NAME

AnyEvent::MP - multi-processing/message-passing framework

SYNOPSIS

   use AnyEvent::MP;

   $NODE      # contains this node's noderef
   NODE       # returns this node's noderef
   NODE $port # returns the noderef of the port

   $SELF      # receiving/own port id in rcv callbacks

   # initialise the node so it can send/receive messages
   initialise_node;

   # ports are message endpoints

   # sending messages
   snd $port, type => data...;
   snd $port, @msg;
   snd @msg_with_first_element_being_a_port;

   # creating/using ports, the simple way
   my $simple_port = port { my @msg = @_; 0 };

   # creating/using ports, tagged message matching
   my $port = port;
   rcv $port, ping => sub { snd $_[0], "pong"; 0 };
   rcv $port, pong => sub { warn "pong received\n"; 0 };

   # create a port on another node
   my $port = spawn $node, $initfunc, @initdata;

   # monitoring
   mon $port, $cb->(@msg)      # callback is invoked on death
   mon $port, $otherport       # kill otherport on abnormal death
   mon $port, $otherport, @msg # send message on death

CURRENT STATUS

   AnyEvent::MP            - stable API, should work
   AnyEvent::MP::Intro     - outdated
   AnyEvent::MP::Kernel    - mostly stable
   AnyEvent::MP::Global    - mostly stable
   AnyEvent::MP::Node      - mostly stable, but internal anyways
   AnyEvent::MP::Transport - mostly stable, but internal anyways

   stay tuned.

DESCRIPTION

This module (-family) implements a simple message passing framework.

Despite its simplicity, you can securely message other processes running on the same or other hosts, and you can supervise entities remotely.

For an introduction to this module family, see the AnyEvent::MP::Intro manual page and the examples under eg/.

At the moment, this module family is a bit underdocumented.

CONCEPTS

port

A port is something you can send messages to (with the snd function).

Ports allow you to register rcv handlers that can match all or just some messages. Messages send to ports will not be queued, regardless of anything was listening for them or not.

port ID - nodeid#portname

A port ID is the concatenation of a node ID, a hash-mark (#) as separator, and a port name (a printable string of unspecified format).

node

A node is a single process containing at least one port - the node port, which enables nodes to manage each other remotely, and to create new ports.

Nodes are either public (have one or more listening ports) or private (no listening ports). Private nodes cannot talk to other private nodes currently.

node ID - [a-za-Z0-9_\-.:]+

A node ID is a string that uniquely identifies the node within a network. Depending on the configuration used, node IDs can look like a hostname, a hostname and a port, or a random string. AnyEvent::MP itself doesn't interpret node IDs in any way.

binds - ip:port

Nodes can only talk to each other by creating some kind of connection to each other. To do this, nodes should listen on one or more local transport endpoints - binds. Currently, only standard ip:port specifications can be used, which specify TCP ports to listen on.

seeds - host:port

When a node starts, it knows nothing about the network. To teach the node about the network it first has to contact some other node within the network. This node is called a seed.

Seeds are transport endpoint(s) of as many nodes as one wants. Those nodes are expected to be long-running, and at least one of those should always be available. When nodes run out of connections (e.g. due to a network error), they try to re-establish connections to some seednodes again to join the network.

Apart from being sued for seeding, seednodes are not special in any way - every public node can be a seednode.

VARIABLES/FUNCTIONS

$thisnode = NODE / $NODE

The NODE function returns, and the $NODE variable contains, the node ID of the node running in the current process. This value is initialised by a call to initialise_node.

$nodeid = node_of $port

Extracts and returns the node ID from a port ID or a node ID.

initialise_node $profile_name

Before a node can talk to other nodes on the network (i.e. enter "distributed mode") it has to initialise itself - the minimum a node needs to know is its own name, and optionally it should know the addresses of some other nodes in the network to discover other nodes.

This function initialises a node - it must be called exactly once (or never) before calling other AnyEvent::MP functions.

The first argument is a profile name. If it is undef or missing, then the current nodename will be used instead (i.e. uname -n).

The function then looks up the profile in the aemp configuration (see the aemp commandline utility).

If the profile specifies a node ID, then this will become the node ID of this process. If not, then the profile name will be used as node ID. The special node ID of anon/ will be replaced by a random node ID.

The next step is to look up the binds in the profile, followed by binding aemp protocol listeners on all binds specified (it is possible and valid to have no binds, meaning that the node cannot be contacted form the outside. This means the node cannot talk to other nodes that also have no binds, but it can still talk to all "normal" nodes).

If the profile does not specify a binds list, then the node ID will be treated as if it were of the form host:port, which will be resolved and used as binds list.

Lastly, the seeds list from the profile is passed to the AnyEvent::MP::Global module, which will then use it to keep connectivity with at least on of those seed nodes at any point in time.

Example: become a distributed node listening on the guessed noderef, or the one specified via aemp for the current node. This should be the most common form of invocation for "daemon"-type nodes.

   initialise_node;

Example: become an anonymous node. This form is often used for commandline clients.

   initialise_node "anon/";

Example: become a distributed node. If there is no profile of the given name, or no binds list was specified, resolve localhost:4044 and bind on the resulting addresses.

   initialise_node "localhost:4044";

$SELF

Contains the current port id while executing rcv callbacks or psub blocks.

*SELF, SELF, %SELF, @SELF...

Due to some quirks in how perl exports variables, it is impossible to just export $SELF, all the symbols named SELF are exported by this module, but only $SELF is currently used.

snd $port, type => @data

snd $port, @msg

Send the given message to the given port, which can identify either a local or a remote port, and must be a port ID.

While the message can be almost anything, it is highly recommended to use a string as first element (a port ID, or some word that indicates a request type etc.) and to consist if only simple perl values (scalars, arrays, hashes) - if you think you need to pass an object, think again.

The message data logically becomes read-only after a call to this function: modifying any argument (or values referenced by them) is forbidden, as there can be considerable time between the call to snd and the time the message is actually being serialised - in fact, it might never be copied as within the same process it is simply handed to the receiving port.

The type of data you can transfer depends on the transport protocol: when JSON is used, then only strings, numbers and arrays and hashes consisting of those are allowed (no objects). When Storable is used, then anything that Storable can serialise and deserialise is allowed, and for the local node, anything can be passed. Best rely only on the common denominator of these.

$local_port = port

Create a new local port object and returns its port ID. Initially it has no callbacks set and will throw an error when it receives messages.

$local_port = port { my @msg = @_ }

Creates a new local port, and returns its ID. Semantically the same as creating a port and calling rcv $port, $callback on it.

The block will be called for every message received on the port, with the global variable $SELF set to the port ID. Runtime errors will cause the port to be kiled. The message will be passed as-is, no extra argument (i.e. no port ID) will be passed to the callback.

If you want to stop/destroy the port, simply kil it:

   my $port = port {
      my @msg = @_;
      ...
      kil $SELF;
   };

rcv $local_port, $callback->(@msg)

Replaces the default callback on the specified port. There is no way to remove the default callback: use sub { } to disable it, or better kil the port when it is no longer needed.

The global $SELF (exported by this module) contains $port while executing the callback. Runtime errors during callback execution will result in the port being kiled.

The default callback received all messages not matched by a more specific tag match.

rcv $local_port, tag => $callback->(@msg_without_tag), ...

Register (or replace) callbacks to be called on messages starting with the given tag on the given port (and return the port), or unregister it (when $callback is $undef or missing). There can only be one callback registered for each tag.

The original message will be passed to the callback, after the first element (the tag) has been removed. The callback will use the same environment as the default callback (see above).

Example: create a port and bind receivers on it in one go.

  my $port = rcv port,
     msg1 => sub { ... },
     msg2 => sub { ... },
  ;

Example: create a port, bind receivers and send it in a message elsewhere in one go:

   snd $otherport, reply =>
      rcv port,
         msg1 => sub { ... },
         ...
   ;

Example: temporarily register a rcv callback for a tag matching some port (e.g. for a rpc reply) and unregister it after a message was received.

   rcv $port, $otherport => sub {
      my @reply = @_;

      rcv $SELF, $otherport;
   };

$closure = psub { BLOCK }

Remembers $SELF and creates a closure out of the BLOCK. When the closure is executed, sets up the environment in the same way as in rcv callbacks, i.e. runtime errors will cause the port to get kiled.

This is useful when you register callbacks from rcv callbacks:

   rcv delayed_reply => sub {
      my ($delay, @reply) = @_;
      my $timer = AE::timer $delay, 0, psub {
         snd @reply, $SELF;
      };
   };

$guard = mon $port, $cb->(@reason) # call $cb when $port dies

$guard = mon $port, $rcvport # kill $rcvport when $port dies

$guard = mon $port # kill $SELF when $port dies

$guard = mon $port, $rcvport, @msg # send a message when $port dies

Monitor the given port and do something when the port is killed or messages to it were lost, and optionally return a guard that can be used to stop monitoring again.

mon effectively guarantees that, in the absence of hardware failures, after starting the monitor, either all messages sent to the port will arrive, or the monitoring action will be invoked after possible message loss has been detected. No messages will be lost "in between" (after the first lost message no further messages will be received by the port). After the monitoring action was invoked, further messages might get delivered again.

Note that monitoring-actions are one-shot: once messages are lost (and a monitoring alert was raised), they are removed and will not trigger again.

In the first form (callback), the callback is simply called with any number of @reason elements (no @reason means that the port was deleted "normally"). Note also that the callback must never die, so use eval if unsure.

In the second form (another port given), the other port ($rcvport) will be kil'ed with @reason, iff a @reason was specified, i.e. on "normal" kils nothing happens, while under all other conditions, the other port is killed with the same reason.

The third form (kill self) is the same as the second form, except that $rvport defaults to $SELF.

In the last form (message), a message of the form @msg, @reason will be snd.

As a rule of thumb, monitoring requests should always monitor a port from a local port (or callback). The reason is that kill messages might get lost, just like any other message. Another less obvious reason is that even monitoring requests can get lost (for exmaple, when the connection to the other node goes down permanently). When monitoring a port locally these problems do not exist.

Example: call a given callback when $port is killed.

   mon $port, sub { warn "port died because of <@_>\n" };

Example: kill ourselves when $port is killed abnormally.

   mon $port;

Example: send us a restart message when another $port is killed.

   mon $port, $self => "restart";

$guard = mon_guard $port, $ref, $ref...

Monitors the given $port and keeps the passed references. When the port is killed, the references will be freed.

Optionally returns a guard that will stop the monitoring.

This function is useful when you create e.g. timers or other watchers and want to free them when the port gets killed (note the use of psub):

  $port->rcv (start => sub {
     my $timer; $timer = mon_guard $port, AE::timer 1, 1, psub {
        undef $timer if 0.9 < rand;
     });
  });

kil $port[, @reason]

Kill the specified port with the given @reason.

If no @reason is specified, then the port is killed "normally" (ports monitoring other ports will not necessarily die because a port dies "normally").

Otherwise, linked ports get killed with the same reason (second form of mon, see above).

Runtime errors while evaluating rcv callbacks or inside psub blocks will be reported as reason die => $@.

Transport/communication errors are reported as transport_error => $message.

$port = spawn $node, $initfunc[, @initdata]

Creates a port on the node $node (which can also be a port ID, in which case it's the node where that port resides).

The port ID of the newly created port is returned immediately, and it is possible to immediately start sending messages or to monitor the port.

After the port has been created, the init function is called on the remote node, in the same context as a rcv callback. This function must be a fully-qualified function name (e.g. MyApp::Chat::Server::init). To specify a function in the main program, use ::name.

If the function doesn't exist, then the node tries to require the package, then the package above the package and so on (e.g. MyApp::Chat::Server, MyApp::Chat, MyApp) until the function exists or it runs out of package names.

The init function is then called with the newly-created port as context object ($SELF) and the @initdata values as arguments.

A common idiom is to pass a local port, immediately monitor the spawned port, and in the remote init function, immediately monitor the passed local port. This two-way monitoring ensures that both ports get cleaned up when there is a problem.

Example: spawn a chat server port on $othernode.

   # this node, executed from within a port context:
   my $server = spawn $othernode, "MyApp::Chat::Server::connect", $SELF;
   mon $server;

   # init function on C<$othernode>
   sub connect {
      my ($srcport) = @_;

      mon $srcport;

      rcv $SELF, sub {
         ...
      };
   }

after $timeout, @msg

after $timeout, $callback

Either sends the given message, or call the given callback, after the specified number of seconds.

This is simply a utility function that comes in handy at times - the AnyEvent::MP author is not convinced of the wisdom of having it, though, so it may go away in the future.

AnyEvent::MP vs. Distributed Erlang

AnyEvent::MP got lots of its ideas from distributed Erlang (Erlang node == aemp node, Erlang process == aemp port), so many of the documents and programming techniques employed by Erlang apply to AnyEvent::MP. Here is a sample:

   http://www.Erlang.se/doc/programming_rules.shtml
   http://Erlang.org/doc/getting_started/part_frame.html # chapters 3 and 4
   http://Erlang.org/download/Erlang-book-part1.pdf      # chapters 5 and 6
   http://Erlang.org/download/armstrong_thesis_2003.pdf  # chapters 4 and 5

Despite the similarities, there are also some important differences:

• Node IDs are arbitrary strings in AEMP.

  Erlang relies on special naming and DNS to work everywhere in the same way. AEMP relies on each node somehow knowing its own address(es) (e.g. by configuraiton or DNS), but will otherwise discover other odes itself.

• Erlang has a "remote ports are like local ports" philosophy, AEMP uses "local ports are like remote ports".

  The failure modes for local ports are quite different (runtime errors only) then for remote ports - when a local port dies, you know it dies, when a connection to another node dies, you know nothing about the other port.

  Erlang pretends remote ports are as reliable as local ports, even when they are not.

  AEMP encourages a "treat remote ports differently" philosophy, with local ports being the special case/exception, where transport errors cannot occur.

• Erlang uses processes and a mailbox, AEMP does not queue.

  Erlang uses processes that selectively receive messages, and therefore needs a queue. AEMP is event based, queuing messages would serve no useful purpose. For the same reason the pattern-matching abilities of AnyEvent::MP are more limited, as there is little need to be able to filter messages without dequeing them.

  (But see Coro::MP for a more Erlang-like process model on top of AEMP).

• Erlang sends are synchronous, AEMP sends are asynchronous.

  Sending messages in Erlang is synchronous and blocks the process (and so does not need a queue that can overflow). AEMP sends are immediate, connection establishment is handled in the background.

• Erlang suffers from silent message loss, AEMP does not.

  Erlang makes few guarantees on messages delivery - messages can get lost without any of the processes realising it (i.e. you send messages a, b, and c, and the other side only receives messages a and c).

  AEMP guarantees correct ordering, and the guarantee that after one message is lost, all following ones sent to the same port are lost as well, until monitoring raises an error, so there are no silent "holes" in the message sequence.

• Erlang can send messages to the wrong port, AEMP does not.

  In Erlang it is quite likely that a node that restarts reuses a process ID known to other nodes for a completely different process, causing messages destined for that process to end up in an unrelated process.

  AEMP never reuses port IDs, so old messages or old port IDs floating around in the network will not be sent to an unrelated port.

• Erlang uses unprotected connections, AEMP uses secure authentication and can use TLS.

  AEMP can use a proven protocol - TLS - to protect connections and securely authenticate nodes.

• The AEMP protocol is optimised for both text-based and binary communications.

  The AEMP protocol, unlike the Erlang protocol, supports both programming language independent text-only protocols (good for debugging) and binary, language-specific serialisers (e.g. Storable). By default, unless TLS is used, the protocol is actually completely text-based.

  It has also been carefully designed to be implementable in other languages with a minimum of work while gracefully degrading functionality to make the protocol simple.

• AEMP has more flexible monitoring options than Erlang.

  In Erlang, you can chose to receive all exit signals as messages or none, there is no in-between, so monitoring single processes is difficult to implement. Monitoring in AEMP is more flexible than in Erlang, as one can choose between automatic kill, exit message or callback on a per-process basis.

• Erlang tries to hide remote/local connections, AEMP does not.

  Monitoring in Erlang is not an indicator of process death/crashes, in the same way as linking is (except linking is unreliable in Erlang).

  In AEMP, you don't "look up" registered port names or send to named ports that might or might not be persistent. Instead, you normally spawn a port on the remote node. The init function monitors you, and you monitor the remote port. Since both monitors are local to the node, they are much more reliable (no need for spawn_link).

  This also saves round-trips and avoids sending messages to the wrong port (hard to do in Erlang).

RATIONALE

Why strings for port and node IDs, why not objects?

We considered "objects", but found that the actual number of methods that can be called are quite low. Since port and node IDs travel over the network frequently, the serialising/deserialising would add lots of overhead, as well as having to keep a proxy object everywhere.

Strings can easily be printed, easily serialised etc. and need no special procedures to be "valid".

And as a result, a miniport consists of a single closure stored in a global hash - it can't become much cheaper.

Why favour JSON, why not a real serialising format such as Storable?

In fact, any AnyEvent::MP node will happily accept Storable as framing format, but currently there is no way to make a node use Storable by default (although all nodes will accept it).

The default framing protocol is JSON because a) JSON::XS is many times faster for small messages and b) most importantly, after years of experience we found that object serialisation is causing more problems than it solves: Just like function calls, objects simply do not travel easily over the network, mostly because they will always be a copy, so you always have to re-think your design.

Keeping your messages simple, concentrating on data structures rather than objects, will keep your messages clean, tidy and efficient.

SEE ALSO

AnyEvent::MP::Intro - a gentle introduction.

AnyEvent::MP::Kernel - more, lower-level, stuff.

AnyEvent::MP::Global - network maintainance and port groups, to find your applications.

AnyEvent.

AUTHOR

 Marc Lehmann <schmorp@schmorp.de>
 http://home.schmorp.de/