, all modules and all cdoe used by the worker, as well as |
230 | L<AnyEvent::Fork::RPC>. They are only briefly mentioned here, for |
169 | any shared data structures should be loaded into the template process, to |
231 | their full documentation please refer to the L<AnyEvent::Fork::RPC> |
170 | take advantage of data sharing via fork. |
232 | documentation. Also, the default values mentioned here are only documented |
171 | |
233 | as a best effort - the L<AnyEvent::Fork::RPC> documentation is binding. |
172 | You can create your own template process by creating a L<AnyEvent::Fork> |
|
|
173 | object yourself and passing it as the C<template> parameter, but |
|
|
174 | C<AnyEvent::Fork::Pool> can create one for you, including some standard |
|
|
175 | options. |
|
|
176 | |
234 | |
177 | =over 4 |
235 | =over 4 |
178 | |
236 | |
179 | =item template => $fork (default: C<< AnyEvent::Fork->new >>) |
|
|
180 | |
|
|
181 | The template process to use, if you want to create your own. |
|
|
182 | |
|
|
183 | =item require => \@modules (default: C<[]>) |
|
|
184 | |
|
|
185 | The modules in this list will be laoded into the template process. |
|
|
186 | |
|
|
187 | =item eval => "# perl code to execute in template" (default: none) |
|
|
188 | |
|
|
189 | This is a perl string that is evaluated after creating the template |
|
|
190 | process and after requiring the modules. It can do whatever it wants to |
|
|
191 | configure the process, but it must not do anything that would keep a later |
|
|
192 | fork from working (so must not create event handlers or (real) threads for |
|
|
193 | example). |
|
|
194 | |
|
|
195 | =back |
|
|
196 | |
|
|
197 | =item AnyEvent::Fork::RPC Parameters |
|
|
198 | |
|
|
199 | These parameters are all passed directly to L<AnyEvent::Fork::RPC>. They |
|
|
200 | are only briefly mentioned here, for their full documentation |
|
|
201 | please refer to the L<AnyEvent::Fork::RPC> documentation. Also, the |
|
|
202 | default values mentioned here are only documented as a best effort - |
|
|
203 | L<AnyEvent::Fork::RPC> documentation is binding. |
|
|
204 | |
|
|
205 | =over 4 |
|
|
206 | |
|
|
207 | =item async => $boolean (default: 0) |
237 | =item async => $boolean (default: 0) |
208 | |
238 | |
209 | Whether to sue the synchronous or asynchronous RPC backend. |
239 | Whether to use the synchronous or asynchronous RPC backend. |
210 | |
240 | |
211 | =item on_error => $callback->($message) (default: die with message) |
241 | =item on_error => $callback->($message) (default: die with message) |
212 | |
242 | |
213 | The callback to call on any (fatal) errors. |
243 | The callback to call on any (fatal) errors. |
214 | |
244 | |
… | |
… | |
235 | |
265 | |
236 | my $max = $arg{max} || 4; |
266 | my $max = $arg{max} || 4; |
237 | my $idle = $arg{idle} || 0, |
267 | my $idle = $arg{idle} || 0, |
238 | my $load = $arg{load} || 2, |
268 | my $load = $arg{load} || 2, |
239 | my $start = $arg{start} || 0.1, |
269 | my $start = $arg{start} || 0.1, |
240 | my $stop = $arg{stop} || 1, |
270 | my $stop = $arg{stop} || 10, |
241 | my $on_event = $arg{on_event} || sub { }, |
271 | my $on_event = $arg{on_event} || sub { }, |
242 | my $on_destroy = $arg{on_destroy}; |
272 | my $on_destroy = $arg{on_destroy}; |
243 | |
273 | |
244 | my @rpc = ( |
274 | my @rpc = ( |
245 | async => $arg{async}, |
275 | async => $arg{async}, |
… | |
… | |
258 | |
288 | |
259 | $template |
289 | $template |
260 | ->require ("AnyEvent::Fork::RPC::" . ($arg{async} ? "Async" : "Sync")) |
290 | ->require ("AnyEvent::Fork::RPC::" . ($arg{async} ? "Async" : "Sync")) |
261 | ->eval (' |
291 | ->eval (' |
262 | my ($magic0, $magic1) = @_; |
292 | my ($magic0, $magic1) = @_; |
263 | sub AnyEvent::Fork::Pool::quit() { |
293 | sub AnyEvent::Fork::Pool::retire() { |
264 | AnyEvent::Fork::RPC::on_event $magic0, "quit", $magic1; |
294 | AnyEvent::Fork::RPC::event $magic0, "quit", $magic1; |
265 | } |
295 | } |
266 | ', $magic0, $magic1) |
296 | ', $magic0, $magic1) |
267 | ->eval ($arg{eval}); |
297 | ; |
268 | |
298 | |
269 | $start_worker = sub { |
299 | $start_worker = sub { |
270 | my $proc = [0, 0, undef]; # load, index, rpc |
300 | my $proc = [0, 0, undef]; # load, index, rpc |
271 | |
301 | |
272 | $proc->[2] = $template |
302 | $proc->[2] = $template |
… | |
… | |
397 | |
427 | |
398 | =item $pool->(..., $cb->(...)) |
428 | =item $pool->(..., $cb->(...)) |
399 | |
429 | |
400 | Call the RPC function of a worker with the given arguments, and when the |
430 | Call the RPC function of a worker with the given arguments, and when the |
401 | worker is done, call the C<$cb> with the results, just like calling the |
431 | worker is done, call the C<$cb> with the results, just like calling the |
402 | L<AnyEvent::Fork::RPC> object directly. |
432 | RPC object durectly - see the L<AnyEvent::Fork::RPC> documentation for |
|
|
433 | details on the RPC API. |
403 | |
434 | |
404 | If there is no free worker, the call will be queued. |
435 | If there is no free worker, the call will be queued until a worker becomes |
|
|
436 | available. |
405 | |
437 | |
406 | Note that there can be considerable time between calling this method and |
438 | Note that there can be considerable time between calling this method and |
407 | the call actually being executed. During this time, the parameters passed |
439 | the call actually being executed. During this time, the parameters passed |
408 | to this function are effectively read-only - modifying them after the call |
440 | to this function are effectively read-only - modifying them after the call |
409 | and before the callback is invoked causes undefined behaviour. |
441 | and before the callback is invoked causes undefined behaviour. |
410 | |
442 | |
411 | =cut |
443 | =cut |
412 | |
444 | |
413 | =back |
445 | =back |
414 | |
446 | |
|
|
447 | =head1 CHILD USAGE |
|
|
448 | |
|
|
449 | In addition to the L<AnyEvent::Fork::RPC> API, this module implements one |
|
|
450 | more child-side function: |
|
|
451 | |
|
|
452 | =over 4 |
|
|
453 | |
|
|
454 | =item AnyEvent::Fork::Pool::retire () |
|
|
455 | |
|
|
456 | This function sends an event to the parent process to request retirement: |
|
|
457 | the worker is removed from the pool and no new jobs will be sent to it, |
|
|
458 | but it has to handle the jobs that are already queued. |
|
|
459 | |
|
|
460 | The parentheses are part of the syntax: the function usually isn't defined |
|
|
461 | when you compile your code (because that happens I<before> handing the |
|
|
462 | template process over to C<AnyEvent::Fork::Pool::run>, so you need the |
|
|
463 | empty parentheses to tell Perl that the function is indeed a function. |
|
|
464 | |
|
|
465 | Retiring a worker can be useful to gracefully shut it down when the worker |
|
|
466 | deems this useful. For example, after executing a job, one could check |
|
|
467 | the process size or the number of jobs handled so far, and if either is |
|
|
468 | too high, the worker could ask to get retired, to avoid memory leaks to |
|
|
469 | accumulate. |
|
|
470 | |
|
|
471 | =back |
|
|
472 | |
|
|
473 | =head1 POOL PARAMETERS RECIPES |
|
|
474 | |
|
|
475 | This section describes some recipes for pool paramaters. These are mostly |
|
|
476 | meant for the synchronous RPC backend, as the asynchronous RPC backend |
|
|
477 | changes the rules considerably, making workers themselves responsible for |
|
|
478 | their scheduling. |
|
|
479 | |
|
|
480 | =over 4 |
|
|
481 | |
|
|
482 | =item low latency - set load = 1 |
|
|
483 | |
|
|
484 | If you need a deterministic low latency, you should set the C<load> |
|
|
485 | parameter to C<1>. This ensures that never more than one job is sent to |
|
|
486 | each worker. This avoids having to wait for a previous job to finish. |
|
|
487 | |
|
|
488 | This makes most sense with the synchronous (default) backend, as the |
|
|
489 | asynchronous backend can handle multiple requests concurrently. |
|
|
490 | |
|
|
491 | =item lowest latency - set load = 1 and idle = max |
|
|
492 | |
|
|
493 | To achieve the lowest latency, you additionally should disable any dynamic |
|
|
494 | resizing of the pool by setting C<idle> to the same value as C<max>. |
|
|
495 | |
|
|
496 | =item high throughput, cpu bound jobs - set load >= 2, max = #cpus |
|
|
497 | |
|
|
498 | To get high throughput with cpu-bound jobs, you should set the maximum |
|
|
499 | pool size to the number of cpus in your system, and C<load> to at least |
|
|
500 | C<2>, to make sure there can be another job waiting for the worker when it |
|
|
501 | has finished one. |
|
|
502 | |
|
|
503 | The value of C<2> for C<load> is the minimum value that I<can> achieve |
|
|
504 | 100% throughput, but if your parent process itself is sometimes busy, you |
|
|
505 | might need higher values. Also there is a limit on the amount of data that |
|
|
506 | can be "in flight" to the worker, so if you send big blobs of data to your |
|
|
507 | worker, C<load> might have much less of an effect. |
|
|
508 | |
|
|
509 | =item high throughput, I/O bound jobs - set load >= 2, max = 1, or very high |
|
|
510 | |
|
|
511 | When your jobs are I/O bound, using more workers usually boils down to |
|
|
512 | higher throughput, depending very much on your actual workload - sometimes |
|
|
513 | having only one worker is best, for example, when you read or write big |
|
|
514 | files at maixmum speed, as a second worker will increase seek times. |
|
|
515 | |
|
|
516 | =back |
|
|
517 | |
415 | =head1 SEE ALSO |
518 | =head1 SEE ALSO |
416 | |
519 | |
417 | L<AnyEvent::Fork>, to create the processes in the first place. |
520 | L<AnyEvent::Fork>, to create the processes in the first place. |
418 | |
521 | |
419 | L<AnyEvent::Fork::RPC>, which implements the RPC protocol and API. |
522 | L<AnyEvent::Fork::RPC>, which implements the RPC protocol and API. |