1 |
=head1 NAME |
2 |
|
3 |
AnyEvent::MP - multi-processing/message-passing framework |
4 |
|
5 |
=head1 SYNOPSIS |
6 |
|
7 |
use AnyEvent::MP; |
8 |
|
9 |
$NODE # contains this node's noderef |
10 |
NODE # returns this node's noderef |
11 |
NODE $port # returns the noderef of the port |
12 |
|
13 |
$SELF # receiving/own port id in rcv callbacks |
14 |
|
15 |
# initialise the node so it can send/receive messages |
16 |
initialise_node; # -OR- |
17 |
initialise_node "localhost:4040"; # -OR- |
18 |
initialise_node "slave/", "localhost:4040" |
19 |
|
20 |
# ports are message endpoints |
21 |
|
22 |
# sending messages |
23 |
snd $port, type => data...; |
24 |
snd $port, @msg; |
25 |
snd @msg_with_first_element_being_a_port; |
26 |
|
27 |
# creating/using ports, the simple way |
28 |
my $simple_port = port { my @msg = @_; 0 }; |
29 |
|
30 |
# creating/using ports, tagged message matching |
31 |
my $port = port; |
32 |
rcv $port, ping => sub { snd $_[0], "pong"; 0 }; |
33 |
rcv $port, pong => sub { warn "pong received\n"; 0 }; |
34 |
|
35 |
# create a port on another node |
36 |
my $port = spawn $node, $initfunc, @initdata; |
37 |
|
38 |
# monitoring |
39 |
mon $port, $cb->(@msg) # callback is invoked on death |
40 |
mon $port, $otherport # kill otherport on abnormal death |
41 |
mon $port, $otherport, @msg # send message on death |
42 |
|
43 |
=head1 CURRENT STATUS |
44 |
|
45 |
AnyEvent::MP - stable API, should work |
46 |
AnyEvent::MP::Intro - outdated |
47 |
AnyEvent::MP::Kernel - WIP |
48 |
AnyEvent::MP::Transport - mostly stable |
49 |
|
50 |
stay tuned. |
51 |
|
52 |
=head1 DESCRIPTION |
53 |
|
54 |
This module (-family) implements a simple message passing framework. |
55 |
|
56 |
Despite its simplicity, you can securely message other processes running |
57 |
on the same or other hosts. |
58 |
|
59 |
For an introduction to this module family, see the L<AnyEvent::MP::Intro> |
60 |
manual page. |
61 |
|
62 |
At the moment, this module family is severly broken and underdocumented, |
63 |
so do not use. This was uploaded mainly to reserve the CPAN namespace - |
64 |
stay tuned! |
65 |
|
66 |
=head1 CONCEPTS |
67 |
|
68 |
=over 4 |
69 |
|
70 |
=item port |
71 |
|
72 |
A port is something you can send messages to (with the C<snd> function). |
73 |
|
74 |
Ports allow you to register C<rcv> handlers that can match all or just |
75 |
some messages. Messages will not be queued. |
76 |
|
77 |
=item port id - C<noderef#portname> |
78 |
|
79 |
A port ID is the concatenation of a noderef, a hash-mark (C<#>) as |
80 |
separator, and a port name (a printable string of unspecified format). An |
81 |
exception is the the node port, whose ID is identical to its node |
82 |
reference. |
83 |
|
84 |
=item node |
85 |
|
86 |
A node is a single process containing at least one port - the node port, |
87 |
which provides nodes to manage each other remotely, and to create new |
88 |
ports. |
89 |
|
90 |
Nodes are either private (single-process only), slaves (connected to a |
91 |
master node only) or public nodes (connectable from unrelated nodes). |
92 |
|
93 |
=item noderef - C<host:port,host:port...>, C<id@noderef>, C<id> |
94 |
|
95 |
A node reference is a string that either simply identifies the node (for |
96 |
private and slave nodes), or contains a recipe on how to reach a given |
97 |
node (for public nodes). |
98 |
|
99 |
This recipe is simply a comma-separated list of C<address:port> pairs (for |
100 |
TCP/IP, other protocols might look different). |
101 |
|
102 |
Node references come in two flavours: resolved (containing only numerical |
103 |
addresses) or unresolved (where hostnames are used instead of addresses). |
104 |
|
105 |
Before using an unresolved node reference in a message you first have to |
106 |
resolve it. |
107 |
|
108 |
=back |
109 |
|
110 |
=head1 VARIABLES/FUNCTIONS |
111 |
|
112 |
=over 4 |
113 |
|
114 |
=cut |
115 |
|
116 |
package AnyEvent::MP; |
117 |
|
118 |
use AnyEvent::MP::Kernel; |
119 |
|
120 |
use common::sense; |
121 |
|
122 |
use Carp (); |
123 |
|
124 |
use AE (); |
125 |
|
126 |
use base "Exporter"; |
127 |
|
128 |
our $VERSION = $AnyEvent::MP::Kernel::VERSION; |
129 |
|
130 |
our @EXPORT = qw( |
131 |
NODE $NODE *SELF node_of _any_ |
132 |
resolve_node initialise_node |
133 |
snd rcv mon kil reg psub spawn |
134 |
port |
135 |
); |
136 |
|
137 |
our $SELF; |
138 |
|
139 |
sub _self_die() { |
140 |
my $msg = $@; |
141 |
$msg =~ s/\n+$// unless ref $msg; |
142 |
kil $SELF, die => $msg; |
143 |
} |
144 |
|
145 |
=item $thisnode = NODE / $NODE |
146 |
|
147 |
The C<NODE> function returns, and the C<$NODE> variable contains the |
148 |
noderef of the local node. The value is initialised by a call to |
149 |
C<initialise_node>. |
150 |
|
151 |
=item $noderef = node_of $port |
152 |
|
153 |
Extracts and returns the noderef from a port ID or a noderef. |
154 |
|
155 |
=item initialise_node $noderef, $seednode, $seednode... |
156 |
|
157 |
=item initialise_node "slave/", $master, $master... |
158 |
|
159 |
Before a node can talk to other nodes on the network it has to initialise |
160 |
itself - the minimum a node needs to know is it's own name, and optionally |
161 |
it should know the noderefs of some other nodes in the network. |
162 |
|
163 |
This function initialises a node - it must be called exactly once (or |
164 |
never) before calling other AnyEvent::MP functions. |
165 |
|
166 |
All arguments (optionally except for the first) are noderefs, which can be |
167 |
either resolved or unresolved. |
168 |
|
169 |
The first argument will be looked up in the configuration database first |
170 |
(if it is C<undef> then the current nodename will be used instead) to find |
171 |
the relevant configuration profile (see L<aemp>). If none is found then |
172 |
the default configuration is used. The configuration supplies additional |
173 |
seed/master nodes and can override the actual noderef. |
174 |
|
175 |
There are two types of networked nodes, public nodes and slave nodes: |
176 |
|
177 |
=over 4 |
178 |
|
179 |
=item public nodes |
180 |
|
181 |
For public nodes, C<$noderef> (supplied either directly to |
182 |
C<initialise_node> or indirectly via a profile or the nodename) must be a |
183 |
noderef (possibly unresolved, in which case it will be resolved). |
184 |
|
185 |
After resolving, the node will bind itself on all endpoints and try to |
186 |
connect to all additional C<$seednodes> that are specified. Seednodes are |
187 |
optional and can be used to quickly bootstrap the node into an existing |
188 |
network. |
189 |
|
190 |
=item slave nodes |
191 |
|
192 |
When the C<$noderef> (either as given or overriden by the config file) |
193 |
is the special string C<slave/>, then the node will become a slave |
194 |
node. Slave nodes cannot be contacted from outside and will route most of |
195 |
their traffic to the master node that they attach to. |
196 |
|
197 |
At least one additional noderef is required (either by specifying it |
198 |
directly or because it is part of the configuration profile): The node |
199 |
will try to connect to all of them and will become a slave attached to the |
200 |
first node it can successfully connect to. |
201 |
|
202 |
Note that slave nodes cannot change their name, and consequently, their |
203 |
master, so if the master goes down, the slave node will not function well |
204 |
anymore until it can re-establish conenciton to its master. This makes |
205 |
slave nodes unsuitable for long-term nodes or fault-tolerant networks. |
206 |
|
207 |
=back |
208 |
|
209 |
This function will block until all nodes have been resolved and, for slave |
210 |
nodes, until it has successfully established a connection to a master |
211 |
server. |
212 |
|
213 |
All the seednodes will also be specially marked to automatically retry |
214 |
connecting to them infinitely. |
215 |
|
216 |
Example: become a public node listening on the guessed noderef, or the one |
217 |
specified via C<aemp> for the current node. This should be the most common |
218 |
form of invocation for "daemon"-type nodes. |
219 |
|
220 |
initialise_node; |
221 |
|
222 |
Example: become a slave node to any of the the seednodes specified via |
223 |
C<aemp>. This form is often used for commandline clients. |
224 |
|
225 |
initialise_node "slave/"; |
226 |
|
227 |
Example: become a slave node to any of the specified master servers. This |
228 |
form is also often used for commandline clients. |
229 |
|
230 |
initialise_node "slave/", "master1", "192.168.13.17", "mp.example.net"; |
231 |
|
232 |
Example: become a public node, and try to contact some well-known master |
233 |
servers to become part of the network. |
234 |
|
235 |
initialise_node undef, "master1", "master2"; |
236 |
|
237 |
Example: become a public node listening on port C<4041>. |
238 |
|
239 |
initialise_node 4041; |
240 |
|
241 |
Example: become a public node, only visible on localhost port 4044. |
242 |
|
243 |
initialise_node "localhost:4044"; |
244 |
|
245 |
=item $cv = resolve_node $noderef |
246 |
|
247 |
Takes an unresolved node reference that may contain hostnames and |
248 |
abbreviated IDs, resolves all of them and returns a resolved node |
249 |
reference. |
250 |
|
251 |
In addition to C<address:port> pairs allowed in resolved noderefs, the |
252 |
following forms are supported: |
253 |
|
254 |
=over 4 |
255 |
|
256 |
=item the empty string |
257 |
|
258 |
An empty-string component gets resolved as if the default port (4040) was |
259 |
specified. |
260 |
|
261 |
=item naked port numbers (e.g. C<1234>) |
262 |
|
263 |
These are resolved by prepending the local nodename and a colon, to be |
264 |
further resolved. |
265 |
|
266 |
=item hostnames (e.g. C<localhost:1234>, C<localhost>) |
267 |
|
268 |
These are resolved by using AnyEvent::DNS to resolve them, optionally |
269 |
looking up SRV records for the C<aemp=4040> port, if no port was |
270 |
specified. |
271 |
|
272 |
=back |
273 |
|
274 |
=item $SELF |
275 |
|
276 |
Contains the current port id while executing C<rcv> callbacks or C<psub> |
277 |
blocks. |
278 |
|
279 |
=item SELF, %SELF, @SELF... |
280 |
|
281 |
Due to some quirks in how perl exports variables, it is impossible to |
282 |
just export C<$SELF>, all the symbols called C<SELF> are exported by this |
283 |
module, but only C<$SELF> is currently used. |
284 |
|
285 |
=item snd $port, type => @data |
286 |
|
287 |
=item snd $port, @msg |
288 |
|
289 |
Send the given message to the given port ID, which can identify either |
290 |
a local or a remote port, and must be a port ID. |
291 |
|
292 |
While the message can be about anything, it is highly recommended to use a |
293 |
string as first element (a port ID, or some word that indicates a request |
294 |
type etc.). |
295 |
|
296 |
The message data effectively becomes read-only after a call to this |
297 |
function: modifying any argument is not allowed and can cause many |
298 |
problems. |
299 |
|
300 |
The type of data you can transfer depends on the transport protocol: when |
301 |
JSON is used, then only strings, numbers and arrays and hashes consisting |
302 |
of those are allowed (no objects). When Storable is used, then anything |
303 |
that Storable can serialise and deserialise is allowed, and for the local |
304 |
node, anything can be passed. |
305 |
|
306 |
=item $local_port = port |
307 |
|
308 |
Create a new local port object and returns its port ID. Initially it has |
309 |
no callbacks set and will throw an error when it receives messages. |
310 |
|
311 |
=item $local_port = port { my @msg = @_ } |
312 |
|
313 |
Creates a new local port, and returns its ID. Semantically the same as |
314 |
creating a port and calling C<rcv $port, $callback> on it. |
315 |
|
316 |
The block will be called for every message received on the port, with the |
317 |
global variable C<$SELF> set to the port ID. Runtime errors will cause the |
318 |
port to be C<kil>ed. The message will be passed as-is, no extra argument |
319 |
(i.e. no port ID) will be passed to the callback. |
320 |
|
321 |
If you want to stop/destroy the port, simply C<kil> it: |
322 |
|
323 |
my $port = port { |
324 |
my @msg = @_; |
325 |
... |
326 |
kil $SELF; |
327 |
}; |
328 |
|
329 |
=cut |
330 |
|
331 |
sub rcv($@); |
332 |
|
333 |
sub _kilme { |
334 |
die "received message on port without callback"; |
335 |
} |
336 |
|
337 |
sub port(;&) { |
338 |
my $id = "$UNIQ." . $ID++; |
339 |
my $port = "$NODE#$id"; |
340 |
|
341 |
rcv $port, shift || \&_kilme; |
342 |
|
343 |
$port |
344 |
} |
345 |
|
346 |
=item rcv $local_port, $callback->(@msg) |
347 |
|
348 |
Replaces the default callback on the specified port. There is no way to |
349 |
remove the default callback: use C<sub { }> to disable it, or better |
350 |
C<kil> the port when it is no longer needed. |
351 |
|
352 |
The global C<$SELF> (exported by this module) contains C<$port> while |
353 |
executing the callback. Runtime errors during callback execution will |
354 |
result in the port being C<kil>ed. |
355 |
|
356 |
The default callback received all messages not matched by a more specific |
357 |
C<tag> match. |
358 |
|
359 |
=item rcv $local_port, tag => $callback->(@msg_without_tag), ... |
360 |
|
361 |
Register (or replace) callbacks to be called on messages starting with the |
362 |
given tag on the given port (and return the port), or unregister it (when |
363 |
C<$callback> is C<$undef> or missing). There can only be one callback |
364 |
registered for each tag. |
365 |
|
366 |
The original message will be passed to the callback, after the first |
367 |
element (the tag) has been removed. The callback will use the same |
368 |
environment as the default callback (see above). |
369 |
|
370 |
Example: create a port and bind receivers on it in one go. |
371 |
|
372 |
my $port = rcv port, |
373 |
msg1 => sub { ... }, |
374 |
msg2 => sub { ... }, |
375 |
; |
376 |
|
377 |
Example: create a port, bind receivers and send it in a message elsewhere |
378 |
in one go: |
379 |
|
380 |
snd $otherport, reply => |
381 |
rcv port, |
382 |
msg1 => sub { ... }, |
383 |
... |
384 |
; |
385 |
|
386 |
Example: temporarily register a rcv callback for a tag matching some port |
387 |
(e.g. for a rpc reply) and unregister it after a message was received. |
388 |
|
389 |
rcv $port, $otherport => sub { |
390 |
my @reply = @_; |
391 |
|
392 |
rcv $SELF, $otherport; |
393 |
}; |
394 |
|
395 |
=cut |
396 |
|
397 |
sub rcv($@) { |
398 |
my $port = shift; |
399 |
my ($noderef, $portid) = split /#/, $port, 2; |
400 |
|
401 |
($NODE{$noderef} || add_node $noderef) == $NODE{""} |
402 |
or Carp::croak "$port: rcv can only be called on local ports, caught"; |
403 |
|
404 |
while (@_) { |
405 |
if (ref $_[0]) { |
406 |
if (my $self = $PORT_DATA{$portid}) { |
407 |
"AnyEvent::MP::Port" eq ref $self |
408 |
or Carp::croak "$port: rcv can only be called on message matching ports, caught"; |
409 |
|
410 |
$self->[2] = shift; |
411 |
} else { |
412 |
my $cb = shift; |
413 |
$PORT{$portid} = sub { |
414 |
local $SELF = $port; |
415 |
eval { &$cb }; _self_die if $@; |
416 |
}; |
417 |
} |
418 |
} elsif (defined $_[0]) { |
419 |
my $self = $PORT_DATA{$portid} ||= do { |
420 |
my $self = bless [$PORT{$port} || sub { }, { }, $port], "AnyEvent::MP::Port"; |
421 |
|
422 |
$PORT{$portid} = sub { |
423 |
local $SELF = $port; |
424 |
|
425 |
if (my $cb = $self->[1]{$_[0]}) { |
426 |
shift; |
427 |
eval { &$cb }; _self_die if $@; |
428 |
} else { |
429 |
&{ $self->[0] }; |
430 |
} |
431 |
}; |
432 |
|
433 |
$self |
434 |
}; |
435 |
|
436 |
"AnyEvent::MP::Port" eq ref $self |
437 |
or Carp::croak "$port: rcv can only be called on message matching ports, caught"; |
438 |
|
439 |
my ($tag, $cb) = splice @_, 0, 2; |
440 |
|
441 |
if (defined $cb) { |
442 |
$self->[1]{$tag} = $cb; |
443 |
} else { |
444 |
delete $self->[1]{$tag}; |
445 |
} |
446 |
} |
447 |
} |
448 |
|
449 |
$port |
450 |
} |
451 |
|
452 |
=item $closure = psub { BLOCK } |
453 |
|
454 |
Remembers C<$SELF> and creates a closure out of the BLOCK. When the |
455 |
closure is executed, sets up the environment in the same way as in C<rcv> |
456 |
callbacks, i.e. runtime errors will cause the port to get C<kil>ed. |
457 |
|
458 |
This is useful when you register callbacks from C<rcv> callbacks: |
459 |
|
460 |
rcv delayed_reply => sub { |
461 |
my ($delay, @reply) = @_; |
462 |
my $timer = AE::timer $delay, 0, psub { |
463 |
snd @reply, $SELF; |
464 |
}; |
465 |
}; |
466 |
|
467 |
=cut |
468 |
|
469 |
sub psub(&) { |
470 |
my $cb = shift; |
471 |
|
472 |
my $port = $SELF |
473 |
or Carp::croak "psub can only be called from within rcv or psub callbacks, not"; |
474 |
|
475 |
sub { |
476 |
local $SELF = $port; |
477 |
|
478 |
if (wantarray) { |
479 |
my @res = eval { &$cb }; |
480 |
_self_die if $@; |
481 |
@res |
482 |
} else { |
483 |
my $res = eval { &$cb }; |
484 |
_self_die if $@; |
485 |
$res |
486 |
} |
487 |
} |
488 |
} |
489 |
|
490 |
=item $guard = mon $port, $cb->(@reason) |
491 |
|
492 |
=item $guard = mon $port, $rcvport |
493 |
|
494 |
=item $guard = mon $port |
495 |
|
496 |
=item $guard = mon $port, $rcvport, @msg |
497 |
|
498 |
Monitor the given port and do something when the port is killed or |
499 |
messages to it were lost, and optionally return a guard that can be used |
500 |
to stop monitoring again. |
501 |
|
502 |
C<mon> effectively guarantees that, in the absence of hardware failures, |
503 |
that after starting the monitor, either all messages sent to the port |
504 |
will arrive, or the monitoring action will be invoked after possible |
505 |
message loss has been detected. No messages will be lost "in between" |
506 |
(after the first lost message no further messages will be received by the |
507 |
port). After the monitoring action was invoked, further messages might get |
508 |
delivered again. |
509 |
|
510 |
In the first form (callback), the callback is simply called with any |
511 |
number of C<@reason> elements (no @reason means that the port was deleted |
512 |
"normally"). Note also that I<< the callback B<must> never die >>, so use |
513 |
C<eval> if unsure. |
514 |
|
515 |
In the second form (another port given), the other port (C<$rcvport>) |
516 |
will be C<kil>'ed with C<@reason>, iff a @reason was specified, i.e. on |
517 |
"normal" kils nothing happens, while under all other conditions, the other |
518 |
port is killed with the same reason. |
519 |
|
520 |
The third form (kill self) is the same as the second form, except that |
521 |
C<$rvport> defaults to C<$SELF>. |
522 |
|
523 |
In the last form (message), a message of the form C<@msg, @reason> will be |
524 |
C<snd>. |
525 |
|
526 |
As a rule of thumb, monitoring requests should always monitor a port from |
527 |
a local port (or callback). The reason is that kill messages might get |
528 |
lost, just like any other message. Another less obvious reason is that |
529 |
even monitoring requests can get lost (for exmaple, when the connection |
530 |
to the other node goes down permanently). When monitoring a port locally |
531 |
these problems do not exist. |
532 |
|
533 |
Example: call a given callback when C<$port> is killed. |
534 |
|
535 |
mon $port, sub { warn "port died because of <@_>\n" }; |
536 |
|
537 |
Example: kill ourselves when C<$port> is killed abnormally. |
538 |
|
539 |
mon $port; |
540 |
|
541 |
Example: send us a restart message when another C<$port> is killed. |
542 |
|
543 |
mon $port, $self => "restart"; |
544 |
|
545 |
=cut |
546 |
|
547 |
sub mon { |
548 |
my ($noderef, $port) = split /#/, shift, 2; |
549 |
|
550 |
my $node = $NODE{$noderef} || add_node $noderef; |
551 |
|
552 |
my $cb = @_ ? shift : $SELF || Carp::croak 'mon: called with one argument only, but $SELF not set,'; |
553 |
|
554 |
unless (ref $cb) { |
555 |
if (@_) { |
556 |
# send a kill info message |
557 |
my (@msg) = ($cb, @_); |
558 |
$cb = sub { snd @msg, @_ }; |
559 |
} else { |
560 |
# simply kill other port |
561 |
my $port = $cb; |
562 |
$cb = sub { kil $port, @_ if @_ }; |
563 |
} |
564 |
} |
565 |
|
566 |
$node->monitor ($port, $cb); |
567 |
|
568 |
defined wantarray |
569 |
and AnyEvent::Util::guard { $node->unmonitor ($port, $cb) } |
570 |
} |
571 |
|
572 |
=item $guard = mon_guard $port, $ref, $ref... |
573 |
|
574 |
Monitors the given C<$port> and keeps the passed references. When the port |
575 |
is killed, the references will be freed. |
576 |
|
577 |
Optionally returns a guard that will stop the monitoring. |
578 |
|
579 |
This function is useful when you create e.g. timers or other watchers and |
580 |
want to free them when the port gets killed: |
581 |
|
582 |
$port->rcv (start => sub { |
583 |
my $timer; $timer = mon_guard $port, AE::timer 1, 1, sub { |
584 |
undef $timer if 0.9 < rand; |
585 |
}); |
586 |
}); |
587 |
|
588 |
=cut |
589 |
|
590 |
sub mon_guard { |
591 |
my ($port, @refs) = @_; |
592 |
|
593 |
#TODO: mon-less form? |
594 |
|
595 |
mon $port, sub { 0 && @refs } |
596 |
} |
597 |
|
598 |
=item kil $port[, @reason] |
599 |
|
600 |
Kill the specified port with the given C<@reason>. |
601 |
|
602 |
If no C<@reason> is specified, then the port is killed "normally" (linked |
603 |
ports will not be kileld, or even notified). |
604 |
|
605 |
Otherwise, linked ports get killed with the same reason (second form of |
606 |
C<mon>, see below). |
607 |
|
608 |
Runtime errors while evaluating C<rcv> callbacks or inside C<psub> blocks |
609 |
will be reported as reason C<< die => $@ >>. |
610 |
|
611 |
Transport/communication errors are reported as C<< transport_error => |
612 |
$message >>. |
613 |
|
614 |
=cut |
615 |
|
616 |
=item $port = spawn $node, $initfunc[, @initdata] |
617 |
|
618 |
Creates a port on the node C<$node> (which can also be a port ID, in which |
619 |
case it's the node where that port resides). |
620 |
|
621 |
The port ID of the newly created port is return immediately, and it is |
622 |
permissible to immediately start sending messages or monitor the port. |
623 |
|
624 |
After the port has been created, the init function is |
625 |
called. This function must be a fully-qualified function name |
626 |
(e.g. C<MyApp::Chat::Server::init>). To specify a function in the main |
627 |
program, use C<::name>. |
628 |
|
629 |
If the function doesn't exist, then the node tries to C<require> |
630 |
the package, then the package above the package and so on (e.g. |
631 |
C<MyApp::Chat::Server>, C<MyApp::Chat>, C<MyApp>) until the function |
632 |
exists or it runs out of package names. |
633 |
|
634 |
The init function is then called with the newly-created port as context |
635 |
object (C<$SELF>) and the C<@initdata> values as arguments. |
636 |
|
637 |
A common idiom is to pass your own port, monitor the spawned port, and |
638 |
in the init function, monitor the original port. This two-way monitoring |
639 |
ensures that both ports get cleaned up when there is a problem. |
640 |
|
641 |
Example: spawn a chat server port on C<$othernode>. |
642 |
|
643 |
# this node, executed from within a port context: |
644 |
my $server = spawn $othernode, "MyApp::Chat::Server::connect", $SELF; |
645 |
mon $server; |
646 |
|
647 |
# init function on C<$othernode> |
648 |
sub connect { |
649 |
my ($srcport) = @_; |
650 |
|
651 |
mon $srcport; |
652 |
|
653 |
rcv $SELF, sub { |
654 |
... |
655 |
}; |
656 |
} |
657 |
|
658 |
=cut |
659 |
|
660 |
sub _spawn { |
661 |
my $port = shift; |
662 |
my $init = shift; |
663 |
|
664 |
local $SELF = "$NODE#$port"; |
665 |
eval { |
666 |
&{ load_func $init } |
667 |
}; |
668 |
_self_die if $@; |
669 |
} |
670 |
|
671 |
sub spawn(@) { |
672 |
my ($noderef, undef) = split /#/, shift, 2; |
673 |
|
674 |
my $id = "$RUNIQ." . $ID++; |
675 |
|
676 |
$_[0] =~ /::/ |
677 |
or Carp::croak "spawn init function must be a fully-qualified name, caught"; |
678 |
|
679 |
snd_to_func $noderef, "AnyEvent::MP::_spawn" => $id, @_; |
680 |
|
681 |
"$noderef#$id" |
682 |
} |
683 |
|
684 |
=back |
685 |
|
686 |
=head1 NODE MESSAGES |
687 |
|
688 |
Nodes understand the following messages sent to them. Many of them take |
689 |
arguments called C<@reply>, which will simply be used to compose a reply |
690 |
message - C<$reply[0]> is the port to reply to, C<$reply[1]> the type and |
691 |
the remaining arguments are simply the message data. |
692 |
|
693 |
While other messages exist, they are not public and subject to change. |
694 |
|
695 |
=over 4 |
696 |
|
697 |
=cut |
698 |
|
699 |
=item lookup => $name, @reply |
700 |
|
701 |
Replies with the port ID of the specified well-known port, or C<undef>. |
702 |
|
703 |
=item devnull => ... |
704 |
|
705 |
Generic data sink/CPU heat conversion. |
706 |
|
707 |
=item relay => $port, @msg |
708 |
|
709 |
Simply forwards the message to the given port. |
710 |
|
711 |
=item eval => $string[ @reply] |
712 |
|
713 |
Evaluates the given string. If C<@reply> is given, then a message of the |
714 |
form C<@reply, $@, @evalres> is sent. |
715 |
|
716 |
Example: crash another node. |
717 |
|
718 |
snd $othernode, eval => "exit"; |
719 |
|
720 |
=item time => @reply |
721 |
|
722 |
Replies the the current node time to C<@reply>. |
723 |
|
724 |
Example: tell the current node to send the current time to C<$myport> in a |
725 |
C<timereply> message. |
726 |
|
727 |
snd $NODE, time => $myport, timereply => 1, 2; |
728 |
# => snd $myport, timereply => 1, 2, <time> |
729 |
|
730 |
=back |
731 |
|
732 |
=head1 AnyEvent::MP vs. Distributed Erlang |
733 |
|
734 |
AnyEvent::MP got lots of its ideas from distributed Erlang (Erlang node |
735 |
== aemp node, Erlang process == aemp port), so many of the documents and |
736 |
programming techniques employed by Erlang apply to AnyEvent::MP. Here is a |
737 |
sample: |
738 |
|
739 |
http://www.Erlang.se/doc/programming_rules.shtml |
740 |
http://Erlang.org/doc/getting_started/part_frame.html # chapters 3 and 4 |
741 |
http://Erlang.org/download/Erlang-book-part1.pdf # chapters 5 and 6 |
742 |
http://Erlang.org/download/armstrong_thesis_2003.pdf # chapters 4 and 5 |
743 |
|
744 |
Despite the similarities, there are also some important differences: |
745 |
|
746 |
=over 4 |
747 |
|
748 |
=item * Node references contain the recipe on how to contact them. |
749 |
|
750 |
Erlang relies on special naming and DNS to work everywhere in the |
751 |
same way. AEMP relies on each node knowing it's own address(es), with |
752 |
convenience functionality. |
753 |
|
754 |
This means that AEMP requires a less tightly controlled environment at the |
755 |
cost of longer node references and a slightly higher management overhead. |
756 |
|
757 |
=item * Erlang has a "remote ports are like local ports" philosophy, AEMP |
758 |
uses "local ports are like remote ports". |
759 |
|
760 |
The failure modes for local ports are quite different (runtime errors |
761 |
only) then for remote ports - when a local port dies, you I<know> it dies, |
762 |
when a connection to another node dies, you know nothing about the other |
763 |
port. |
764 |
|
765 |
Erlang pretends remote ports are as reliable as local ports, even when |
766 |
they are not. |
767 |
|
768 |
AEMP encourages a "treat remote ports differently" philosophy, with local |
769 |
ports being the special case/exception, where transport errors cannot |
770 |
occur. |
771 |
|
772 |
=item * Erlang uses processes and a mailbox, AEMP does not queue. |
773 |
|
774 |
Erlang uses processes that selectively receive messages, and therefore |
775 |
needs a queue. AEMP is event based, queuing messages would serve no |
776 |
useful purpose. For the same reason the pattern-matching abilities of |
777 |
AnyEvent::MP are more limited, as there is little need to be able to |
778 |
filter messages without dequeing them. |
779 |
|
780 |
(But see L<Coro::MP> for a more Erlang-like process model on top of AEMP). |
781 |
|
782 |
=item * Erlang sends are synchronous, AEMP sends are asynchronous. |
783 |
|
784 |
Sending messages in Erlang is synchronous and blocks the process (and |
785 |
so does not need a queue that can overflow). AEMP sends are immediate, |
786 |
connection establishment is handled in the background. |
787 |
|
788 |
=item * Erlang suffers from silent message loss, AEMP does not. |
789 |
|
790 |
Erlang makes few guarantees on messages delivery - messages can get lost |
791 |
without any of the processes realising it (i.e. you send messages a, b, |
792 |
and c, and the other side only receives messages a and c). |
793 |
|
794 |
AEMP guarantees correct ordering, and the guarantee that there are no |
795 |
holes in the message sequence. |
796 |
|
797 |
=item * In Erlang, processes can be declared dead and later be found to be |
798 |
alive. |
799 |
|
800 |
In Erlang it can happen that a monitored process is declared dead and |
801 |
linked processes get killed, but later it turns out that the process is |
802 |
still alive - and can receive messages. |
803 |
|
804 |
In AEMP, when port monitoring detects a port as dead, then that port will |
805 |
eventually be killed - it cannot happen that a node detects a port as dead |
806 |
and then later sends messages to it, finding it is still alive. |
807 |
|
808 |
=item * Erlang can send messages to the wrong port, AEMP does not. |
809 |
|
810 |
In Erlang it is quite likely that a node that restarts reuses a process ID |
811 |
known to other nodes for a completely different process, causing messages |
812 |
destined for that process to end up in an unrelated process. |
813 |
|
814 |
AEMP never reuses port IDs, so old messages or old port IDs floating |
815 |
around in the network will not be sent to an unrelated port. |
816 |
|
817 |
=item * Erlang uses unprotected connections, AEMP uses secure |
818 |
authentication and can use TLS. |
819 |
|
820 |
AEMP can use a proven protocol - SSL/TLS - to protect connections and |
821 |
securely authenticate nodes. |
822 |
|
823 |
=item * The AEMP protocol is optimised for both text-based and binary |
824 |
communications. |
825 |
|
826 |
The AEMP protocol, unlike the Erlang protocol, supports both |
827 |
language-independent text-only protocols (good for debugging) and binary, |
828 |
language-specific serialisers (e.g. Storable). |
829 |
|
830 |
It has also been carefully designed to be implementable in other languages |
831 |
with a minimum of work while gracefully degrading fucntionality to make the |
832 |
protocol simple. |
833 |
|
834 |
=item * AEMP has more flexible monitoring options than Erlang. |
835 |
|
836 |
In Erlang, you can chose to receive I<all> exit signals as messages |
837 |
or I<none>, there is no in-between, so monitoring single processes is |
838 |
difficult to implement. Monitoring in AEMP is more flexible than in |
839 |
Erlang, as one can choose between automatic kill, exit message or callback |
840 |
on a per-process basis. |
841 |
|
842 |
=item * Erlang tries to hide remote/local connections, AEMP does not. |
843 |
|
844 |
Monitoring in Erlang is not an indicator of process death/crashes, |
845 |
as linking is (except linking is unreliable in Erlang). |
846 |
|
847 |
In AEMP, you don't "look up" registered port names or send to named ports |
848 |
that might or might not be persistent. Instead, you normally spawn a port |
849 |
on the remote node. The init function monitors the you, and you monitor |
850 |
the remote port. Since both monitors are local to the node, they are much |
851 |
more reliable. |
852 |
|
853 |
This also saves round-trips and avoids sending messages to the wrong port |
854 |
(hard to do in Erlang). |
855 |
|
856 |
=back |
857 |
|
858 |
=head1 RATIONALE |
859 |
|
860 |
=over 4 |
861 |
|
862 |
=item Why strings for ports and noderefs, why not objects? |
863 |
|
864 |
We considered "objects", but found that the actual number of methods |
865 |
thatc an be called are very low. Since port IDs and noderefs travel over |
866 |
the network frequently, the serialising/deserialising would add lots of |
867 |
overhead, as well as having to keep a proxy object. |
868 |
|
869 |
Strings can easily be printed, easily serialised etc. and need no special |
870 |
procedures to be "valid". |
871 |
|
872 |
And a a miniport consists of a single closure stored in a global hash - it |
873 |
can't become much cheaper. |
874 |
|
875 |
=item Why favour JSON, why not real serialising format such as Storable? |
876 |
|
877 |
In fact, any AnyEvent::MP node will happily accept Storable as framing |
878 |
format, but currently there is no way to make a node use Storable by |
879 |
default. |
880 |
|
881 |
The default framing protocol is JSON because a) JSON::XS is many times |
882 |
faster for small messages and b) most importantly, after years of |
883 |
experience we found that object serialisation is causing more problems |
884 |
than it gains: Just like function calls, objects simply do not travel |
885 |
easily over the network, mostly because they will always be a copy, so you |
886 |
always have to re-think your design. |
887 |
|
888 |
Keeping your messages simple, concentrating on data structures rather than |
889 |
objects, will keep your messages clean, tidy and efficient. |
890 |
|
891 |
=back |
892 |
|
893 |
=head1 SEE ALSO |
894 |
|
895 |
L<AnyEvent>. |
896 |
|
897 |
=head1 AUTHOR |
898 |
|
899 |
Marc Lehmann <schmorp@schmorp.de> |
900 |
http://home.schmorp.de/ |
901 |
|
902 |
=cut |
903 |
|
904 |
1 |
905 |
|