1 |
=head1 NAME |
2 |
|
3 |
AnyEvent::MP - erlang-style multi-processing/message-passing framework |
4 |
|
5 |
=head1 SYNOPSIS |
6 |
|
7 |
use AnyEvent::MP; |
8 |
|
9 |
$NODE # contains this node's node ID |
10 |
NODE # returns this node's node ID |
11 |
|
12 |
$SELF # receiving/own port id in rcv callbacks |
13 |
|
14 |
# initialise the node so it can send/receive messages |
15 |
configure; |
16 |
|
17 |
# ports are message destinations |
18 |
|
19 |
# sending messages |
20 |
snd $port, type => data...; |
21 |
snd $port, @msg; |
22 |
snd @msg_with_first_element_being_a_port; |
23 |
|
24 |
# creating/using ports, the simple way |
25 |
my $simple_port = port { my @msg = @_ }; |
26 |
|
27 |
# creating/using ports, tagged message matching |
28 |
my $port = port; |
29 |
rcv $port, ping => sub { snd $_[0], "pong" }; |
30 |
rcv $port, pong => sub { warn "pong received\n" }; |
31 |
|
32 |
# create a port on another node |
33 |
my $port = spawn $node, $initfunc, @initdata; |
34 |
|
35 |
# destroy a port again |
36 |
kil $port; # "normal" kill |
37 |
kil $port, my_error => "everything is broken"; # error kill |
38 |
|
39 |
# monitoring |
40 |
mon $port, $cb->(@msg) # callback is invoked on death |
41 |
mon $port, $localport # kill localport on abnormal death |
42 |
mon $port, $localport, @msg # send message on death |
43 |
|
44 |
# temporarily execute code in port context |
45 |
peval $port, sub { die "kill the port!" }; |
46 |
|
47 |
# execute callbacks in $SELF port context |
48 |
my $timer = AE::timer 1, 0, psub { |
49 |
die "kill the port, delayed"; |
50 |
}; |
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, and you can supervise entities remotely. |
58 |
|
59 |
For an introduction to this module family, see the L<AnyEvent::MP::Intro> |
60 |
manual page and the examples under F<eg/>. |
61 |
|
62 |
=head1 CONCEPTS |
63 |
|
64 |
=over 4 |
65 |
|
66 |
=item port |
67 |
|
68 |
Not to be confused with a TCP port, a "port" is something you can send |
69 |
messages to (with the C<snd> function). |
70 |
|
71 |
Ports allow you to register C<rcv> handlers that can match all or just |
72 |
some messages. Messages send to ports will not be queued, regardless of |
73 |
anything was listening for them or not. |
74 |
|
75 |
Ports are represented by (printable) strings called "port IDs". |
76 |
|
77 |
=item port ID - C<nodeid#portname> |
78 |
|
79 |
A port ID is the concatenation of a node ID, a hash-mark (C<#>) |
80 |
as separator, and a port name (a printable string of unspecified |
81 |
format created by AnyEvent::MP). |
82 |
|
83 |
=item node |
84 |
|
85 |
A node is a single process containing at least one port - the node port, |
86 |
which enables nodes to manage each other remotely, and to create new |
87 |
ports. |
88 |
|
89 |
Nodes are either public (have one or more listening ports) or private |
90 |
(no listening ports). Private nodes cannot talk to other private nodes |
91 |
currently, but all nodes can talk to public nodes. |
92 |
|
93 |
Nodes is represented by (printable) strings called "node IDs". |
94 |
|
95 |
=item node ID - C<[A-Za-z0-9_\-.:]*> |
96 |
|
97 |
A node ID is a string that uniquely identifies the node within a |
98 |
network. Depending on the configuration used, node IDs can look like a |
99 |
hostname, a hostname and a port, or a random string. AnyEvent::MP itself |
100 |
doesn't interpret node IDs in any way except to uniquely identify a node. |
101 |
|
102 |
=item binds - C<ip:port> |
103 |
|
104 |
Nodes can only talk to each other by creating some kind of connection to |
105 |
each other. To do this, nodes should listen on one or more local transport |
106 |
endpoints - binds. |
107 |
|
108 |
Currently, only standard C<ip:port> specifications can be used, which |
109 |
specify TCP ports to listen on. So a bind is basically just a tcp socket |
110 |
in listening mode thta accepts conenctions form other nodes. |
111 |
|
112 |
=item seed nodes |
113 |
|
114 |
When a node starts, it knows nothing about the network it is in - it |
115 |
needs to connect to at least one other node that is already in the |
116 |
network. These other nodes are called "seed nodes". |
117 |
|
118 |
Seed nodes themselves are not special - they are seed nodes only because |
119 |
some other node I<uses> them as such, but any node can be used as seed |
120 |
node for other nodes, and eahc node cna use a different set of seed nodes. |
121 |
|
122 |
In addition to discovering the network, seed nodes are also used to |
123 |
maintain the network - all nodes using the same seed node form are part of |
124 |
the same network. If a network is split into multiple subnets because e.g. |
125 |
the network link between the parts goes down, then using the same seed |
126 |
nodes for all nodes ensures that eventually the subnets get merged again. |
127 |
|
128 |
Seed nodes are expected to be long-running, and at least one seed node |
129 |
should always be available. They should also be relatively responsive - a |
130 |
seed node that blocks for long periods will slow down everybody else. |
131 |
|
132 |
For small networks, it's best if every node uses the same set of seed |
133 |
nodes. For large networks, it can be useful to specify "regional" seed |
134 |
nodes for most nodes in an area, and use all seed nodes as seed nodes for |
135 |
each other. What's important is that all seed nodes connections form a |
136 |
complete graph, so that the network cannot split into separate subnets |
137 |
forever. |
138 |
|
139 |
Seed nodes are represented by seed IDs. |
140 |
|
141 |
=item seed IDs - C<host:port> |
142 |
|
143 |
Seed IDs are transport endpoint(s) (usually a hostname/IP address and a |
144 |
TCP port) of nodes that should be used as seed nodes. |
145 |
|
146 |
=item global nodes |
147 |
|
148 |
An AEMP network needs a discovery service - nodes need to know how to |
149 |
connect to other nodes they only know by name. In addition, AEMP offers a |
150 |
distributed "group database", which maps group names to a list of strings |
151 |
- for example, to register worker ports. |
152 |
|
153 |
A network needs at least one global node to work, and allows every node to |
154 |
be a global node. |
155 |
|
156 |
Any node that loads the L<AnyEvent::MP::Global> module becomes a global |
157 |
node and tries to keep connections to all other nodes. So while it can |
158 |
make sense to make every node "global" in small networks, it usually makes |
159 |
sense to only make seed nodes into global nodes in large networks (nodes |
160 |
keep connections to seed nodes and global nodes, so makign them the same |
161 |
reduces overhead). |
162 |
|
163 |
=back |
164 |
|
165 |
=head1 VARIABLES/FUNCTIONS |
166 |
|
167 |
=over 4 |
168 |
|
169 |
=cut |
170 |
|
171 |
package AnyEvent::MP; |
172 |
|
173 |
use AnyEvent::MP::Config (); |
174 |
use AnyEvent::MP::Kernel; |
175 |
use AnyEvent::MP::Kernel qw(%NODE %PORT %PORT_DATA $UNIQ $RUNIQ $ID); |
176 |
|
177 |
use common::sense; |
178 |
|
179 |
use Carp (); |
180 |
|
181 |
use AnyEvent (); |
182 |
use Guard (); |
183 |
|
184 |
use base "Exporter"; |
185 |
|
186 |
our $VERSION = $AnyEvent::MP::Config::VERSION; |
187 |
|
188 |
our @EXPORT = qw( |
189 |
NODE $NODE *SELF node_of after |
190 |
configure |
191 |
snd rcv mon mon_guard kil psub peval spawn cal |
192 |
port |
193 |
db_set db_del db_reg |
194 |
db_mon db_family db_keys db_values |
195 |
); |
196 |
|
197 |
our $SELF; |
198 |
|
199 |
sub _self_die() { |
200 |
my $msg = $@; |
201 |
$msg =~ s/\n+$// unless ref $msg; |
202 |
kil $SELF, die => $msg; |
203 |
} |
204 |
|
205 |
=item $thisnode = NODE / $NODE |
206 |
|
207 |
The C<NODE> function returns, and the C<$NODE> variable contains, the node |
208 |
ID of the node running in the current process. This value is initialised by |
209 |
a call to C<configure>. |
210 |
|
211 |
=item $nodeid = node_of $port |
212 |
|
213 |
Extracts and returns the node ID from a port ID or a node ID. |
214 |
|
215 |
=item configure $profile, key => value... |
216 |
|
217 |
=item configure key => value... |
218 |
|
219 |
Before a node can talk to other nodes on the network (i.e. enter |
220 |
"distributed mode") it has to configure itself - the minimum a node needs |
221 |
to know is its own name, and optionally it should know the addresses of |
222 |
some other nodes in the network to discover other nodes. |
223 |
|
224 |
This function configures a node - it must be called exactly once (or |
225 |
never) before calling other AnyEvent::MP functions. |
226 |
|
227 |
The key/value pairs are basically the same ones as documented for the |
228 |
F<aemp> command line utility (sans the set/del prefix), with these additions: |
229 |
|
230 |
=over 4 |
231 |
|
232 |
=item norc => $boolean (default false) |
233 |
|
234 |
If true, then the rc file (e.g. F<~/.perl-anyevent-mp>) will I<not> |
235 |
be consulted - all configuraiton options must be specified in the |
236 |
C<configure> call. |
237 |
|
238 |
=item force => $boolean (default false) |
239 |
|
240 |
IF true, then the values specified in the C<configure> will take |
241 |
precedence over any values configured via the rc file. The default is for |
242 |
the rc file to override any options specified in the program. |
243 |
|
244 |
=item secure => $pass->(@msg) |
245 |
|
246 |
In addition to specifying a boolean, you can specify a code reference that |
247 |
is called for every code execution attempt - the execution request is |
248 |
granted iff the callback returns a true value. |
249 |
|
250 |
Most of the time the callback should look only at |
251 |
C<$AnyEvent::MP::Kernel::SRCNODE> to make a decision, and not at the |
252 |
actual message (which can be about anything, and is mostly provided for |
253 |
diagnostic purposes). |
254 |
|
255 |
See F<semp setsecure> for more info. |
256 |
|
257 |
=back |
258 |
|
259 |
=over 4 |
260 |
|
261 |
=item step 1, gathering configuration from profiles |
262 |
|
263 |
The function first looks up a profile in the aemp configuration (see the |
264 |
L<aemp> commandline utility). The profile name can be specified via the |
265 |
named C<profile> parameter or can simply be the first parameter). If it is |
266 |
missing, then the nodename (F<uname -n>) will be used as profile name. |
267 |
|
268 |
The profile data is then gathered as follows: |
269 |
|
270 |
First, all remaining key => value pairs (all of which are conveniently |
271 |
undocumented at the moment) will be interpreted as configuration |
272 |
data. Then they will be overwritten by any values specified in the global |
273 |
default configuration (see the F<aemp> utility), then the chain of |
274 |
profiles chosen by the profile name (and any C<parent> attributes). |
275 |
|
276 |
That means that the values specified in the profile have highest priority |
277 |
and the values specified directly via C<configure> have lowest priority, |
278 |
and can only be used to specify defaults. |
279 |
|
280 |
If the profile specifies a node ID, then this will become the node ID of |
281 |
this process. If not, then the profile name will be used as node ID, with |
282 |
a unique randoms tring (C</%u>) appended. |
283 |
|
284 |
The node ID can contain some C<%> sequences that are expanded: C<%n> |
285 |
is expanded to the local nodename, C<%u> is replaced by a random |
286 |
strign to make the node unique. For example, the F<aemp> commandline |
287 |
utility uses C<aemp/%n/%u> as nodename, which might expand to |
288 |
C<aemp/cerebro/ZQDGSIkRhEZQDGSIkRhE>. |
289 |
|
290 |
=item step 2, bind listener sockets |
291 |
|
292 |
The next step is to look up the binds in the profile, followed by binding |
293 |
aemp protocol listeners on all binds specified (it is possible and valid |
294 |
to have no binds, meaning that the node cannot be contacted form the |
295 |
outside. This means the node cannot talk to other nodes that also have no |
296 |
binds, but it can still talk to all "normal" nodes). |
297 |
|
298 |
If the profile does not specify a binds list, then a default of C<*> is |
299 |
used, meaning the node will bind on a dynamically-assigned port on every |
300 |
local IP address it finds. |
301 |
|
302 |
=item step 3, connect to seed nodes |
303 |
|
304 |
As the last step, the seed ID list from the profile is passed to the |
305 |
L<AnyEvent::MP::Global> module, which will then use it to keep |
306 |
connectivity with at least one node at any point in time. |
307 |
|
308 |
=back |
309 |
|
310 |
Example: become a distributed node using the local node name as profile. |
311 |
This should be the most common form of invocation for "daemon"-type nodes. |
312 |
|
313 |
configure |
314 |
|
315 |
Example: become a semi-anonymous node. This form is often used for |
316 |
commandline clients. |
317 |
|
318 |
configure nodeid => "myscript/%n/%u"; |
319 |
|
320 |
Example: configure a node using a profile called seed, which is suitable |
321 |
for a seed node as it binds on all local addresses on a fixed port (4040, |
322 |
customary for aemp). |
323 |
|
324 |
# use the aemp commandline utility |
325 |
# aemp profile seed binds '*:4040' |
326 |
|
327 |
# then use it |
328 |
configure profile => "seed"; |
329 |
|
330 |
# or simply use aemp from the shell again: |
331 |
# aemp run profile seed |
332 |
|
333 |
# or provide a nicer-to-remember nodeid |
334 |
# aemp run profile seed nodeid "$(hostname)" |
335 |
|
336 |
=item $SELF |
337 |
|
338 |
Contains the current port id while executing C<rcv> callbacks or C<psub> |
339 |
blocks. |
340 |
|
341 |
=item *SELF, SELF, %SELF, @SELF... |
342 |
|
343 |
Due to some quirks in how perl exports variables, it is impossible to |
344 |
just export C<$SELF>, all the symbols named C<SELF> are exported by this |
345 |
module, but only C<$SELF> is currently used. |
346 |
|
347 |
=item snd $port, type => @data |
348 |
|
349 |
=item snd $port, @msg |
350 |
|
351 |
Send the given message to the given port, which can identify either a |
352 |
local or a remote port, and must be a port ID. |
353 |
|
354 |
While the message can be almost anything, it is highly recommended to |
355 |
use a string as first element (a port ID, or some word that indicates a |
356 |
request type etc.) and to consist if only simple perl values (scalars, |
357 |
arrays, hashes) - if you think you need to pass an object, think again. |
358 |
|
359 |
The message data logically becomes read-only after a call to this |
360 |
function: modifying any argument (or values referenced by them) is |
361 |
forbidden, as there can be considerable time between the call to C<snd> |
362 |
and the time the message is actually being serialised - in fact, it might |
363 |
never be copied as within the same process it is simply handed to the |
364 |
receiving port. |
365 |
|
366 |
The type of data you can transfer depends on the transport protocol: when |
367 |
JSON is used, then only strings, numbers and arrays and hashes consisting |
368 |
of those are allowed (no objects). When Storable is used, then anything |
369 |
that Storable can serialise and deserialise is allowed, and for the local |
370 |
node, anything can be passed. Best rely only on the common denominator of |
371 |
these. |
372 |
|
373 |
=item $local_port = port |
374 |
|
375 |
Create a new local port object and returns its port ID. Initially it has |
376 |
no callbacks set and will throw an error when it receives messages. |
377 |
|
378 |
=item $local_port = port { my @msg = @_ } |
379 |
|
380 |
Creates a new local port, and returns its ID. Semantically the same as |
381 |
creating a port and calling C<rcv $port, $callback> on it. |
382 |
|
383 |
The block will be called for every message received on the port, with the |
384 |
global variable C<$SELF> set to the port ID. Runtime errors will cause the |
385 |
port to be C<kil>ed. The message will be passed as-is, no extra argument |
386 |
(i.e. no port ID) will be passed to the callback. |
387 |
|
388 |
If you want to stop/destroy the port, simply C<kil> it: |
389 |
|
390 |
my $port = port { |
391 |
my @msg = @_; |
392 |
... |
393 |
kil $SELF; |
394 |
}; |
395 |
|
396 |
=cut |
397 |
|
398 |
sub rcv($@); |
399 |
|
400 |
my $KILME = sub { |
401 |
(my $tag = substr $_[0], 0, 30) =~ s/([\x20-\x7e])/./g; |
402 |
kil $SELF, unhandled_message => "no callback found for message '$tag'"; |
403 |
}; |
404 |
|
405 |
sub port(;&) { |
406 |
my $id = $UNIQ . ++$ID; |
407 |
my $port = "$NODE#$id"; |
408 |
|
409 |
rcv $port, shift || $KILME; |
410 |
|
411 |
$port |
412 |
} |
413 |
|
414 |
=item rcv $local_port, $callback->(@msg) |
415 |
|
416 |
Replaces the default callback on the specified port. There is no way to |
417 |
remove the default callback: use C<sub { }> to disable it, or better |
418 |
C<kil> the port when it is no longer needed. |
419 |
|
420 |
The global C<$SELF> (exported by this module) contains C<$port> while |
421 |
executing the callback. Runtime errors during callback execution will |
422 |
result in the port being C<kil>ed. |
423 |
|
424 |
The default callback receives all messages not matched by a more specific |
425 |
C<tag> match. |
426 |
|
427 |
=item rcv $local_port, tag => $callback->(@msg_without_tag), ... |
428 |
|
429 |
Register (or replace) callbacks to be called on messages starting with the |
430 |
given tag on the given port (and return the port), or unregister it (when |
431 |
C<$callback> is C<$undef> or missing). There can only be one callback |
432 |
registered for each tag. |
433 |
|
434 |
The original message will be passed to the callback, after the first |
435 |
element (the tag) has been removed. The callback will use the same |
436 |
environment as the default callback (see above). |
437 |
|
438 |
Example: create a port and bind receivers on it in one go. |
439 |
|
440 |
my $port = rcv port, |
441 |
msg1 => sub { ... }, |
442 |
msg2 => sub { ... }, |
443 |
; |
444 |
|
445 |
Example: create a port, bind receivers and send it in a message elsewhere |
446 |
in one go: |
447 |
|
448 |
snd $otherport, reply => |
449 |
rcv port, |
450 |
msg1 => sub { ... }, |
451 |
... |
452 |
; |
453 |
|
454 |
Example: temporarily register a rcv callback for a tag matching some port |
455 |
(e.g. for an rpc reply) and unregister it after a message was received. |
456 |
|
457 |
rcv $port, $otherport => sub { |
458 |
my @reply = @_; |
459 |
|
460 |
rcv $SELF, $otherport; |
461 |
}; |
462 |
|
463 |
=cut |
464 |
|
465 |
sub rcv($@) { |
466 |
my $port = shift; |
467 |
my ($nodeid, $portid) = split /#/, $port, 2; |
468 |
|
469 |
$NODE{$nodeid} == $NODE{""} |
470 |
or Carp::croak "$port: rcv can only be called on local ports, caught"; |
471 |
|
472 |
while (@_) { |
473 |
if (ref $_[0]) { |
474 |
if (my $self = $PORT_DATA{$portid}) { |
475 |
"AnyEvent::MP::Port" eq ref $self |
476 |
or Carp::croak "$port: rcv can only be called on message matching ports, caught"; |
477 |
|
478 |
$self->[0] = shift; |
479 |
} else { |
480 |
my $cb = shift; |
481 |
$PORT{$portid} = sub { |
482 |
local $SELF = $port; |
483 |
eval { &$cb }; _self_die if $@; |
484 |
}; |
485 |
} |
486 |
} elsif (defined $_[0]) { |
487 |
my $self = $PORT_DATA{$portid} ||= do { |
488 |
my $self = bless [$PORT{$portid} || sub { }, { }, $port], "AnyEvent::MP::Port"; |
489 |
|
490 |
$PORT{$portid} = sub { |
491 |
local $SELF = $port; |
492 |
|
493 |
if (my $cb = $self->[1]{$_[0]}) { |
494 |
shift; |
495 |
eval { &$cb }; _self_die if $@; |
496 |
} else { |
497 |
&{ $self->[0] }; |
498 |
} |
499 |
}; |
500 |
|
501 |
$self |
502 |
}; |
503 |
|
504 |
"AnyEvent::MP::Port" eq ref $self |
505 |
or Carp::croak "$port: rcv can only be called on message matching ports, caught"; |
506 |
|
507 |
my ($tag, $cb) = splice @_, 0, 2; |
508 |
|
509 |
if (defined $cb) { |
510 |
$self->[1]{$tag} = $cb; |
511 |
} else { |
512 |
delete $self->[1]{$tag}; |
513 |
} |
514 |
} |
515 |
} |
516 |
|
517 |
$port |
518 |
} |
519 |
|
520 |
=item peval $port, $coderef[, @args] |
521 |
|
522 |
Evaluates the given C<$codref> within the contetx of C<$port>, that is, |
523 |
when the code throews an exception the C<$port> will be killed. |
524 |
|
525 |
Any remaining args will be passed to the callback. Any return values will |
526 |
be returned to the caller. |
527 |
|
528 |
This is useful when you temporarily want to execute code in the context of |
529 |
a port. |
530 |
|
531 |
Example: create a port and run some initialisation code in it's context. |
532 |
|
533 |
my $port = port { ... }; |
534 |
|
535 |
peval $port, sub { |
536 |
init |
537 |
or die "unable to init"; |
538 |
}; |
539 |
|
540 |
=cut |
541 |
|
542 |
sub peval($$) { |
543 |
local $SELF = shift; |
544 |
my $cb = shift; |
545 |
|
546 |
if (wantarray) { |
547 |
my @res = eval { &$cb }; |
548 |
_self_die if $@; |
549 |
@res |
550 |
} else { |
551 |
my $res = eval { &$cb }; |
552 |
_self_die if $@; |
553 |
$res |
554 |
} |
555 |
} |
556 |
|
557 |
=item $closure = psub { BLOCK } |
558 |
|
559 |
Remembers C<$SELF> and creates a closure out of the BLOCK. When the |
560 |
closure is executed, sets up the environment in the same way as in C<rcv> |
561 |
callbacks, i.e. runtime errors will cause the port to get C<kil>ed. |
562 |
|
563 |
The effect is basically as if it returned C<< sub { peval $SELF, sub { |
564 |
BLOCK }, @_ } >>. |
565 |
|
566 |
This is useful when you register callbacks from C<rcv> callbacks: |
567 |
|
568 |
rcv delayed_reply => sub { |
569 |
my ($delay, @reply) = @_; |
570 |
my $timer = AE::timer $delay, 0, psub { |
571 |
snd @reply, $SELF; |
572 |
}; |
573 |
}; |
574 |
|
575 |
=cut |
576 |
|
577 |
sub psub(&) { |
578 |
my $cb = shift; |
579 |
|
580 |
my $port = $SELF |
581 |
or Carp::croak "psub can only be called from within rcv or psub callbacks, not"; |
582 |
|
583 |
sub { |
584 |
local $SELF = $port; |
585 |
|
586 |
if (wantarray) { |
587 |
my @res = eval { &$cb }; |
588 |
_self_die if $@; |
589 |
@res |
590 |
} else { |
591 |
my $res = eval { &$cb }; |
592 |
_self_die if $@; |
593 |
$res |
594 |
} |
595 |
} |
596 |
} |
597 |
|
598 |
=item $guard = mon $port, $rcvport # kill $rcvport when $port dies |
599 |
|
600 |
=item $guard = mon $port # kill $SELF when $port dies |
601 |
|
602 |
=item $guard = mon $port, $cb->(@reason) # call $cb when $port dies |
603 |
|
604 |
=item $guard = mon $port, $rcvport, @msg # send a message when $port dies |
605 |
|
606 |
Monitor the given port and do something when the port is killed or |
607 |
messages to it were lost, and optionally return a guard that can be used |
608 |
to stop monitoring again. |
609 |
|
610 |
The first two forms distinguish between "normal" and "abnormal" kil's: |
611 |
|
612 |
In the first form (another port given), if the C<$port> is C<kil>'ed with |
613 |
a non-empty reason, the other port (C<$rcvport>) will be kil'ed with the |
614 |
same reason. That is, on "normal" kil's nothing happens, while under all |
615 |
other conditions, the other port is killed with the same reason. |
616 |
|
617 |
The second form (kill self) is the same as the first form, except that |
618 |
C<$rvport> defaults to C<$SELF>. |
619 |
|
620 |
The remaining forms don't distinguish between "normal" and "abnormal" kil's |
621 |
- it's up to the callback or receiver to check whether the C<@reason> is |
622 |
empty and act accordingly. |
623 |
|
624 |
In the third form (callback), the callback is simply called with any |
625 |
number of C<@reason> elements (empty @reason means that the port was deleted |
626 |
"normally"). Note also that I<< the callback B<must> never die >>, so use |
627 |
C<eval> if unsure. |
628 |
|
629 |
In the last form (message), a message of the form C<$rcvport, @msg, |
630 |
@reason> will be C<snd>. |
631 |
|
632 |
Monitoring-actions are one-shot: once messages are lost (and a monitoring |
633 |
alert was raised), they are removed and will not trigger again, even if it |
634 |
turns out that the port is still alive. |
635 |
|
636 |
As a rule of thumb, monitoring requests should always monitor a remote |
637 |
port locally (using a local C<$rcvport> or a callback). The reason is that |
638 |
kill messages might get lost, just like any other message. Another less |
639 |
obvious reason is that even monitoring requests can get lost (for example, |
640 |
when the connection to the other node goes down permanently). When |
641 |
monitoring a port locally these problems do not exist. |
642 |
|
643 |
C<mon> effectively guarantees that, in the absence of hardware failures, |
644 |
after starting the monitor, either all messages sent to the port will |
645 |
arrive, or the monitoring action will be invoked after possible message |
646 |
loss has been detected. No messages will be lost "in between" (after |
647 |
the first lost message no further messages will be received by the |
648 |
port). After the monitoring action was invoked, further messages might get |
649 |
delivered again. |
650 |
|
651 |
Inter-host-connection timeouts and monitoring depend on the transport |
652 |
used. The only transport currently implemented is TCP, and AnyEvent::MP |
653 |
relies on TCP to detect node-downs (this can take 10-15 minutes on a |
654 |
non-idle connection, and usually around two hours for idle connections). |
655 |
|
656 |
This means that monitoring is good for program errors and cleaning up |
657 |
stuff eventually, but they are no replacement for a timeout when you need |
658 |
to ensure some maximum latency. |
659 |
|
660 |
Example: call a given callback when C<$port> is killed. |
661 |
|
662 |
mon $port, sub { warn "port died because of <@_>\n" }; |
663 |
|
664 |
Example: kill ourselves when C<$port> is killed abnormally. |
665 |
|
666 |
mon $port; |
667 |
|
668 |
Example: send us a restart message when another C<$port> is killed. |
669 |
|
670 |
mon $port, $self => "restart"; |
671 |
|
672 |
=cut |
673 |
|
674 |
sub mon { |
675 |
my ($nodeid, $port) = split /#/, shift, 2; |
676 |
|
677 |
my $node = $NODE{$nodeid} || add_node $nodeid; |
678 |
|
679 |
my $cb = @_ ? shift : $SELF || Carp::croak 'mon: called with one argument only, but $SELF not set,'; |
680 |
|
681 |
unless (ref $cb) { |
682 |
if (@_) { |
683 |
# send a kill info message |
684 |
my (@msg) = ($cb, @_); |
685 |
$cb = sub { snd @msg, @_ }; |
686 |
} else { |
687 |
# simply kill other port |
688 |
my $port = $cb; |
689 |
$cb = sub { kil $port, @_ if @_ }; |
690 |
} |
691 |
} |
692 |
|
693 |
$node->monitor ($port, $cb); |
694 |
|
695 |
defined wantarray |
696 |
and ($cb += 0, Guard::guard { $node->unmonitor ($port, $cb) }) |
697 |
} |
698 |
|
699 |
=item $guard = mon_guard $port, $ref, $ref... |
700 |
|
701 |
Monitors the given C<$port> and keeps the passed references. When the port |
702 |
is killed, the references will be freed. |
703 |
|
704 |
Optionally returns a guard that will stop the monitoring. |
705 |
|
706 |
This function is useful when you create e.g. timers or other watchers and |
707 |
want to free them when the port gets killed (note the use of C<psub>): |
708 |
|
709 |
$port->rcv (start => sub { |
710 |
my $timer; $timer = mon_guard $port, AE::timer 1, 1, psub { |
711 |
undef $timer if 0.9 < rand; |
712 |
}); |
713 |
}); |
714 |
|
715 |
=cut |
716 |
|
717 |
sub mon_guard { |
718 |
my ($port, @refs) = @_; |
719 |
|
720 |
#TODO: mon-less form? |
721 |
|
722 |
mon $port, sub { 0 && @refs } |
723 |
} |
724 |
|
725 |
=item kil $port[, @reason] |
726 |
|
727 |
Kill the specified port with the given C<@reason>. |
728 |
|
729 |
If no C<@reason> is specified, then the port is killed "normally" - |
730 |
monitor callback will be invoked, but the kil will not cause linked ports |
731 |
(C<mon $mport, $lport> form) to get killed. |
732 |
|
733 |
If a C<@reason> is specified, then linked ports (C<mon $mport, $lport> |
734 |
form) get killed with the same reason. |
735 |
|
736 |
Runtime errors while evaluating C<rcv> callbacks or inside C<psub> blocks |
737 |
will be reported as reason C<< die => $@ >>. |
738 |
|
739 |
Transport/communication errors are reported as C<< transport_error => |
740 |
$message >>. |
741 |
|
742 |
Common idioms: |
743 |
|
744 |
# silently remove yourself, do not kill linked ports |
745 |
kil $SELF; |
746 |
|
747 |
# report a failure in some detail |
748 |
kil $SELF, failure_mode_1 => "it failed with too high temperature"; |
749 |
|
750 |
# do not waste much time with killing, just die when something goes wrong |
751 |
open my $fh, "<file" |
752 |
or die "file: $!"; |
753 |
|
754 |
=item $port = spawn $node, $initfunc[, @initdata] |
755 |
|
756 |
Creates a port on the node C<$node> (which can also be a port ID, in which |
757 |
case it's the node where that port resides). |
758 |
|
759 |
The port ID of the newly created port is returned immediately, and it is |
760 |
possible to immediately start sending messages or to monitor the port. |
761 |
|
762 |
After the port has been created, the init function is called on the remote |
763 |
node, in the same context as a C<rcv> callback. This function must be a |
764 |
fully-qualified function name (e.g. C<MyApp::Chat::Server::init>). To |
765 |
specify a function in the main program, use C<::name>. |
766 |
|
767 |
If the function doesn't exist, then the node tries to C<require> |
768 |
the package, then the package above the package and so on (e.g. |
769 |
C<MyApp::Chat::Server>, C<MyApp::Chat>, C<MyApp>) until the function |
770 |
exists or it runs out of package names. |
771 |
|
772 |
The init function is then called with the newly-created port as context |
773 |
object (C<$SELF>) and the C<@initdata> values as arguments. It I<must> |
774 |
call one of the C<rcv> functions to set callbacks on C<$SELF>, otherwise |
775 |
the port might not get created. |
776 |
|
777 |
A common idiom is to pass a local port, immediately monitor the spawned |
778 |
port, and in the remote init function, immediately monitor the passed |
779 |
local port. This two-way monitoring ensures that both ports get cleaned up |
780 |
when there is a problem. |
781 |
|
782 |
C<spawn> guarantees that the C<$initfunc> has no visible effects on the |
783 |
caller before C<spawn> returns (by delaying invocation when spawn is |
784 |
called for the local node). |
785 |
|
786 |
Example: spawn a chat server port on C<$othernode>. |
787 |
|
788 |
# this node, executed from within a port context: |
789 |
my $server = spawn $othernode, "MyApp::Chat::Server::connect", $SELF; |
790 |
mon $server; |
791 |
|
792 |
# init function on C<$othernode> |
793 |
sub connect { |
794 |
my ($srcport) = @_; |
795 |
|
796 |
mon $srcport; |
797 |
|
798 |
rcv $SELF, sub { |
799 |
... |
800 |
}; |
801 |
} |
802 |
|
803 |
=cut |
804 |
|
805 |
sub _spawn { |
806 |
my $port = shift; |
807 |
my $init = shift; |
808 |
|
809 |
# rcv will create the actual port |
810 |
local $SELF = "$NODE#$port"; |
811 |
eval { |
812 |
&{ load_func $init } |
813 |
}; |
814 |
_self_die if $@; |
815 |
} |
816 |
|
817 |
sub spawn(@) { |
818 |
my ($nodeid, undef) = split /#/, shift, 2; |
819 |
|
820 |
my $id = $RUNIQ . ++$ID; |
821 |
|
822 |
$_[0] =~ /::/ |
823 |
or Carp::croak "spawn init function must be a fully-qualified name, caught"; |
824 |
|
825 |
snd_to_func $nodeid, "AnyEvent::MP::_spawn" => $id, @_; |
826 |
|
827 |
"$nodeid#$id" |
828 |
} |
829 |
|
830 |
|
831 |
=item after $timeout, @msg |
832 |
|
833 |
=item after $timeout, $callback |
834 |
|
835 |
Either sends the given message, or call the given callback, after the |
836 |
specified number of seconds. |
837 |
|
838 |
This is simply a utility function that comes in handy at times - the |
839 |
AnyEvent::MP author is not convinced of the wisdom of having it, though, |
840 |
so it may go away in the future. |
841 |
|
842 |
=cut |
843 |
|
844 |
sub after($@) { |
845 |
my ($timeout, @action) = @_; |
846 |
|
847 |
my $t; $t = AE::timer $timeout, 0, sub { |
848 |
undef $t; |
849 |
ref $action[0] |
850 |
? $action[0]() |
851 |
: snd @action; |
852 |
}; |
853 |
} |
854 |
|
855 |
#=item $cb2 = timeout $seconds, $cb[, @args] |
856 |
|
857 |
=item cal $port, @msg, $callback[, $timeout] |
858 |
|
859 |
A simple form of RPC - sends a message to the given C<$port> with the |
860 |
given contents (C<@msg>), but adds a reply port to the message. |
861 |
|
862 |
The reply port is created temporarily just for the purpose of receiving |
863 |
the reply, and will be C<kil>ed when no longer needed. |
864 |
|
865 |
A reply message sent to the port is passed to the C<$callback> as-is. |
866 |
|
867 |
If an optional time-out (in seconds) is given and it is not C<undef>, |
868 |
then the callback will be called without any arguments after the time-out |
869 |
elapsed and the port is C<kil>ed. |
870 |
|
871 |
If no time-out is given (or it is C<undef>), then the local port will |
872 |
monitor the remote port instead, so it eventually gets cleaned-up. |
873 |
|
874 |
Currently this function returns the temporary port, but this "feature" |
875 |
might go in future versions unless you can make a convincing case that |
876 |
this is indeed useful for something. |
877 |
|
878 |
=cut |
879 |
|
880 |
sub cal(@) { |
881 |
my $timeout = ref $_[-1] ? undef : pop; |
882 |
my $cb = pop; |
883 |
|
884 |
my $port = port { |
885 |
undef $timeout; |
886 |
kil $SELF; |
887 |
&$cb; |
888 |
}; |
889 |
|
890 |
if (defined $timeout) { |
891 |
$timeout = AE::timer $timeout, 0, sub { |
892 |
undef $timeout; |
893 |
kil $port; |
894 |
$cb->(); |
895 |
}; |
896 |
} else { |
897 |
mon $_[0], sub { |
898 |
kil $port; |
899 |
$cb->(); |
900 |
}; |
901 |
} |
902 |
|
903 |
push @_, $port; |
904 |
&snd; |
905 |
|
906 |
$port |
907 |
} |
908 |
|
909 |
=back |
910 |
|
911 |
=head1 DISTRIBUTED DATABASE |
912 |
|
913 |
AnyEvent::MP comes with a simple distributed database. The database will |
914 |
be mirrored asynchronously on all global nodes. Other nodes bind to one |
915 |
of the global nodes for their needs. Every node has a "local database" |
916 |
which contains all the values that are set locally. All local databases |
917 |
are merged together to form the global database, which can be queried. |
918 |
|
919 |
The database structure is that of a two-level hash - the database hash |
920 |
contains hashes which contain values, similarly to a perl hash of hashes, |
921 |
i.e.: |
922 |
|
923 |
$DATABASE{$family}{$subkey} = $value |
924 |
|
925 |
The top level hash key is called "family", and the second-level hash key |
926 |
is called "subkey" or simply "key". |
927 |
|
928 |
The family must be alphanumeric, i.e. start with a letter and consist |
929 |
of letters, digits, underscores and colons (C<[A-Za-z][A-Za-z0-9_:]*>, |
930 |
pretty much like Perl module names. |
931 |
|
932 |
As the family namespace is global, it is recommended to prefix family names |
933 |
with the name of the application or module using it. |
934 |
|
935 |
The subkeys must be non-empty strings, with no further restrictions. |
936 |
|
937 |
The values should preferably be strings, but other perl scalars should |
938 |
work as well (such as C<undef>, arrays and hashes). |
939 |
|
940 |
Every database entry is owned by one node - adding the same family/subkey |
941 |
combination on multiple nodes will not cause discomfort for AnyEvent::MP, |
942 |
but the result might be nondeterministic, i.e. the key might have |
943 |
different values on different nodes. |
944 |
|
945 |
Different subkeys in the same family can be owned by different nodes |
946 |
without problems, and in fact, this is the common method to create worker |
947 |
pools. For example, a worker port for image scaling might do this: |
948 |
|
949 |
db_set my_image_scalers => $port; |
950 |
|
951 |
And clients looking for an image scaler will want to get the |
952 |
C<my_image_scalers> keys from time to time: |
953 |
|
954 |
db_keys my_image_scalers => sub { |
955 |
@ports = @{ $_[0] }; |
956 |
}; |
957 |
|
958 |
Or better yet, they want to monitor the database family, so they always |
959 |
have a reasonable up-to-date copy: |
960 |
|
961 |
db_mon my_image_scalers => sub { |
962 |
@ports = keys %{ $_[0] }; |
963 |
}; |
964 |
|
965 |
In general, you can set or delete single subkeys, but query and monitor |
966 |
whole families only. |
967 |
|
968 |
If you feel the need to monitor or query a single subkey, try giving it |
969 |
it's own family. |
970 |
|
971 |
=over |
972 |
|
973 |
=item $guard = db_set $family => $subkey [=> $value] |
974 |
|
975 |
Sets (or replaces) a key to the database - if C<$value> is omitted, |
976 |
C<undef> is used instead. |
977 |
|
978 |
When called in non-void context, C<db_set> returns a guard that |
979 |
automatically calls C<db_del> when it is destroyed. |
980 |
|
981 |
=item db_del $family => $subkey... |
982 |
|
983 |
Deletes one or more subkeys from the database family. |
984 |
|
985 |
=item $guard = db_reg $family => $port => $value |
986 |
|
987 |
=item $guard = db_reg $family => $port |
988 |
|
989 |
=item $guard = db_reg $family |
990 |
|
991 |
Registers a port in the given family and optionally returns a guard to |
992 |
remove it. |
993 |
|
994 |
This function basically does the same as: |
995 |
|
996 |
db_set $family => $port => $value |
997 |
|
998 |
Except that the port is monitored and automatically removed from the |
999 |
database family when it is kil'ed. |
1000 |
|
1001 |
If C<$value> is missing, C<undef> is used. If C<$port> is missing, then |
1002 |
C<$SELF> is used. |
1003 |
|
1004 |
This function is most useful to register a port in some port group (which |
1005 |
is just another name for a database family), and have it removed when the |
1006 |
port is gone. This works best when the port is a local port. |
1007 |
|
1008 |
=cut |
1009 |
|
1010 |
sub db_reg($$;$) { |
1011 |
my $family = shift; |
1012 |
my $port = @_ ? shift : $SELF; |
1013 |
|
1014 |
my $clr = sub { db_del $family => $port }; |
1015 |
mon $port, $clr; |
1016 |
|
1017 |
db_set $family => $port => $_[0]; |
1018 |
|
1019 |
defined wantarray |
1020 |
and &Guard::guard ($clr) |
1021 |
} |
1022 |
|
1023 |
=item db_family $family => $cb->(\%familyhash) |
1024 |
|
1025 |
Queries the named database C<$family> and call the callback with the |
1026 |
family represented as a hash. You can keep and freely modify the hash. |
1027 |
|
1028 |
=item db_keys $family => $cb->(\@keys) |
1029 |
|
1030 |
Same as C<db_family>, except it only queries the family I<subkeys> and passes |
1031 |
them as array reference to the callback. |
1032 |
|
1033 |
=item db_values $family => $cb->(\@values) |
1034 |
|
1035 |
Same as C<db_family>, except it only queries the family I<values> and passes them |
1036 |
as array reference to the callback. |
1037 |
|
1038 |
=item $guard = db_mon $family => $cb->($familyhash, \@added, \@changed, \@deleted) |
1039 |
|
1040 |
Creates a monitor on the given database family. Each time a key is set |
1041 |
or or is deleted the callback is called with a hash containing the |
1042 |
database family and three lists of added, changed and deleted subkeys, |
1043 |
respectively. If no keys have changed then the array reference might be |
1044 |
C<undef> or even missing. |
1045 |
|
1046 |
If not called in void context, a guard object is returned that, when |
1047 |
destroyed, stops the monitor. |
1048 |
|
1049 |
The family hash reference and the key arrays belong to AnyEvent::MP and |
1050 |
B<must not be modified or stored> by the callback. When in doubt, make a |
1051 |
copy. |
1052 |
|
1053 |
As soon as possible after the monitoring starts, the callback will be |
1054 |
called with the intiial contents of the family, even if it is empty, |
1055 |
i.e. there will always be a timely call to the callback with the current |
1056 |
contents. |
1057 |
|
1058 |
It is possible that the callback is called with a change event even though |
1059 |
the subkey is already present and the value has not changed. |
1060 |
|
1061 |
The monitoring stops when the guard object is destroyed. |
1062 |
|
1063 |
Example: on every change to the family "mygroup", print out all keys. |
1064 |
|
1065 |
my $guard = db_mon mygroup => sub { |
1066 |
my ($family, $a, $c, $d) = @_; |
1067 |
print "mygroup members: ", (join " ", keys %$family), "\n"; |
1068 |
}; |
1069 |
|
1070 |
Exmaple: wait until the family "My::Module::workers" is non-empty. |
1071 |
|
1072 |
my $guard; $guard = db_mon My::Module::workers => sub { |
1073 |
my ($family, $a, $c, $d) = @_; |
1074 |
return unless %$family; |
1075 |
undef $guard; |
1076 |
print "My::Module::workers now nonempty\n"; |
1077 |
}; |
1078 |
|
1079 |
Example: print all changes to the family "AnyRvent::Fantasy::Module". |
1080 |
|
1081 |
my $guard = db_mon AnyRvent::Fantasy::Module => sub { |
1082 |
my ($family, $a, $c, $d) = @_; |
1083 |
|
1084 |
print "+$_=$family->{$_}\n" for @$a; |
1085 |
print "*$_=$family->{$_}\n" for @$c; |
1086 |
print "-$_=$family->{$_}\n" for @$d; |
1087 |
}; |
1088 |
|
1089 |
=cut |
1090 |
|
1091 |
=back |
1092 |
|
1093 |
=head1 AnyEvent::MP vs. Distributed Erlang |
1094 |
|
1095 |
AnyEvent::MP got lots of its ideas from distributed Erlang (Erlang node |
1096 |
== aemp node, Erlang process == aemp port), so many of the documents and |
1097 |
programming techniques employed by Erlang apply to AnyEvent::MP. Here is a |
1098 |
sample: |
1099 |
|
1100 |
http://www.erlang.se/doc/programming_rules.shtml |
1101 |
http://erlang.org/doc/getting_started/part_frame.html # chapters 3 and 4 |
1102 |
http://erlang.org/download/erlang-book-part1.pdf # chapters 5 and 6 |
1103 |
http://erlang.org/download/armstrong_thesis_2003.pdf # chapters 4 and 5 |
1104 |
|
1105 |
Despite the similarities, there are also some important differences: |
1106 |
|
1107 |
=over 4 |
1108 |
|
1109 |
=item * Node IDs are arbitrary strings in AEMP. |
1110 |
|
1111 |
Erlang relies on special naming and DNS to work everywhere in the same |
1112 |
way. AEMP relies on each node somehow knowing its own address(es) (e.g. by |
1113 |
configuration or DNS), and possibly the addresses of some seed nodes, but |
1114 |
will otherwise discover other nodes (and their IDs) itself. |
1115 |
|
1116 |
=item * Erlang has a "remote ports are like local ports" philosophy, AEMP |
1117 |
uses "local ports are like remote ports". |
1118 |
|
1119 |
The failure modes for local ports are quite different (runtime errors |
1120 |
only) then for remote ports - when a local port dies, you I<know> it dies, |
1121 |
when a connection to another node dies, you know nothing about the other |
1122 |
port. |
1123 |
|
1124 |
Erlang pretends remote ports are as reliable as local ports, even when |
1125 |
they are not. |
1126 |
|
1127 |
AEMP encourages a "treat remote ports differently" philosophy, with local |
1128 |
ports being the special case/exception, where transport errors cannot |
1129 |
occur. |
1130 |
|
1131 |
=item * Erlang uses processes and a mailbox, AEMP does not queue. |
1132 |
|
1133 |
Erlang uses processes that selectively receive messages out of order, and |
1134 |
therefore needs a queue. AEMP is event based, queuing messages would serve |
1135 |
no useful purpose. For the same reason the pattern-matching abilities |
1136 |
of AnyEvent::MP are more limited, as there is little need to be able to |
1137 |
filter messages without dequeuing them. |
1138 |
|
1139 |
This is not a philosophical difference, but simply stems from AnyEvent::MP |
1140 |
being event-based, while Erlang is process-based. |
1141 |
|
1142 |
You cna have a look at L<Coro::MP> for a more Erlang-like process model on |
1143 |
top of AEMP and Coro threads. |
1144 |
|
1145 |
=item * Erlang sends are synchronous, AEMP sends are asynchronous. |
1146 |
|
1147 |
Sending messages in Erlang is synchronous and blocks the process until |
1148 |
a conenction has been established and the message sent (and so does not |
1149 |
need a queue that can overflow). AEMP sends return immediately, connection |
1150 |
establishment is handled in the background. |
1151 |
|
1152 |
=item * Erlang suffers from silent message loss, AEMP does not. |
1153 |
|
1154 |
Erlang implements few guarantees on messages delivery - messages can get |
1155 |
lost without any of the processes realising it (i.e. you send messages a, |
1156 |
b, and c, and the other side only receives messages a and c). |
1157 |
|
1158 |
AEMP guarantees (modulo hardware errors) correct ordering, and the |
1159 |
guarantee that after one message is lost, all following ones sent to the |
1160 |
same port are lost as well, until monitoring raises an error, so there are |
1161 |
no silent "holes" in the message sequence. |
1162 |
|
1163 |
If you want your software to be very reliable, you have to cope with |
1164 |
corrupted and even out-of-order messages in both Erlang and AEMP. AEMP |
1165 |
simply tries to work better in common error cases, such as when a network |
1166 |
link goes down. |
1167 |
|
1168 |
=item * Erlang can send messages to the wrong port, AEMP does not. |
1169 |
|
1170 |
In Erlang it is quite likely that a node that restarts reuses an Erlang |
1171 |
process ID known to other nodes for a completely different process, |
1172 |
causing messages destined for that process to end up in an unrelated |
1173 |
process. |
1174 |
|
1175 |
AEMP does not reuse port IDs, so old messages or old port IDs floating |
1176 |
around in the network will not be sent to an unrelated port. |
1177 |
|
1178 |
=item * Erlang uses unprotected connections, AEMP uses secure |
1179 |
authentication and can use TLS. |
1180 |
|
1181 |
AEMP can use a proven protocol - TLS - to protect connections and |
1182 |
securely authenticate nodes. |
1183 |
|
1184 |
=item * The AEMP protocol is optimised for both text-based and binary |
1185 |
communications. |
1186 |
|
1187 |
The AEMP protocol, unlike the Erlang protocol, supports both programming |
1188 |
language independent text-only protocols (good for debugging), and binary, |
1189 |
language-specific serialisers (e.g. Storable). By default, unless TLS is |
1190 |
used, the protocol is actually completely text-based. |
1191 |
|
1192 |
It has also been carefully designed to be implementable in other languages |
1193 |
with a minimum of work while gracefully degrading functionality to make the |
1194 |
protocol simple. |
1195 |
|
1196 |
=item * AEMP has more flexible monitoring options than Erlang. |
1197 |
|
1198 |
In Erlang, you can chose to receive I<all> exit signals as messages or |
1199 |
I<none>, there is no in-between, so monitoring single Erlang processes is |
1200 |
difficult to implement. |
1201 |
|
1202 |
Monitoring in AEMP is more flexible than in Erlang, as one can choose |
1203 |
between automatic kill, exit message or callback on a per-port basis. |
1204 |
|
1205 |
=item * Erlang tries to hide remote/local connections, AEMP does not. |
1206 |
|
1207 |
Monitoring in Erlang is not an indicator of process death/crashes, in the |
1208 |
same way as linking is (except linking is unreliable in Erlang). |
1209 |
|
1210 |
In AEMP, you don't "look up" registered port names or send to named ports |
1211 |
that might or might not be persistent. Instead, you normally spawn a port |
1212 |
on the remote node. The init function monitors you, and you monitor the |
1213 |
remote port. Since both monitors are local to the node, they are much more |
1214 |
reliable (no need for C<spawn_link>). |
1215 |
|
1216 |
This also saves round-trips and avoids sending messages to the wrong port |
1217 |
(hard to do in Erlang). |
1218 |
|
1219 |
=back |
1220 |
|
1221 |
=head1 RATIONALE |
1222 |
|
1223 |
=over 4 |
1224 |
|
1225 |
=item Why strings for port and node IDs, why not objects? |
1226 |
|
1227 |
We considered "objects", but found that the actual number of methods |
1228 |
that can be called are quite low. Since port and node IDs travel over |
1229 |
the network frequently, the serialising/deserialising would add lots of |
1230 |
overhead, as well as having to keep a proxy object everywhere. |
1231 |
|
1232 |
Strings can easily be printed, easily serialised etc. and need no special |
1233 |
procedures to be "valid". |
1234 |
|
1235 |
And as a result, a port with just a default receiver consists of a single |
1236 |
code reference stored in a global hash - it can't become much cheaper. |
1237 |
|
1238 |
=item Why favour JSON, why not a real serialising format such as Storable? |
1239 |
|
1240 |
In fact, any AnyEvent::MP node will happily accept Storable as framing |
1241 |
format, but currently there is no way to make a node use Storable by |
1242 |
default (although all nodes will accept it). |
1243 |
|
1244 |
The default framing protocol is JSON because a) JSON::XS is many times |
1245 |
faster for small messages and b) most importantly, after years of |
1246 |
experience we found that object serialisation is causing more problems |
1247 |
than it solves: Just like function calls, objects simply do not travel |
1248 |
easily over the network, mostly because they will always be a copy, so you |
1249 |
always have to re-think your design. |
1250 |
|
1251 |
Keeping your messages simple, concentrating on data structures rather than |
1252 |
objects, will keep your messages clean, tidy and efficient. |
1253 |
|
1254 |
=back |
1255 |
|
1256 |
=head1 PORTING FROM AnyEvent::MP VERSION 1.X |
1257 |
|
1258 |
AEMP version 2 has a few major incompatible changes compared to version 1: |
1259 |
|
1260 |
=over 4 |
1261 |
|
1262 |
=item AnyEvent::MP::Global no longer has group management functions. |
1263 |
|
1264 |
At least not officially - the grp_* functions are still exported and might |
1265 |
work, but they will be removed in some later release. |
1266 |
|
1267 |
AnyEvent::MP now comes with a distributed database that is more |
1268 |
powerful. Its database families map closely to port groups, but the API |
1269 |
has changed (the functions are also now exported by AnyEvent::MP). Here is |
1270 |
a rough porting guide: |
1271 |
|
1272 |
grp_reg $group, $port # old |
1273 |
db_reg $group, $port # new |
1274 |
|
1275 |
$list = grp_get $group # old |
1276 |
db_keys $group, sub { my $list = shift } # new |
1277 |
|
1278 |
grp_mon $group, $cb->(\@ports, $add, $del) # old |
1279 |
db_mon $group, $cb->(\%ports, $add, $change, $del) # new |
1280 |
|
1281 |
C<grp_reg> is a no-brainer (just replace by C<db_reg>), but C<grp_get> is |
1282 |
no longer instant, because the local node might not have a copy of the |
1283 |
group. You can either modify your code to allow for a callback, or use |
1284 |
C<db_mon> to keep an updated copy of the group: |
1285 |
|
1286 |
my $local_group_copy; |
1287 |
db_mon $group => sub { $local_group_copy = $_[0] }; |
1288 |
|
1289 |
# now "keys %$local_group_copy" always returns the most up-to-date |
1290 |
# list of ports in the group. |
1291 |
|
1292 |
C<grp_mon> can be replaced by C<db_mon> with minor changes - C<db_mon> |
1293 |
passes a hash as first argument, and an extra C<$chg> argument that can be |
1294 |
ignored: |
1295 |
|
1296 |
db_mon $group => sub { |
1297 |
my ($ports, $add, $chg, $lde) = @_; |
1298 |
$ports = [keys %$ports]; |
1299 |
|
1300 |
# now $ports, $add and $del are the same as |
1301 |
# were originally passed by grp_mon. |
1302 |
... |
1303 |
}; |
1304 |
|
1305 |
=item Nodes not longer connect to all other nodes. |
1306 |
|
1307 |
In AEMP 1.x, every node automatically loads the L<AnyEvent::MP::Global> |
1308 |
module, which in turn would create connections to all other nodes in the |
1309 |
network (helped by the seed nodes). |
1310 |
|
1311 |
In version 2.x, global nodes still connect to all other global nodes, but |
1312 |
other nodes don't - now every node either is a global node itself, or |
1313 |
attaches itself to another global node. |
1314 |
|
1315 |
If a node isn't a global node itself, then it attaches itself to one |
1316 |
of its seed nodes. If that seed node isn't a global node yet, it will |
1317 |
automatically be upgraded to a global node. |
1318 |
|
1319 |
So in many cases, nothing needs to be changed - one just has to make sure |
1320 |
that all seed nodes are meshed together with the other seed nodes (as with |
1321 |
AEMP 1.x), and other nodes specify them as seed nodes. This is most easily |
1322 |
achieved by specifying the same set of seed nodes for all nodes in the |
1323 |
network. |
1324 |
|
1325 |
Not opening a connection to every other node is usually an advantage, |
1326 |
except when you need the lower latency of an already established |
1327 |
connection. To ensure a node establishes a connection to another node, |
1328 |
you can monitor the node port (C<mon $node, ...>), which will attempt to |
1329 |
create the connection (and notify you when the connection fails). |
1330 |
|
1331 |
=item Listener-less nodes (nodes without binds) are gone. |
1332 |
|
1333 |
And are not coming back, at least not in their old form. If no C<binds> |
1334 |
are specified for a node, AnyEvent::MP assumes a default of C<*:*>. |
1335 |
|
1336 |
There are vague plans to implement some form of routing domains, which |
1337 |
might or might not bring back listener-less nodes, but don't count on it. |
1338 |
|
1339 |
The fact that most connections are now optional somewhat mitigates this, |
1340 |
as a node can be effectively unreachable from the outside without any |
1341 |
problems, as long as it isn't a global node and only reaches out to other |
1342 |
nodes (as opposed to being contacted from other nodes). |
1343 |
|
1344 |
=item $AnyEvent::MP::Kernel::WARN has gone. |
1345 |
|
1346 |
AnyEvent has acquired a logging framework (L<AnyEvent::Log>), and AEMP now |
1347 |
uses this, and so should your programs. |
1348 |
|
1349 |
Every module now documents what kinds of messages it generates, with |
1350 |
AnyEvent::MP acting as a catch all. |
1351 |
|
1352 |
On the positive side, this means that instead of setting |
1353 |
C<PERL_ANYEVENT_MP_WARNLEVEL>, you can get away by setting C<AE_VERBOSE> - |
1354 |
much less to type. |
1355 |
|
1356 |
=back |
1357 |
|
1358 |
=head1 LOGGING |
1359 |
|
1360 |
AnyEvent::MP does not normally log anything by itself, but sinc eit is the |
1361 |
root of the contetx hierarchy for AnyEvent::MP modules, it will receive |
1362 |
all log messages by submodules. |
1363 |
|
1364 |
=head1 SEE ALSO |
1365 |
|
1366 |
L<AnyEvent::MP::Intro> - a gentle introduction. |
1367 |
|
1368 |
L<AnyEvent::MP::Kernel> - more, lower-level, stuff. |
1369 |
|
1370 |
L<AnyEvent::MP::Global> - network maintenance and port groups, to find |
1371 |
your applications. |
1372 |
|
1373 |
L<AnyEvent::MP::DataConn> - establish data connections between nodes. |
1374 |
|
1375 |
L<AnyEvent::MP::LogCatcher> - simple service to display log messages from |
1376 |
all nodes. |
1377 |
|
1378 |
L<AnyEvent>. |
1379 |
|
1380 |
=head1 AUTHOR |
1381 |
|
1382 |
Marc Lehmann <schmorp@schmorp.de> |
1383 |
http://home.schmorp.de/ |
1384 |
|
1385 |
=cut |
1386 |
|
1387 |
1 |
1388 |
|