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 $localport, $cb->(@msg) # callback is invoked on death |
41 |
mon $localport, $otherport # kill otherport on abnormal death |
42 |
mon $localport, $otherport, @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 CURRENT STATUS |
53 |
|
54 |
bin/aemp - stable. |
55 |
AnyEvent::MP - stable API, should work. |
56 |
AnyEvent::MP::Intro - explains most concepts. |
57 |
AnyEvent::MP::Kernel - mostly stable API. |
58 |
AnyEvent::MP::Global - stable API. |
59 |
|
60 |
=head1 DESCRIPTION |
61 |
|
62 |
This module (-family) implements a simple message passing framework. |
63 |
|
64 |
Despite its simplicity, you can securely message other processes running |
65 |
on the same or other hosts, and you can supervise entities remotely. |
66 |
|
67 |
For an introduction to this module family, see the L<AnyEvent::MP::Intro> |
68 |
manual page and the examples under F<eg/>. |
69 |
|
70 |
=head1 CONCEPTS |
71 |
|
72 |
=over 4 |
73 |
|
74 |
=item port |
75 |
|
76 |
Not to be confused with a TCP port, a "port" is something you can send |
77 |
messages to (with the C<snd> function). |
78 |
|
79 |
Ports allow you to register C<rcv> handlers that can match all or just |
80 |
some messages. Messages send to ports will not be queued, regardless of |
81 |
anything was listening for them or not. |
82 |
|
83 |
Ports are represented by (printable) strings called "port IDs". |
84 |
|
85 |
=item port ID - C<nodeid#portname> |
86 |
|
87 |
A port ID is the concatenation of a node ID, a hash-mark (C<#>) |
88 |
as separator, and a port name (a printable string of unspecified |
89 |
format created by AnyEvent::MP). |
90 |
|
91 |
=item node |
92 |
|
93 |
A node is a single process containing at least one port - the node port, |
94 |
which enables nodes to manage each other remotely, and to create new |
95 |
ports. |
96 |
|
97 |
Nodes are either public (have one or more listening ports) or private |
98 |
(no listening ports). Private nodes cannot talk to other private nodes |
99 |
currently, but all nodes can talk to public nodes. |
100 |
|
101 |
Nodes is represented by (printable) strings called "node IDs". |
102 |
|
103 |
=item node ID - C<[A-Za-z0-9_\-.:]*> |
104 |
|
105 |
A node ID is a string that uniquely identifies the node within a |
106 |
network. Depending on the configuration used, node IDs can look like a |
107 |
hostname, a hostname and a port, or a random string. AnyEvent::MP itself |
108 |
doesn't interpret node IDs in any way except to uniquely identify a node. |
109 |
|
110 |
=item binds - C<ip:port> |
111 |
|
112 |
Nodes can only talk to each other by creating some kind of connection to |
113 |
each other. To do this, nodes should listen on one or more local transport |
114 |
endpoints - binds. |
115 |
|
116 |
Currently, only standard C<ip:port> specifications can be used, which |
117 |
specify TCP ports to listen on. So a bind is basically just a tcp socket |
118 |
in listening mode thta accepts conenctions form other nodes. |
119 |
|
120 |
=item seed nodes |
121 |
|
122 |
When a node starts, it knows nothing about the network it is in - it |
123 |
needs to connect to at least one other node that is already in the |
124 |
network. These other nodes are called "seed nodes". |
125 |
|
126 |
Seed nodes themselves are not special - they are seed nodes only because |
127 |
some other node I<uses> them as such, but any node can be used as seed |
128 |
node for other nodes, and eahc node cna use a different set of seed nodes. |
129 |
|
130 |
In addition to discovering the network, seed nodes are also used to |
131 |
maintain the network - all nodes using the same seed node form are part of |
132 |
the same network. If a network is split into multiple subnets because e.g. |
133 |
the network link between the parts goes down, then using the same seed |
134 |
nodes for all nodes ensures that eventually the subnets get merged again. |
135 |
|
136 |
Seed nodes are expected to be long-running, and at least one seed node |
137 |
should always be available. They should also be relatively responsive - a |
138 |
seed node that blocks for long periods will slow down everybody else. |
139 |
|
140 |
For small networks, it's best if every node uses the same set of seed |
141 |
nodes. For large networks, it can be useful to specify "regional" seed |
142 |
nodes for most nodes in an area, and use all seed nodes as seed nodes for |
143 |
each other. What's important is that all seed nodes connections form a |
144 |
complete graph, so that the network cannot split into separate subnets |
145 |
forever. |
146 |
|
147 |
Seed nodes are represented by seed IDs. |
148 |
|
149 |
=item seed IDs - C<host:port> |
150 |
|
151 |
Seed IDs are transport endpoint(s) (usually a hostname/IP address and a |
152 |
TCP port) of nodes that should be used as seed nodes. |
153 |
|
154 |
=item global nodes |
155 |
|
156 |
An AEMP network needs a discovery service - nodes need to know how to |
157 |
connect to other nodes they only know by name. In addition, AEMP offers a |
158 |
distributed "group database", which maps group names to a list of strings |
159 |
- for example, to register worker ports. |
160 |
|
161 |
A network needs at least one global node to work, and allows every node to |
162 |
be a global node. |
163 |
|
164 |
Any node that loads the L<AnyEvent::MP::Global> module becomes a global |
165 |
node and tries to keep connections to all other nodes. So while it can |
166 |
make sense to make every node "global" in small networks, it usually makes |
167 |
sense to only make seed nodes into global nodes in large networks (nodes |
168 |
keep connections to seed nodes and global nodes, so makign them the same |
169 |
reduces overhead). |
170 |
|
171 |
=back |
172 |
|
173 |
=head1 VARIABLES/FUNCTIONS |
174 |
|
175 |
=over 4 |
176 |
|
177 |
=cut |
178 |
|
179 |
package AnyEvent::MP; |
180 |
|
181 |
use AnyEvent::MP::Config (); |
182 |
use AnyEvent::MP::Kernel; |
183 |
use AnyEvent::MP::Kernel qw(%NODE %PORT %PORT_DATA $UNIQ $RUNIQ $ID); |
184 |
|
185 |
use common::sense; |
186 |
|
187 |
use Carp (); |
188 |
|
189 |
use AE (); |
190 |
use Guard (); |
191 |
|
192 |
use base "Exporter"; |
193 |
|
194 |
our $VERSION = $AnyEvent::MP::Config::VERSION; |
195 |
|
196 |
our @EXPORT = qw( |
197 |
NODE $NODE *SELF node_of after |
198 |
configure |
199 |
snd rcv mon mon_guard kil psub peval spawn cal |
200 |
port |
201 |
db_set db_del db_reg |
202 |
); |
203 |
|
204 |
our $SELF; |
205 |
|
206 |
sub _self_die() { |
207 |
my $msg = $@; |
208 |
$msg =~ s/\n+$// unless ref $msg; |
209 |
kil $SELF, die => $msg; |
210 |
} |
211 |
|
212 |
=item $thisnode = NODE / $NODE |
213 |
|
214 |
The C<NODE> function returns, and the C<$NODE> variable contains, the node |
215 |
ID of the node running in the current process. This value is initialised by |
216 |
a call to C<configure>. |
217 |
|
218 |
=item $nodeid = node_of $port |
219 |
|
220 |
Extracts and returns the node ID from a port ID or a node ID. |
221 |
|
222 |
=item configure $profile, key => value... |
223 |
|
224 |
=item configure key => value... |
225 |
|
226 |
Before a node can talk to other nodes on the network (i.e. enter |
227 |
"distributed mode") it has to configure itself - the minimum a node needs |
228 |
to know is its own name, and optionally it should know the addresses of |
229 |
some other nodes in the network to discover other nodes. |
230 |
|
231 |
This function configures a node - it must be called exactly once (or |
232 |
never) before calling other AnyEvent::MP functions. |
233 |
|
234 |
The key/value pairs are basically the same ones as documented for the |
235 |
F<aemp> command line utility (sans the set/del prefix), with these additions: |
236 |
|
237 |
=over 4 |
238 |
|
239 |
=item norc => $boolean (default false) |
240 |
|
241 |
If true, then the rc file (e.g. F<~/.perl-anyevent-mp>) will I<not> |
242 |
be consulted - all configuraiton options must be specified in the |
243 |
C<configure> call. |
244 |
|
245 |
=item force => $boolean (default false) |
246 |
|
247 |
IF true, then the values specified in the C<configure> will take |
248 |
precedence over any values configured via the rc file. The default is for |
249 |
the rc file to override any options specified in the program. |
250 |
|
251 |
=item secure => $pass->($nodeid) |
252 |
|
253 |
In addition to specifying a boolean, you can specify a code reference that |
254 |
is called for every remote execution attempt - the execution request is |
255 |
granted iff the callback returns a true value. |
256 |
|
257 |
See F<semp setsecure> for more info. |
258 |
|
259 |
=back |
260 |
|
261 |
=over 4 |
262 |
|
263 |
=item step 1, gathering configuration from profiles |
264 |
|
265 |
The function first looks up a profile in the aemp configuration (see the |
266 |
L<aemp> commandline utility). The profile name can be specified via the |
267 |
named C<profile> parameter or can simply be the first parameter). If it is |
268 |
missing, then the nodename (F<uname -n>) will be used as profile name. |
269 |
|
270 |
The profile data is then gathered as follows: |
271 |
|
272 |
First, all remaining key => value pairs (all of which are conveniently |
273 |
undocumented at the moment) will be interpreted as configuration |
274 |
data. Then they will be overwritten by any values specified in the global |
275 |
default configuration (see the F<aemp> utility), then the chain of |
276 |
profiles chosen by the profile name (and any C<parent> attributes). |
277 |
|
278 |
That means that the values specified in the profile have highest priority |
279 |
and the values specified directly via C<configure> have lowest priority, |
280 |
and can only be used to specify defaults. |
281 |
|
282 |
If the profile specifies a node ID, then this will become the node ID of |
283 |
this process. If not, then the profile name will be used as node ID, with |
284 |
a unique randoms tring (C</%u>) appended. |
285 |
|
286 |
The node ID can contain some C<%> sequences that are expanded: C<%n> |
287 |
is expanded to the local nodename, C<%u> is replaced by a random |
288 |
strign to make the node unique. For example, the F<aemp> commandline |
289 |
utility uses C<aemp/%n/%u> as nodename, which might expand to |
290 |
C<aemp/cerebro/ZQDGSIkRhEZQDGSIkRhE>. |
291 |
|
292 |
=item step 2, bind listener sockets |
293 |
|
294 |
The next step is to look up the binds in the profile, followed by binding |
295 |
aemp protocol listeners on all binds specified (it is possible and valid |
296 |
to have no binds, meaning that the node cannot be contacted form the |
297 |
outside. This means the node cannot talk to other nodes that also have no |
298 |
binds, but it can still talk to all "normal" nodes). |
299 |
|
300 |
If the profile does not specify a binds list, then a default of C<*> is |
301 |
used, meaning the node will bind on a dynamically-assigned port on every |
302 |
local IP address it finds. |
303 |
|
304 |
=item step 3, connect to seed nodes |
305 |
|
306 |
As the last step, the seed ID list from the profile is passed to the |
307 |
L<AnyEvent::MP::Global> module, which will then use it to keep |
308 |
connectivity with at least one node at any point in time. |
309 |
|
310 |
=back |
311 |
|
312 |
Example: become a distributed node using the local node name as profile. |
313 |
This should be the most common form of invocation for "daemon"-type nodes. |
314 |
|
315 |
configure |
316 |
|
317 |
Example: become a semi-anonymous node. This form is often used for |
318 |
commandline clients. |
319 |
|
320 |
configure nodeid => "myscript/%n/%u"; |
321 |
|
322 |
Example: configure a node using a profile called seed, which is suitable |
323 |
for a seed node as it binds on all local addresses on a fixed port (4040, |
324 |
customary for aemp). |
325 |
|
326 |
# use the aemp commandline utility |
327 |
# aemp profile seed binds '*:4040' |
328 |
|
329 |
# then use it |
330 |
configure profile => "seed"; |
331 |
|
332 |
# or simply use aemp from the shell again: |
333 |
# aemp run profile seed |
334 |
|
335 |
# or provide a nicer-to-remember nodeid |
336 |
# aemp run profile seed nodeid "$(hostname)" |
337 |
|
338 |
=item $SELF |
339 |
|
340 |
Contains the current port id while executing C<rcv> callbacks or C<psub> |
341 |
blocks. |
342 |
|
343 |
=item *SELF, SELF, %SELF, @SELF... |
344 |
|
345 |
Due to some quirks in how perl exports variables, it is impossible to |
346 |
just export C<$SELF>, all the symbols named C<SELF> are exported by this |
347 |
module, but only C<$SELF> is currently used. |
348 |
|
349 |
=item snd $port, type => @data |
350 |
|
351 |
=item snd $port, @msg |
352 |
|
353 |
Send the given message to the given port, which can identify either a |
354 |
local or a remote port, and must be a port ID. |
355 |
|
356 |
While the message can be almost anything, it is highly recommended to |
357 |
use a string as first element (a port ID, or some word that indicates a |
358 |
request type etc.) and to consist if only simple perl values (scalars, |
359 |
arrays, hashes) - if you think you need to pass an object, think again. |
360 |
|
361 |
The message data logically becomes read-only after a call to this |
362 |
function: modifying any argument (or values referenced by them) is |
363 |
forbidden, as there can be considerable time between the call to C<snd> |
364 |
and the time the message is actually being serialised - in fact, it might |
365 |
never be copied as within the same process it is simply handed to the |
366 |
receiving port. |
367 |
|
368 |
The type of data you can transfer depends on the transport protocol: when |
369 |
JSON is used, then only strings, numbers and arrays and hashes consisting |
370 |
of those are allowed (no objects). When Storable is used, then anything |
371 |
that Storable can serialise and deserialise is allowed, and for the local |
372 |
node, anything can be passed. Best rely only on the common denominator of |
373 |
these. |
374 |
|
375 |
=item $local_port = port |
376 |
|
377 |
Create a new local port object and returns its port ID. Initially it has |
378 |
no callbacks set and will throw an error when it receives messages. |
379 |
|
380 |
=item $local_port = port { my @msg = @_ } |
381 |
|
382 |
Creates a new local port, and returns its ID. Semantically the same as |
383 |
creating a port and calling C<rcv $port, $callback> on it. |
384 |
|
385 |
The block will be called for every message received on the port, with the |
386 |
global variable C<$SELF> set to the port ID. Runtime errors will cause the |
387 |
port to be C<kil>ed. The message will be passed as-is, no extra argument |
388 |
(i.e. no port ID) will be passed to the callback. |
389 |
|
390 |
If you want to stop/destroy the port, simply C<kil> it: |
391 |
|
392 |
my $port = port { |
393 |
my @msg = @_; |
394 |
... |
395 |
kil $SELF; |
396 |
}; |
397 |
|
398 |
=cut |
399 |
|
400 |
sub rcv($@); |
401 |
|
402 |
sub _kilme { |
403 |
die "received message on port without callback"; |
404 |
} |
405 |
|
406 |
sub port(;&) { |
407 |
my $id = $UNIQ . ++$ID; |
408 |
my $port = "$NODE#$id"; |
409 |
|
410 |
rcv $port, shift || \&_kilme; |
411 |
|
412 |
$port |
413 |
} |
414 |
|
415 |
=item rcv $local_port, $callback->(@msg) |
416 |
|
417 |
Replaces the default callback on the specified port. There is no way to |
418 |
remove the default callback: use C<sub { }> to disable it, or better |
419 |
C<kil> the port when it is no longer needed. |
420 |
|
421 |
The global C<$SELF> (exported by this module) contains C<$port> while |
422 |
executing the callback. Runtime errors during callback execution will |
423 |
result in the port being C<kil>ed. |
424 |
|
425 |
The default callback received all messages not matched by a more specific |
426 |
C<tag> match. |
427 |
|
428 |
=item rcv $local_port, tag => $callback->(@msg_without_tag), ... |
429 |
|
430 |
Register (or replace) callbacks to be called on messages starting with the |
431 |
given tag on the given port (and return the port), or unregister it (when |
432 |
C<$callback> is C<$undef> or missing). There can only be one callback |
433 |
registered for each tag. |
434 |
|
435 |
The original message will be passed to the callback, after the first |
436 |
element (the tag) has been removed. The callback will use the same |
437 |
environment as the default callback (see above). |
438 |
|
439 |
Example: create a port and bind receivers on it in one go. |
440 |
|
441 |
my $port = rcv port, |
442 |
msg1 => sub { ... }, |
443 |
msg2 => sub { ... }, |
444 |
; |
445 |
|
446 |
Example: create a port, bind receivers and send it in a message elsewhere |
447 |
in one go: |
448 |
|
449 |
snd $otherport, reply => |
450 |
rcv port, |
451 |
msg1 => sub { ... }, |
452 |
... |
453 |
; |
454 |
|
455 |
Example: temporarily register a rcv callback for a tag matching some port |
456 |
(e.g. for an rpc reply) and unregister it after a message was received. |
457 |
|
458 |
rcv $port, $otherport => sub { |
459 |
my @reply = @_; |
460 |
|
461 |
rcv $SELF, $otherport; |
462 |
}; |
463 |
|
464 |
=cut |
465 |
|
466 |
sub rcv($@) { |
467 |
my $port = shift; |
468 |
my ($nodeid, $portid) = split /#/, $port, 2; |
469 |
|
470 |
$NODE{$nodeid} == $NODE{""} |
471 |
or Carp::croak "$port: rcv can only be called on local ports, caught"; |
472 |
|
473 |
while (@_) { |
474 |
if (ref $_[0]) { |
475 |
if (my $self = $PORT_DATA{$portid}) { |
476 |
"AnyEvent::MP::Port" eq ref $self |
477 |
or Carp::croak "$port: rcv can only be called on message matching ports, caught"; |
478 |
|
479 |
$self->[0] = shift; |
480 |
} else { |
481 |
my $cb = shift; |
482 |
$PORT{$portid} = sub { |
483 |
local $SELF = $port; |
484 |
eval { &$cb }; _self_die if $@; |
485 |
}; |
486 |
} |
487 |
} elsif (defined $_[0]) { |
488 |
my $self = $PORT_DATA{$portid} ||= do { |
489 |
my $self = bless [$PORT{$portid} || sub { }, { }, $port], "AnyEvent::MP::Port"; |
490 |
|
491 |
$PORT{$portid} = sub { |
492 |
local $SELF = $port; |
493 |
|
494 |
if (my $cb = $self->[1]{$_[0]}) { |
495 |
shift; |
496 |
eval { &$cb }; _self_die if $@; |
497 |
} else { |
498 |
&{ $self->[0] }; |
499 |
} |
500 |
}; |
501 |
|
502 |
$self |
503 |
}; |
504 |
|
505 |
"AnyEvent::MP::Port" eq ref $self |
506 |
or Carp::croak "$port: rcv can only be called on message matching ports, caught"; |
507 |
|
508 |
my ($tag, $cb) = splice @_, 0, 2; |
509 |
|
510 |
if (defined $cb) { |
511 |
$self->[1]{$tag} = $cb; |
512 |
} else { |
513 |
delete $self->[1]{$tag}; |
514 |
} |
515 |
} |
516 |
} |
517 |
|
518 |
$port |
519 |
} |
520 |
|
521 |
=item peval $port, $coderef[, @args] |
522 |
|
523 |
Evaluates the given C<$codref> within the contetx of C<$port>, that is, |
524 |
when the code throews an exception the C<$port> will be killed. |
525 |
|
526 |
Any remaining args will be passed to the callback. Any return values will |
527 |
be returned to the caller. |
528 |
|
529 |
This is useful when you temporarily want to execute code in the context of |
530 |
a port. |
531 |
|
532 |
Example: create a port and run some initialisation code in it's context. |
533 |
|
534 |
my $port = port { ... }; |
535 |
|
536 |
peval $port, sub { |
537 |
init |
538 |
or die "unable to init"; |
539 |
}; |
540 |
|
541 |
=cut |
542 |
|
543 |
sub peval($$) { |
544 |
local $SELF = shift; |
545 |
my $cb = shift; |
546 |
|
547 |
if (wantarray) { |
548 |
my @res = eval { &$cb }; |
549 |
_self_die if $@; |
550 |
@res |
551 |
} else { |
552 |
my $res = eval { &$cb }; |
553 |
_self_die if $@; |
554 |
$res |
555 |
} |
556 |
} |
557 |
|
558 |
=item $closure = psub { BLOCK } |
559 |
|
560 |
Remembers C<$SELF> and creates a closure out of the BLOCK. When the |
561 |
closure is executed, sets up the environment in the same way as in C<rcv> |
562 |
callbacks, i.e. runtime errors will cause the port to get C<kil>ed. |
563 |
|
564 |
The effect is basically as if it returned C<< sub { peval $SELF, sub { |
565 |
BLOCK }, @_ } >>. |
566 |
|
567 |
This is useful when you register callbacks from C<rcv> callbacks: |
568 |
|
569 |
rcv delayed_reply => sub { |
570 |
my ($delay, @reply) = @_; |
571 |
my $timer = AE::timer $delay, 0, psub { |
572 |
snd @reply, $SELF; |
573 |
}; |
574 |
}; |
575 |
|
576 |
=cut |
577 |
|
578 |
sub psub(&) { |
579 |
my $cb = shift; |
580 |
|
581 |
my $port = $SELF |
582 |
or Carp::croak "psub can only be called from within rcv or psub callbacks, not"; |
583 |
|
584 |
sub { |
585 |
local $SELF = $port; |
586 |
|
587 |
if (wantarray) { |
588 |
my @res = eval { &$cb }; |
589 |
_self_die if $@; |
590 |
@res |
591 |
} else { |
592 |
my $res = eval { &$cb }; |
593 |
_self_die if $@; |
594 |
$res |
595 |
} |
596 |
} |
597 |
} |
598 |
|
599 |
=item $guard = mon $port, $cb->(@reason) # call $cb when $port dies |
600 |
|
601 |
=item $guard = mon $port, $rcvport # kill $rcvport when $port dies |
602 |
|
603 |
=item $guard = mon $port # kill $SELF when $port dies |
604 |
|
605 |
=item $guard = mon $port, $rcvport, @msg # send a message when $port dies |
606 |
|
607 |
Monitor the given port and do something when the port is killed or |
608 |
messages to it were lost, and optionally return a guard that can be used |
609 |
to stop monitoring again. |
610 |
|
611 |
In the first form (callback), the callback is simply called with any |
612 |
number of C<@reason> elements (no @reason means that the port was deleted |
613 |
"normally"). Note also that I<< the callback B<must> never die >>, so use |
614 |
C<eval> if unsure. |
615 |
|
616 |
In the second form (another port given), the other port (C<$rcvport>) |
617 |
will be C<kil>'ed with C<@reason>, if a @reason was specified, i.e. on |
618 |
"normal" kils nothing happens, while under all other conditions, the other |
619 |
port is killed with the same reason. |
620 |
|
621 |
The third form (kill self) is the same as the second form, except that |
622 |
C<$rvport> defaults to C<$SELF>. |
623 |
|
624 |
In the last form (message), a message of the form C<@msg, @reason> will be |
625 |
C<snd>. |
626 |
|
627 |
Monitoring-actions are one-shot: once messages are lost (and a monitoring |
628 |
alert was raised), they are removed and will not trigger again. |
629 |
|
630 |
As a rule of thumb, monitoring requests should always monitor a port from |
631 |
a local port (or callback). The reason is that kill messages might get |
632 |
lost, just like any other message. Another less obvious reason is that |
633 |
even monitoring requests can get lost (for example, when the connection |
634 |
to the other node goes down permanently). When monitoring a port locally |
635 |
these problems do not exist. |
636 |
|
637 |
C<mon> effectively guarantees that, in the absence of hardware failures, |
638 |
after starting the monitor, either all messages sent to the port will |
639 |
arrive, or the monitoring action will be invoked after possible message |
640 |
loss has been detected. No messages will be lost "in between" (after |
641 |
the first lost message no further messages will be received by the |
642 |
port). After the monitoring action was invoked, further messages might get |
643 |
delivered again. |
644 |
|
645 |
Inter-host-connection timeouts and monitoring depend on the transport |
646 |
used. The only transport currently implemented is TCP, and AnyEvent::MP |
647 |
relies on TCP to detect node-downs (this can take 10-15 minutes on a |
648 |
non-idle connection, and usually around two hours for idle connections). |
649 |
|
650 |
This means that monitoring is good for program errors and cleaning up |
651 |
stuff eventually, but they are no replacement for a timeout when you need |
652 |
to ensure some maximum latency. |
653 |
|
654 |
Example: call a given callback when C<$port> is killed. |
655 |
|
656 |
mon $port, sub { warn "port died because of <@_>\n" }; |
657 |
|
658 |
Example: kill ourselves when C<$port> is killed abnormally. |
659 |
|
660 |
mon $port; |
661 |
|
662 |
Example: send us a restart message when another C<$port> is killed. |
663 |
|
664 |
mon $port, $self => "restart"; |
665 |
|
666 |
=cut |
667 |
|
668 |
sub mon { |
669 |
my ($nodeid, $port) = split /#/, shift, 2; |
670 |
|
671 |
my $node = $NODE{$nodeid} || add_node $nodeid; |
672 |
|
673 |
my $cb = @_ ? shift : $SELF || Carp::croak 'mon: called with one argument only, but $SELF not set,'; |
674 |
|
675 |
unless (ref $cb) { |
676 |
if (@_) { |
677 |
# send a kill info message |
678 |
my (@msg) = ($cb, @_); |
679 |
$cb = sub { snd @msg, @_ }; |
680 |
} else { |
681 |
# simply kill other port |
682 |
my $port = $cb; |
683 |
$cb = sub { kil $port, @_ if @_ }; |
684 |
} |
685 |
} |
686 |
|
687 |
$node->monitor ($port, $cb); |
688 |
|
689 |
defined wantarray |
690 |
and ($cb += 0, Guard::guard { $node->unmonitor ($port, $cb) }) |
691 |
} |
692 |
|
693 |
=item $guard = mon_guard $port, $ref, $ref... |
694 |
|
695 |
Monitors the given C<$port> and keeps the passed references. When the port |
696 |
is killed, the references will be freed. |
697 |
|
698 |
Optionally returns a guard that will stop the monitoring. |
699 |
|
700 |
This function is useful when you create e.g. timers or other watchers and |
701 |
want to free them when the port gets killed (note the use of C<psub>): |
702 |
|
703 |
$port->rcv (start => sub { |
704 |
my $timer; $timer = mon_guard $port, AE::timer 1, 1, psub { |
705 |
undef $timer if 0.9 < rand; |
706 |
}); |
707 |
}); |
708 |
|
709 |
=cut |
710 |
|
711 |
sub mon_guard { |
712 |
my ($port, @refs) = @_; |
713 |
|
714 |
#TODO: mon-less form? |
715 |
|
716 |
mon $port, sub { 0 && @refs } |
717 |
} |
718 |
|
719 |
=item kil $port[, @reason] |
720 |
|
721 |
Kill the specified port with the given C<@reason>. |
722 |
|
723 |
If no C<@reason> is specified, then the port is killed "normally" - |
724 |
monitor callback will be invoked, but the kil will not cause linked ports |
725 |
(C<mon $mport, $lport> form) to get killed. |
726 |
|
727 |
If a C<@reason> is specified, then linked ports (C<mon $mport, $lport> |
728 |
form) get killed with the same reason. |
729 |
|
730 |
Runtime errors while evaluating C<rcv> callbacks or inside C<psub> blocks |
731 |
will be reported as reason C<< die => $@ >>. |
732 |
|
733 |
Transport/communication errors are reported as C<< transport_error => |
734 |
$message >>. |
735 |
|
736 |
=cut |
737 |
|
738 |
=item $port = spawn $node, $initfunc[, @initdata] |
739 |
|
740 |
Creates a port on the node C<$node> (which can also be a port ID, in which |
741 |
case it's the node where that port resides). |
742 |
|
743 |
The port ID of the newly created port is returned immediately, and it is |
744 |
possible to immediately start sending messages or to monitor the port. |
745 |
|
746 |
After the port has been created, the init function is called on the remote |
747 |
node, in the same context as a C<rcv> callback. This function must be a |
748 |
fully-qualified function name (e.g. C<MyApp::Chat::Server::init>). To |
749 |
specify a function in the main program, use C<::name>. |
750 |
|
751 |
If the function doesn't exist, then the node tries to C<require> |
752 |
the package, then the package above the package and so on (e.g. |
753 |
C<MyApp::Chat::Server>, C<MyApp::Chat>, C<MyApp>) until the function |
754 |
exists or it runs out of package names. |
755 |
|
756 |
The init function is then called with the newly-created port as context |
757 |
object (C<$SELF>) and the C<@initdata> values as arguments. It I<must> |
758 |
call one of the C<rcv> functions to set callbacks on C<$SELF>, otherwise |
759 |
the port might not get created. |
760 |
|
761 |
A common idiom is to pass a local port, immediately monitor the spawned |
762 |
port, and in the remote init function, immediately monitor the passed |
763 |
local port. This two-way monitoring ensures that both ports get cleaned up |
764 |
when there is a problem. |
765 |
|
766 |
C<spawn> guarantees that the C<$initfunc> has no visible effects on the |
767 |
caller before C<spawn> returns (by delaying invocation when spawn is |
768 |
called for the local node). |
769 |
|
770 |
Example: spawn a chat server port on C<$othernode>. |
771 |
|
772 |
# this node, executed from within a port context: |
773 |
my $server = spawn $othernode, "MyApp::Chat::Server::connect", $SELF; |
774 |
mon $server; |
775 |
|
776 |
# init function on C<$othernode> |
777 |
sub connect { |
778 |
my ($srcport) = @_; |
779 |
|
780 |
mon $srcport; |
781 |
|
782 |
rcv $SELF, sub { |
783 |
... |
784 |
}; |
785 |
} |
786 |
|
787 |
=cut |
788 |
|
789 |
sub _spawn { |
790 |
my $port = shift; |
791 |
my $init = shift; |
792 |
|
793 |
# rcv will create the actual port |
794 |
local $SELF = "$NODE#$port"; |
795 |
eval { |
796 |
&{ load_func $init } |
797 |
}; |
798 |
_self_die if $@; |
799 |
} |
800 |
|
801 |
sub spawn(@) { |
802 |
my ($nodeid, undef) = split /#/, shift, 2; |
803 |
|
804 |
my $id = $RUNIQ . ++$ID; |
805 |
|
806 |
$_[0] =~ /::/ |
807 |
or Carp::croak "spawn init function must be a fully-qualified name, caught"; |
808 |
|
809 |
snd_to_func $nodeid, "AnyEvent::MP::_spawn" => $id, @_; |
810 |
|
811 |
"$nodeid#$id" |
812 |
} |
813 |
|
814 |
|
815 |
=item after $timeout, @msg |
816 |
|
817 |
=item after $timeout, $callback |
818 |
|
819 |
Either sends the given message, or call the given callback, after the |
820 |
specified number of seconds. |
821 |
|
822 |
This is simply a utility function that comes in handy at times - the |
823 |
AnyEvent::MP author is not convinced of the wisdom of having it, though, |
824 |
so it may go away in the future. |
825 |
|
826 |
=cut |
827 |
|
828 |
sub after($@) { |
829 |
my ($timeout, @action) = @_; |
830 |
|
831 |
my $t; $t = AE::timer $timeout, 0, sub { |
832 |
undef $t; |
833 |
ref $action[0] |
834 |
? $action[0]() |
835 |
: snd @action; |
836 |
}; |
837 |
} |
838 |
|
839 |
=item cal $port, @msg, $callback[, $timeout] |
840 |
|
841 |
A simple form of RPC - sends a message to the given C<$port> with the |
842 |
given contents (C<@msg>), but adds a reply port to the message. |
843 |
|
844 |
The reply port is created temporarily just for the purpose of receiving |
845 |
the reply, and will be C<kil>ed when no longer needed. |
846 |
|
847 |
A reply message sent to the port is passed to the C<$callback> as-is. |
848 |
|
849 |
If an optional time-out (in seconds) is given and it is not C<undef>, |
850 |
then the callback will be called without any arguments after the time-out |
851 |
elapsed and the port is C<kil>ed. |
852 |
|
853 |
If no time-out is given (or it is C<undef>), then the local port will |
854 |
monitor the remote port instead, so it eventually gets cleaned-up. |
855 |
|
856 |
Currently this function returns the temporary port, but this "feature" |
857 |
might go in future versions unless you can make a convincing case that |
858 |
this is indeed useful for something. |
859 |
|
860 |
=cut |
861 |
|
862 |
sub cal(@) { |
863 |
my $timeout = ref $_[-1] ? undef : pop; |
864 |
my $cb = pop; |
865 |
|
866 |
my $port = port { |
867 |
undef $timeout; |
868 |
kil $SELF; |
869 |
&$cb; |
870 |
}; |
871 |
|
872 |
if (defined $timeout) { |
873 |
$timeout = AE::timer $timeout, 0, sub { |
874 |
undef $timeout; |
875 |
kil $port; |
876 |
$cb->(); |
877 |
}; |
878 |
} else { |
879 |
mon $_[0], sub { |
880 |
kil $port; |
881 |
$cb->(); |
882 |
}; |
883 |
} |
884 |
|
885 |
push @_, $port; |
886 |
&snd; |
887 |
|
888 |
$port |
889 |
} |
890 |
|
891 |
=back |
892 |
|
893 |
=head1 DISTRIBUTED DATABASE |
894 |
|
895 |
AnyEvent::MP comes with a simple distributed database. The database will |
896 |
be mirrored asynchronously at all global nodes. Other nodes bind to one of |
897 |
the global nodes for their needs. |
898 |
|
899 |
The database consists of a two-level hash - a hash contains a hash which |
900 |
contains values. |
901 |
|
902 |
The top level hash key is called "family", and the second-level hash key |
903 |
is called "subkey" or simply "key". |
904 |
|
905 |
The family must be alphanumeric, i.e. start with a letter and consist |
906 |
of letters, digits, underscores and colons (C<[A-Za-z][A-Za-z0-9_:]*>, |
907 |
pretty much like Perl module names. |
908 |
|
909 |
As the family namespace is global, it is recommended to prefix family names |
910 |
with the name of the application or module using it. |
911 |
|
912 |
The subkeys must be non-empty strings, with no further restrictions. |
913 |
|
914 |
The values should preferably be strings, but other perl scalars should |
915 |
work as well (such as undef, arrays and hashes). |
916 |
|
917 |
Every database entry is owned by one node - adding the same family/subkey |
918 |
combination on multiple nodes will not cause discomfort for AnyEvent::MP, |
919 |
but the result might be nondeterministic, i.e. the key might have |
920 |
different values on different nodes. |
921 |
|
922 |
Different subkeys in the same family can be owned by different nodes |
923 |
without problems, and in fact, this is the common method to create worker |
924 |
pools. For example, a worker port for image scaling might do this: |
925 |
|
926 |
db_set my_image_scalers => $port; |
927 |
|
928 |
And clients looking for an image scaler will want to get the |
929 |
C<my_image_scalers> keys: |
930 |
|
931 |
db_keys "my_image_scalers" => 60 => sub { |
932 |
#d##TODO# |
933 |
|
934 |
=over |
935 |
|
936 |
=item db_set $family => $subkey [=> $value] |
937 |
|
938 |
Sets (or replaces) a key to the database - if C<$value> is omitted, |
939 |
C<undef> is used instead. |
940 |
|
941 |
=item db_del $family => $subkey |
942 |
|
943 |
Deletes a key from the database. |
944 |
|
945 |
=item $guard = db_reg $family => $subkey [=> $value] |
946 |
|
947 |
Sets the key on the database and returns a guard. When the guard is |
948 |
destroyed, the key is deleted from the database. If C<$value> is missing, |
949 |
then C<undef> is used. |
950 |
|
951 |
=cut |
952 |
|
953 |
=back |
954 |
|
955 |
=head1 AnyEvent::MP vs. Distributed Erlang |
956 |
|
957 |
AnyEvent::MP got lots of its ideas from distributed Erlang (Erlang node |
958 |
== aemp node, Erlang process == aemp port), so many of the documents and |
959 |
programming techniques employed by Erlang apply to AnyEvent::MP. Here is a |
960 |
sample: |
961 |
|
962 |
http://www.erlang.se/doc/programming_rules.shtml |
963 |
http://erlang.org/doc/getting_started/part_frame.html # chapters 3 and 4 |
964 |
http://erlang.org/download/erlang-book-part1.pdf # chapters 5 and 6 |
965 |
http://erlang.org/download/armstrong_thesis_2003.pdf # chapters 4 and 5 |
966 |
|
967 |
Despite the similarities, there are also some important differences: |
968 |
|
969 |
=over 4 |
970 |
|
971 |
=item * Node IDs are arbitrary strings in AEMP. |
972 |
|
973 |
Erlang relies on special naming and DNS to work everywhere in the same |
974 |
way. AEMP relies on each node somehow knowing its own address(es) (e.g. by |
975 |
configuration or DNS), and possibly the addresses of some seed nodes, but |
976 |
will otherwise discover other nodes (and their IDs) itself. |
977 |
|
978 |
=item * Erlang has a "remote ports are like local ports" philosophy, AEMP |
979 |
uses "local ports are like remote ports". |
980 |
|
981 |
The failure modes for local ports are quite different (runtime errors |
982 |
only) then for remote ports - when a local port dies, you I<know> it dies, |
983 |
when a connection to another node dies, you know nothing about the other |
984 |
port. |
985 |
|
986 |
Erlang pretends remote ports are as reliable as local ports, even when |
987 |
they are not. |
988 |
|
989 |
AEMP encourages a "treat remote ports differently" philosophy, with local |
990 |
ports being the special case/exception, where transport errors cannot |
991 |
occur. |
992 |
|
993 |
=item * Erlang uses processes and a mailbox, AEMP does not queue. |
994 |
|
995 |
Erlang uses processes that selectively receive messages out of order, and |
996 |
therefore needs a queue. AEMP is event based, queuing messages would serve |
997 |
no useful purpose. For the same reason the pattern-matching abilities |
998 |
of AnyEvent::MP are more limited, as there is little need to be able to |
999 |
filter messages without dequeuing them. |
1000 |
|
1001 |
This is not a philosophical difference, but simply stems from AnyEvent::MP |
1002 |
being event-based, while Erlang is process-based. |
1003 |
|
1004 |
You cna have a look at L<Coro::MP> for a more Erlang-like process model on |
1005 |
top of AEMP and Coro threads. |
1006 |
|
1007 |
=item * Erlang sends are synchronous, AEMP sends are asynchronous. |
1008 |
|
1009 |
Sending messages in Erlang is synchronous and blocks the process until |
1010 |
a conenction has been established and the message sent (and so does not |
1011 |
need a queue that can overflow). AEMP sends return immediately, connection |
1012 |
establishment is handled in the background. |
1013 |
|
1014 |
=item * Erlang suffers from silent message loss, AEMP does not. |
1015 |
|
1016 |
Erlang implements few guarantees on messages delivery - messages can get |
1017 |
lost without any of the processes realising it (i.e. you send messages a, |
1018 |
b, and c, and the other side only receives messages a and c). |
1019 |
|
1020 |
AEMP guarantees (modulo hardware errors) correct ordering, and the |
1021 |
guarantee that after one message is lost, all following ones sent to the |
1022 |
same port are lost as well, until monitoring raises an error, so there are |
1023 |
no silent "holes" in the message sequence. |
1024 |
|
1025 |
If you want your software to be very reliable, you have to cope with |
1026 |
corrupted and even out-of-order messages in both Erlang and AEMP. AEMP |
1027 |
simply tries to work better in common error cases, such as when a network |
1028 |
link goes down. |
1029 |
|
1030 |
=item * Erlang can send messages to the wrong port, AEMP does not. |
1031 |
|
1032 |
In Erlang it is quite likely that a node that restarts reuses an Erlang |
1033 |
process ID known to other nodes for a completely different process, |
1034 |
causing messages destined for that process to end up in an unrelated |
1035 |
process. |
1036 |
|
1037 |
AEMP does not reuse port IDs, so old messages or old port IDs floating |
1038 |
around in the network will not be sent to an unrelated port. |
1039 |
|
1040 |
=item * Erlang uses unprotected connections, AEMP uses secure |
1041 |
authentication and can use TLS. |
1042 |
|
1043 |
AEMP can use a proven protocol - TLS - to protect connections and |
1044 |
securely authenticate nodes. |
1045 |
|
1046 |
=item * The AEMP protocol is optimised for both text-based and binary |
1047 |
communications. |
1048 |
|
1049 |
The AEMP protocol, unlike the Erlang protocol, supports both programming |
1050 |
language independent text-only protocols (good for debugging), and binary, |
1051 |
language-specific serialisers (e.g. Storable). By default, unless TLS is |
1052 |
used, the protocol is actually completely text-based. |
1053 |
|
1054 |
It has also been carefully designed to be implementable in other languages |
1055 |
with a minimum of work while gracefully degrading functionality to make the |
1056 |
protocol simple. |
1057 |
|
1058 |
=item * AEMP has more flexible monitoring options than Erlang. |
1059 |
|
1060 |
In Erlang, you can chose to receive I<all> exit signals as messages or |
1061 |
I<none>, there is no in-between, so monitoring single Erlang processes is |
1062 |
difficult to implement. |
1063 |
|
1064 |
Monitoring in AEMP is more flexible than in Erlang, as one can choose |
1065 |
between automatic kill, exit message or callback on a per-port basis. |
1066 |
|
1067 |
=item * Erlang tries to hide remote/local connections, AEMP does not. |
1068 |
|
1069 |
Monitoring in Erlang is not an indicator of process death/crashes, in the |
1070 |
same way as linking is (except linking is unreliable in Erlang). |
1071 |
|
1072 |
In AEMP, you don't "look up" registered port names or send to named ports |
1073 |
that might or might not be persistent. Instead, you normally spawn a port |
1074 |
on the remote node. The init function monitors you, and you monitor the |
1075 |
remote port. Since both monitors are local to the node, they are much more |
1076 |
reliable (no need for C<spawn_link>). |
1077 |
|
1078 |
This also saves round-trips and avoids sending messages to the wrong port |
1079 |
(hard to do in Erlang). |
1080 |
|
1081 |
=back |
1082 |
|
1083 |
=head1 RATIONALE |
1084 |
|
1085 |
=over 4 |
1086 |
|
1087 |
=item Why strings for port and node IDs, why not objects? |
1088 |
|
1089 |
We considered "objects", but found that the actual number of methods |
1090 |
that can be called are quite low. Since port and node IDs travel over |
1091 |
the network frequently, the serialising/deserialising would add lots of |
1092 |
overhead, as well as having to keep a proxy object everywhere. |
1093 |
|
1094 |
Strings can easily be printed, easily serialised etc. and need no special |
1095 |
procedures to be "valid". |
1096 |
|
1097 |
And as a result, a port with just a default receiver consists of a single |
1098 |
code reference stored in a global hash - it can't become much cheaper. |
1099 |
|
1100 |
=item Why favour JSON, why not a real serialising format such as Storable? |
1101 |
|
1102 |
In fact, any AnyEvent::MP node will happily accept Storable as framing |
1103 |
format, but currently there is no way to make a node use Storable by |
1104 |
default (although all nodes will accept it). |
1105 |
|
1106 |
The default framing protocol is JSON because a) JSON::XS is many times |
1107 |
faster for small messages and b) most importantly, after years of |
1108 |
experience we found that object serialisation is causing more problems |
1109 |
than it solves: Just like function calls, objects simply do not travel |
1110 |
easily over the network, mostly because they will always be a copy, so you |
1111 |
always have to re-think your design. |
1112 |
|
1113 |
Keeping your messages simple, concentrating on data structures rather than |
1114 |
objects, will keep your messages clean, tidy and efficient. |
1115 |
|
1116 |
=back |
1117 |
|
1118 |
=head1 SEE ALSO |
1119 |
|
1120 |
L<AnyEvent::MP::Intro> - a gentle introduction. |
1121 |
|
1122 |
L<AnyEvent::MP::Kernel> - more, lower-level, stuff. |
1123 |
|
1124 |
L<AnyEvent::MP::Global> - network maintenance and port groups, to find |
1125 |
your applications. |
1126 |
|
1127 |
L<AnyEvent::MP::DataConn> - establish data connections between nodes. |
1128 |
|
1129 |
L<AnyEvent::MP::LogCatcher> - simple service to display log messages from |
1130 |
all nodes. |
1131 |
|
1132 |
L<AnyEvent>. |
1133 |
|
1134 |
=head1 AUTHOR |
1135 |
|
1136 |
Marc Lehmann <schmorp@schmorp.de> |
1137 |
http://home.schmorp.de/ |
1138 |
|
1139 |
=cut |
1140 |
|
1141 |
1 |
1142 |
|