ZMQ::FFI(3) version agnostic Perl bindings for zeromq using ffi


version 1.11


#### send/recv ####
use v5.10;
my $endpoint = "ipc://zmq-ffi-$$";
my $ctx = ZMQ::FFI->new();
my $s1 = $ctx->socket(ZMQ_REQ);
my $s2 = $ctx->socket(ZMQ_REP);
say $s2->recv();
# ohhai
#### pub/sub ####
use v5.10;
use Time::HiRes q(usleep);
my $endpoint = "ipc://zmq-ffi-$$";
my $ctx = ZMQ::FFI->new();
my $s = $ctx->socket(ZMQ_SUB);
my $p = $ctx->socket(ZMQ_PUB);
# all topics
until ($s->has_pollin) {
# compensate for slow subscriber
usleep 100_000;
say $s->recv();
# ohhai
# specific topics
until ($s->has_pollin) {
usleep 100_000;
$p->send('topic1 ohhai');
$p->send('topic2 ohhai');
while ($s->has_pollin) {
say join ' ', $s->recv();
# topic1 ohhai
# topic2 ohhai
#### multipart ####
use v5.10;
my $endpoint = "ipc://zmq-ffi-$$";
my $ctx = ZMQ::FFI->new();
my $d = $ctx->socket(ZMQ_DEALER);
my $r = $ctx->socket(ZMQ_ROUTER);
$d->send_multipart([qw(ABC DEF GHI)]);
say join ' ', $r->recv_multipart;
# dealer ABC DEF GHI
#### nonblocking ####
use v5.10;
use AnyEvent;
use EV;
my $endpoint = "ipc://zmq-ffi-$$";
my $ctx = ZMQ::FFI->new();
my @messages = qw(foo bar baz);
my $pull = $ctx->socket(ZMQ_PULL);
my $fd = $pull->get_fd();
my $recv = 0;
my $w = AE::io $fd, 0, sub {
while ( $pull->has_pollin ) {
say $pull->recv();
# foo, bar, baz
if ($recv == 3) {
my $push = $ctx->socket(ZMQ_PUSH);
my $sent = 0;
my $t;
$t = AE::timer 0, .1, sub {
if ($sent == 3) {
undef $t;
#### specifying versions ####
use ZMQ::FFI;
# 2.x context
my $ctx = ZMQ::FFI->new( soname => '' );
my ($major, $minor, $patch) = $ctx->version;
# 3.x context
my $ctx = ZMQ::FFI->new( soname => '' );
my ($major, $minor, $patch) = $ctx->version;


ZMQ::FFI exposes a high level, transparent, OO interface to zeromq independent of the underlying libzmq version. Where semantics differ, it will dispatch to the appropriate backend for you. As it uses ffi, there is no dependency on XS or compilation.

As of 1.00 ZMQ::FFI is implemented using FFI::Platypus. This version has substantial performance improvements and you are encouraged to use 1.00 or newer.



    my $ctx = ZMQ::FFI->new(%options);

returns a new context object, appropriate for the version of libzmq found on your system. It accepts the following optional attributes:


zeromq thread pool size. Default: 1
requires zmq >= 3.x

max number of sockets allowed for context. Default: 1024

    ZMQ::FFI->new( soname => '/path/to/' );
    ZMQ::FFI->new( soname => '' );

specify the libzmq library name to load. By default ZMQ::FFI will first try the generic soname for the system, then the soname for each version of zeromq (e.g. "soname" can also be the path to a particular libzmq so file

It is technically possible to have multiple contexts of different versions in the same process, though the utility of doing such a thing is dubious


    my ($major, $minor, $patch) = $ctx->version();

return the libzmq version as the list "($major, $minor, $patch)"


requires zmq >= 3.x

    my $threads = $ctx->get(ZMQ_IO_THREADS)

get a context option value


requires zmq >= 3.x

    $ctx->set(ZMQ_MAX_SOCKETS, 42)

set a context option value


    my $socket = $ctx->socket(ZMQ_REQ)

returns a socket of the specified type. See ``SOCKET API'' below


    $ctx->proxy($frontend, $backend);
    $ctx->proxy($frontend, $backend, $capture);

sets up and runs a "zmq_proxy". For zmq 2.x this will use a "ZMQ_STREAMER" device to simulate the proxy. The optional $capture is only supported for zmq >= 3.x however


zmq 2.x only

    $ctx->device($type, $frontend, $backend);

sets up and runs a "zmq_device" with specified frontend and backend sockets


destroy the underlying zmq context. In general you shouldn't have to call this directly as it is called automatically for you when the object gets reaped

See ``CLEANUP'' below


The following API is available on socket objects created by "$ctx->socket".

For core attributes and functions, common across all versions of zeromq, convenience methods are provided. Otherwise, generic get/set methods are provided that will work independent of version.

As attributes are constantly being added/removed from zeromq, it is unlikely the 'static' accessors will grow much beyond the current set.


    my ($major, $minor, $patch) = $socket->version();

same as Context "version" above



does socket connect on the specified endpoint


requires zmq >= 3.x


does socket disconnect on the specified endpoint



does socket bind on the specified endpoint


requires zmq >= 3.x


does socket unbind on the specified endpoint

get_linger, set_linger

    my $linger = $socket->get_linger();

get or set the socket linger period. Default: 0 (no linger)

See ``CLEANUP'' below

get_identity, set_identity

    my $ident = $socket->get_identity();

get or set the socket identity for request/reply patterns


    my $fd = $socket->get_fd();

get the file descriptor associated with the socket


    my $option_value = $socket->get($option_name, $option_type);
    my $linger = $socket->get(ZMQ_LINGER, 'int');

generic method to get the value for any socket option. $option_type is the type associated with $option_value in the zeromq API ("zmq_getsockopt" man page)


    $socket->set($option_name, $option_type, $option_value);
    $socket->set(ZMQ_IDENTITY, 'binary', 'foo');

generic method to set the value for any socket option. $option_type is the type associated with $option_value in the zeromq API ("zmq_setsockopt" man page)



add $topic to the subscription list



remove $topic from the subscription list


    $socket->send($msg, $flags);

sends a message using the optional flags


    $socket->send($parts_aref, $flags);

given an array ref of message parts, sends the multipart message using the optional flags. ZMQ_SNDMORE semantics are handled for you


    my $msg = $socket->recv();
    my $msg = $socket->recv($flags);

receives a message using the optional flags


    my @parts = $socket->recv_multipart();
    my @parts = $socket->recv_multipart($flags);

receives a multipart message, returning an array of parts. ZMQ_RCVMORE semantics are handled for you

has_pollin, has_pollout

    while ( $socket->has_pollin ) { ... }

checks ZMQ_EVENTS for ZMQ_POLLIN and ZMQ_POLLOUT respectively, and returns true/false depending on the state


close the underlying zmq socket. In general you shouldn't have to call this directly as it is called automatically for you when the object gets reaped

See ``CLEANUP'' below



controls whether error handling should be exceptional or not. This is set to true by default. See ``ERROR HANDLING'' below


returns true or false depending on whether the last socket operation had an error. This is really just an alias for "last_errno"


returns the system "errno" set by the last socket operation, or 0 if there was no error


returns the human readable system error message associated with the socket "last_errno"


With respect to cleanup "ZMQ::FFI" follows either the zeromq guide <> recommendations or the behavior of other zmq bindings. That is:
  • it uses 0 linger by default (this is the default used by czmq <> and jzmq <>)
  • during object destruction it will call close/destroy for you
  • it arranges the reference hierarchy such that sockets will be properly
          cleaned up before their associated contexts
  • it detects fork/thread situations and ensures sockets/contexts are only
          cleaned up in their originating process/thread
  • it guards against double closes/destroys

Given the above you're probably better off letting "ZMQ::FFI" handle cleanup for you. But if for some reason you want to do explicit cleanup yourself you can. All the below will accomplish the same thing:

    # implicit cleanup
        my $context = ZMQ::FFI->new();
        my $socket  = $ctx->socket($type);
        # close/destroy called in destructors at end of scope
    # explicit cleanup
    # ditto
    undef $socket;
    undef $context;

Regarding "linger", you can always set this to a value you prefer if you don't like the default. Once set the new value will be used when the socket is subsequently closed (either implicitly or explicitly):

    $socket->set_linger(-1); # infinite linger
                             # $context->destroy will block forever
                             # (or until all pending messages have been sent)


By default, ZMQ::FFI checks the return codes of underlying zmq functions for you, and in the case of an error it will die with the human readable system error message.

    # dies with 'zmq_socket: Invalid argument'

Usually this is what you want, but not always. Some zmq operations can return errors that are not fatal and should be handled. For example using "ZMQ_DONTWAIT" with send/recv can return "EAGAIN" and simply means try again, not die.

For situations such as this you can turn off exceptional error handling by setting "die_on_error" to 0. It is then for you to check and manage any zmq errors by checking "last_errno":

    use Errno qw(EAGAIN);
    my $ctx = ZMQ::FFI->new();
    my $s   = $ctx->socket(ZMQ_DEALER);
    $s->die_on_error(0); # turn off exceptional error handling
    while (1) {
        my $msg = $s->recv(ZMQ_DONTWAIT);
        if ($s->last_errno == EAGAIN) {
            sleep 1;
        elsif ($s->last_errno) {
            die $s->last_strerror;
        else {
            warn "recvd: $msg";
    $s->die_on_error(1); # turn back on exceptional error handling


ZMQ::FFI uses FFI::Platypus on the backend. In addition to a friendly, usable interface, FFI::Platypus's killer feature is "attach". "attach" makes it possible to bind ffi functions in memory as first class Perl xsubs. This results in dramatic performance gains and gives you the flexibility of ffi with performance approaching that of XS.

Testing indicates FFI::Platypus xsubs are around 30% slower than ``real'' XS xsubs. That may sound like a lot, but to put it in perspective that means, for zeromq, the XS bindings can send 10 million messages 1-2 seconds faster than the ffi ones.

If you really care about 1-2 seconds over 10 million messages you should be writing your solution in C anyways. An equivalent C implementation will be several hundred percent faster or more.

Keep in mind also that the small speed bump you get using XS can easily be wiped out by crappy and poorly optimized Perl code.

Now that Perl finally has a great ffi interface, it is hard to make the case to continue using XS. The slight speed bump just isn't worth giving up the convenience, flexibility, and portability of ffi.

You can find the detailed performance results that informed this section at: <>


"ZMQ::FFI" is free as in beer in addition to being free as in speech. While I've done my best to ensure it's tasty, high quality beer, it probably isn't perfect. If you encounter problems, or otherwise see room for improvement, please open an issue (or even better a pull request!) on github <>


Dylan Cali <[email protected]>


This software is copyright (c) 2016 by Dylan Cali.

This is free software; you can redistribute it and/or modify it under the same terms as the Perl 5 programming language system itself.