Awesome
enm
is an
Erlang port driver
that wraps the nanomsg C library, allowing Erlang
systems to communicate with other nanomsg endpoints. enm
supports
idioms and approaches common to standard Erlang networking facilities such
as gen_tcp
and gen_udp
.
enm
is currently based on version 1.0.0 of nanomsg, and enm
itself
is new, so its features are experimental and subject to change.
Starting and Stopping
You can start enm
as a normal application, using
application:start(enm)
and application:stop(enm)
. You can also call
enm:start_link/0
or enm:start/0
, and call enm:stop/0
to stop it.
Just Open a Socket
enm
supports all nanomsg scalability protocols and transports. You can
open a socket providing a particular scalability protocol using functions
named for each protocol. For example, the enm:pair/0
function opens a
pair-type socket for one-to-one communication, and the enm:req/0
and
enm:rep/0
functions open the request and reply ends, respectively, of the
reqrep
scalability protocol. The arity 0 versions of the enm
scalability protocol functions listed below use default settings for the
open sockets, while the arity 1 versions allow a list of socket
options to be passed in order to control socket settings.
req
: open the request end of thereqrep
protocolrep
: open the reply end of thereqrep
protocolpair
: open a pair socket for one-to-one communicationsbus
: open a bus socket for many-to-many communicationspub
: open the publication end of thepubsub
protocolsub
: open the subscriber end of thepubsub
protocolpush
: open the pushing end of thepipeline
protocolpull
: open the pulling end of thepipeline
protocolsurveyor
: open the query end of thesurvey
protocolrespondent
: open the response end of thesurvey
protocol
If successful, these functions — both their arity 0 and arity 1
versions — all return {ok,Socket}
.
Once opened, sockets can be bound or connected using the enm:bind/2
or
enm:connect/2
functions respectively. Bind and connect information can
alternatively be provided via socket options when
sockets are first opened via the functions listed above.
Functions
In addition to the scalability protocol functions,
enm
supports the following functions:
send(Socket, Data)
: sendData
onSocket
.Data
is an Erlangiolist
, thus allowing lists of binaries and characters, or nested lists thereof, to be sent.recv(Socket)
: receive data fromSocket
. This function blocks indefinitely until data arrive. Returns{ok,Data}
on success, or an error tuple on failure.Data
defaults to a binary unless the socket was opened inlist
mode or was set intolist
mode viasetopts/2
.recv(Socket, Timeout)
: same asrecv/1
but if no data arrive withinTimeout
milliseconds, return{error,etimedout}
.bind(Socket, Address)
: bindSocket
toAddress
, whereAddress
supports one of the nanomsg transport types: inproc, ipc, or TCP.Address
can be either a string or binary using the nanomsg URL address format, such as"inproc://foo"
to bind to an intraprocess address or"tcp://*:12345"
to listen on all your host's network interfaces on port 12345, or it can be one of theenm
address record types.getopts(Socket, Options)
: return the current setting onSocket
for each of the option names listed inOptions
. If successful, returns{ok, OptionList}
where each element of the list provides the name and setting of one of the requested options. Ifgetopts
fails it return an error tuple.setopts(Socket, OptionList)
: apply each of the option settings inOptionList
toSocket
. Returnsok
if successful or an error tuple on failure.controlling_process(Socket, Pid)
: the current controlling process forSocket
can call this function to transfer its control to the process represented byPid
. The controlling process of aSocket
is initially the one that opens it, and it's the one that receives data messages as Erlang messages if the socket is in an active mode.shutdown(Socket, EndpointId)
: removes the endpoint associated withEndpointId
, created viabind/2
orconnect/2
, fromSocket
.close(Socket)
: closesSocket
.
If you're already familiar with standard Erlang networking capabilities,
you'll find these functions similar to functions supplied by standard
modules such as gen_tcp
, gen_udp
and inet
.
Address Record Types
To help avoid errors with mistyped string and binary address URLs, enm
provides three record types you can use for addresses instead:
#nn_inproc{addr=Address}
: for intra-process addresses.Address
is a name in either string or binary form.#nn_ipc{path=Path}
: for IPC addresses.Path
can be either an absolute pathname or a pathname relative to the current working directory, in either string or binary form.#nn_tcp{interface=Interface, addr=Address, port=Port}
: for TCP addresses.Interface
can be the atomany
, a network address in string or tuple form, or a string representing a network interface name.Address
can be a hostname or a network address in either string or binary form.Port
is a port number.
Using these types, which is completely optional, requires including the
enm.hrl
file.
Socket Options
enm
supports several socket options that can be set either when the
socket is opened, or modified later during operation. Most socket options
can also be read from enm
sockets. enm
supports the following options:
type
: indicates the type of socket. For example, thennreq
type indicates a socket opened via thereq
function, andnnsurveyor
indicates a socket implementing the query end of thesurvey
protocol. This option can only be read from anenm
socket and cannot be set.active
: this controls how messages are delivered from anenm
socket to its controlling Erlang process.- The default setting,
{active,true}
, means that the driver reads data from the socket as soon as they arrive and sends them as Erlang messages to the controlling process. - The
{active,false}
setting puts anenm
socket in passive mode; data from such a socket are retrieved only via theenm:recv/{1,2}
functions. - The
{active,once}
setting allows the driver to deliver one message from the socket to the controlling process, after which the socket flips automatically to{active,false}
mode. This allows the application to receive nanomsg messages as Erlang messages only when it's ready to handle them. - The
{active,N}
mode, whereN
represents an integer, is similar to{active,once}
mode except that it allows the driver to receiveN
messages on the socket and deliver them as Erlang messages to the controlling process before flipping the socket into{active,false}
mode. When the socket flips to passive mode,enm
sends a{X_passive,Socket}
message to the controlling process, with the socket's actual type name substituted for "X" (for example,{nnpair_passive, Socket}
ifSocket
is apair
socket).
- The default setting,
raw
: this option, which defaults to false, controls whether the underlying nanomsg socket is opened with theAF_SP
domain (the default, or set via{raw,false}
) or theAF_SP_RAW
domain (set via{raw,true}
). Using the atomraw
by itself is equivalent to{raw,true}
. See the nanomsg nn_socket man page for more details on theAF_SP
andAF_SP_RAW
socket domains.mode
: this controls the form of the data delivered or retrieved from the socket. The default,binary
, means that data from the socket are delivered to the application as Erlang binaries, whereas thelist
setting means socket data are delivered as Erlang lists. Using the atombinary
by itself is equivalent to{mode,binary}
, andlist
by itself is equivalent to{mode,list}
.bind
: this option allows you to open a socket and then immediately bind it to the given address. See thebind
function description for more details on the allowable forms for the bind address. Note, however, that the bind endpoint identifier is thrown away in this case. If you need to later manage the endpoint viashutdown
, use thebind
function instead.connect
: this option allows you open a socket and then immediately connect it to the given address. See theconnect
function description for more details on the allowable forms for the connect address. Note, however, that the connect endpoint identifier is thrown away in this case. If you need to later manage the endpoint viashutdown
, use theconnect
function instead.deadline
: forsurveyor
sockets, set the surveyor deadline to specify how long, in milliseconds, to wait for responses to arrive.subscribe
: forsub
sockets, subscribe to the named topic, specified either as a string or a binary. Topic names must be less than 256 characters in length (this is anenm
limit, not a nanomsg limit). Applying thesubscribe
option to a socket type other thansub
results in abadarg
exception.unsubscribe
: forsub
sockets, unsubscribe from the named topic, specified either as a string or a binary. As for thesubscribe
option, topic names must be less than 256 characters in length. Applying theunsubscribe
option to a socket type other thansub
results in abadarg
exception.resend_ivl
: forreq
sockets, set the request resend interval to specify how long, in milliseconds, to wait for a reply before resending the request. The default is 60000. Applying theresend_ivl
option to a socket type other thanreq
results in abadarg
exception.sndbuf
: set the send buffer size to the specified number of bytes. Applying this option to a socket that doesn't allow sending, specifically apull
orsub
socket, results in abadarg
exception.rcvbuf
: set the receive buffer size to the specified number of bytes. Applying this option to a socket that doesn't allow receiving, specifically apush
orpub
socket, results in abadarg
exception.nodelay
: if true, set theTCP_NODELAY
option on TCP sockets, or if false, clear it.reconnect_ivl
: set the reconnect interval to specify how long, in milliseconds, to wait before attempting to reconnect a broken socket connection. The supplied value must be greater than 0, otherwise abadarg
exception results. The default is 100.reconnect_ivl_max
: set the maximum reconnect interval in milliseconds. If this value is greater than the default of 0, socket reconnection attempts will use exponential backoff starting with the socket'sreconnect_ivl
value and doubling it on each reconnection attempt, but will ensure the backoff value never exceeds the specifiedreconnect_ivl_max
value. With the default value of 0, no exponential backoff is used, and only thereconnect_ivl
setting controls reconnection wait time.
Currently, most but not all nanomsg socket options are implemented. Please file an issue or submit a pull request if an option you need is missing.
Examples
These following examples are based on
Tim Dysinger's C examples,
but they produce somewhat different output. They are all run with inproc
addresses, thereby taking advantage of Erlang's lightweight processes
rather than using separate OS processes as for Tim's examples (though we
could easily do that with Erlang too).
Note also that each example explicitly starts and stops enm
—
this is for exposition only, and is not something you'd do explicitly in an
actual Erlang application. The output shown comes from an interactive
Erlang shell, and it assumes enm
beam files are on the shell's load
path.
You can find the code for these examples in the repository examples
directory.
Pipeline
-module(pipeline).
-export([start/0]).
start() ->
enm:start_link(),
Url = "inproc://pipeline",
{ok,Pull} = enm:pull([{bind,Url},list]),
{ok,Push} = enm:push([{connect,Url},list]),
Send1 = "Hello, World!",
io:format("pushing message \"~s\"~n", [Send1]),
ok = enm:send(Push, Send1),
receive
{nnpull,Pull,Send1} ->
io:format("pulling message \"~s\"~n", [Send1])
end,
Send2 = "Goodbye.",
io:format("pushing message \"~s\"~n", [Send2]),
ok = enm:send(Push, Send2),
receive
{nnpull,Pull,Send2} ->
io:format("pulling message \"~s\"~n", [Send2])
end,
enm:close(Push),
enm:close(Pull),
enm:stop().
Here, note the pattern matching in the receive
statements where we use
the data variables set for the sent messages as the data to be expected to
be received. We put each socket into list
mode to ensure these pattern
matches succeed, given that Send1
and Send2
are strings. Note also that
both the type of the socket and the socket itself are part of the received
messages, allowing us to use matching to easily distinguish between what
each socket is receiving. If these expected patterns did not match what was
being sent, the receive
statements would wait forever.
Pipeline Results
1> c("examples/pipeline.erl", [{o,"examples"}]).
{ok,pipeline}
2> pipeline:start().
pushing message "Hello, World!"
pulling message "Hello, World!"
pushing message "Goodbye."
pulling message "Goodbye."
ok
Request/Reply
-module(request_reply).
-export([start/0]).
start() ->
enm:start_link(),
Url = "inproc://request_reply",
{ok,Rep} = enm:rep([{bind,Url}]),
{ok,Req} = enm:req([{connect,Url}]),
DateReq = <<"DATE">>,
io:format("sending date request~n"),
ok = enm:send(Req, DateReq),
receive
{nnrep,Rep,DateReq} ->
io:format("received date request~n"),
Now = httpd_util:rfc1123_date(),
io:format("sending date ~s~n", [Now]),
ok = enm:send(Rep, Now)
end,
receive
{nnreq,Req,Date} ->
io:format("received date ~s~n", [Date])
end,
enm:close(Req),
enm:close(Rep),
enm:stop().
This is similar to the pipeline example except that data flows in both directions, and both sockets default to binary mode.
Request/Reply Results
1> c("examples/request_reply.erl", [{o,"examples"}]).
{ok,request_reply}
2> request_reply:start().
sending date request
received date request
sending date Tue, 09 Sep 2014 23:05:26 GMT
received date Tue, 09 Sep 2014 23:05:26 GMT
ok
Pair
-module(pair).
-export([start/0, node/4]).
start() ->
enm:start_link(),
Self = self(),
Url = "inproc://pair",
spawn(?MODULE, node, [Self, Url, bind, "Node0"]),
spawn(?MODULE, node, [Self, Url, connect, "Node1"]),
collect(["Node0","Node1"]).
node(Parent, Url, F, Name) ->
{ok,P} = enm:pair([{active,3}]),
{ok,Id} = enm:F(P,Url),
send_recv(P, Name),
enm:shutdown(P, Id),
Parent ! {done,Name}.
send_recv(Sock, Name) ->
receive
{_,Sock,Buf} ->
io:format("~s received \"~s\"~n", [Name, Buf])
after
100 ->
ok
end,
case enm:getopts(Sock, [active]) of
{ok, [{active,false}]} ->
ok;
{error, Error} ->
error(Error);
_ ->
timer:sleep(1000),
io:format("~s sending \"~s\"~n", [Name, Name]),
ok = enm:send(Sock, Name),
send_recv(Sock, Name)
end.
collect([]) ->
ok;
collect([Name|Names]) ->
receive
{done,Name} ->
collect(Names)
end.
This code is a little more involved than previous examples because we spawn
two child processes that receive and send messages. Note how we use the
{active,N}
socket mode for each end of the pair to eventually break out
of the recursive send_recv/2
function, by using enm:getopts/2
to
check for when each socket flips into {active,false}
mode.
Pair Results
1> c("examples/pair.erl",[{o,"examples"}]).
{ok,pair}
2> pair:start().
Node0 sending "Node0"
Node1 sending "Node1"
Node0 received "Node1"
Node1 received "Node0"
Node1 sending "Node1"
Node0 sending "Node0"
Node0 received "Node1"
Node1 received "Node0"
Node1 sending "Node1"
Node0 sending "Node0"
Node0 received "Node1"
Node1 received "Node0"
ok
Pub/Sub
-module(pubsub).
-export([start/0]).
-define(COUNT, 3).
start() ->
enm:start_link(),
Url = "inproc://pubsub",
Pub = pub(Url),
collect(subs(Url, self())),
enm:close(Pub),
enm:stop().
pub(Url) ->
{ok,Pub} = enm:pub([{bind,Url}]),
spawn_link(fun() -> pub(Pub, ?COUNT) end),
Pub.
pub(_, 0) ->
ok;
pub(Pub, Count) ->
Now = httpd_util:rfc1123_date(),
io:format("publishing date \"~s\"~n", [Now]),
ok = enm:send(Pub, ["DATE: ", Now]),
timer:sleep(1000),
pub(Pub, Count-1).
subs(Url, Parent) ->
subs(Url, Parent, ?COUNT, []).
subs(_, _, 0, Acc) ->
Acc;
subs(Url, Parent, Count, Acc) ->
{ok, Sub} = enm:sub([{connect,Url},{subscribe,"DATE:"},{active,false}]),
Name = "Subscriber" ++ integer_to_list(Count),
spawn_link(fun() -> sub(Sub, Parent, Name) end),
subs(Url, Parent, Count-1, [Name|Acc]).
sub(Sub, Parent, Name) ->
case enm:recv(Sub, 2000) of
{ok,Data} ->
io:format("~s received \"~s\"~n", [Name, Data]),
sub(Sub, Parent, Name);
{error,etimedout} ->
enm:close(Sub),
Parent ! {done, Name},
ok
end.
collect([Sub|Subs]) ->
receive
{done,Sub} ->
collect(Subs)
end;
collect([]) ->
ok.
This code sets up a publisher and 3 subscribers, and the publisher
publishes dates to the subscribers. It includes the text "DATE:" in each
message, and messages containing that text are what the subscribers are
looking to receive. Note the use of {active,false}
mode for the
subscriber sockets; this is done because the Erlang process that creates
the sockets, known as the controlling process for the socket, is not the
same process that receives the messages. Only the controlling process can
receive messages in an active mode from a socket.
Pub/Sub Results
1> c("examples/pubsub.erl", [{o,"examples"}]).
{ok,pubsub}
2> pubsub:start().
publishing date "Tue, 09 Sep 2014 23:08:10 GMT"
Subscriber3 received "DATE: Tue, 09 Sep 2014 23:08:10 GMT"
Subscriber2 received "DATE: Tue, 09 Sep 2014 23:08:10 GMT"
Subscriber1 received "DATE: Tue, 09 Sep 2014 23:08:10 GMT"
publishing date "Tue, 09 Sep 2014 23:08:11 GMT"
Subscriber3 received "DATE: Tue, 09 Sep 2014 23:08:11 GMT"
Subscriber2 received "DATE: Tue, 09 Sep 2014 23:08:11 GMT"
Subscriber1 received "DATE: Tue, 09 Sep 2014 23:08:11 GMT"
publishing date "Tue, 09 Sep 2014 23:08:12 GMT"
Subscriber3 received "DATE: Tue, 09 Sep 2014 23:08:12 GMT"
Subscriber2 received "DATE: Tue, 09 Sep 2014 23:08:12 GMT"
Subscriber1 received "DATE: Tue, 09 Sep 2014 23:08:12 GMT"
ok
Survey
-module(survey).
-export([start/0]).
-define(COUNT, 3).
start() ->
enm:start_link(),
Url = "inproc://survey",
Self = self(),
{ok,Survey} = enm:surveyor([{bind,Url},{deadline,3000}]),
Clients = clients(Url, Self),
ok = enm:send(Survey, httpd_util:rfc1123_date()),
get_responses(Survey),
wait_for_clients(Clients),
enm:close(Survey),
enm:stop().
clients(Url, Parent) ->
clients(Url, Parent, ?COUNT, []).
clients(_, _, 0, Acc) ->
Acc;
clients(Url, Parent, Count, Acc) ->
{ok, Respondent} = enm:respondent([{connect,Url},{active,false},list]),
Name = "Respondent" ++ integer_to_list(Count),
Pid = spawn_link(fun() -> client(Respondent, Name, Parent) end),
clients(Url, Parent, Count-1, [Pid|Acc]).
client(Respondent, Name, Parent) ->
{ok,Msg} = enm:recv(Respondent, 5000),
Date = httpd_util:convert_request_date(Msg),
ok = enm:send(Respondent, term_to_binary(Date)),
io:format("~s got \"~s\"~n", [Name, Msg]),
Parent ! {done, self(), Respondent}.
get_responses(Survey) ->
get_responses(Survey, ?COUNT+1).
get_responses(_, 0) ->
ok;
get_responses(Survey, Count) ->
receive
{nnsurveyor,Survey,BinResp} ->
Response = binary_to_term(BinResp),
io:format("received survey response ~p~n", [Response]);
{nnsurveyor_deadline,Survey} ->
io:format("survey has expired~n")
end,
get_responses(Survey, Count-1).
wait_for_clients([Client|Clients]) ->
receive
{done,Client,Respondent} ->
enm:close(Respondent),
wait_for_clients(Clients)
end;
wait_for_clients([]) ->
ok.
This example creates a surveyor, and several respondents connect to it. The
{deadline,3000}
option used when creating the surveyor socket means
respondents have a maximum of 3 seconds to respond to any survey. The
surveyor sends out the survey, and then collects responses from each of the
respondents. When we hit the survey deadline, the controlling process for
the surveyor socket gets a {nnsurveyor_deadline,Socket}
message.
Survey Results
1> c("examples/survey.erl", [{o,"examples"}]).
{ok,survey}
2> survey:start().
Respondent3 got "Tue, 09 Sep 2014 23:09:34 GMT"
Respondent2 got "Tue, 09 Sep 2014 23:09:34 GMT"
Respondent1 got "Tue, 09 Sep 2014 23:09:34 GMT"
received survey response {{2014,9,9},{23,9,34}}
received survey response {{2014,9,9},{23,9,34}}
received survey response {{2014,9,9},{23,9,34}}
survey has expired
ok
Bus
-module(bus).
-export([start/0]).
-define(COUNT, 4).
start() ->
enm:start_link(),
UrlBase = "inproc://bus",
Buses = connect_buses(UrlBase),
Pids = send_and_receive(Buses, self()),
wait_for_pids(Pids),
enm:stop().
connect_buses(UrlBase) ->
connect_buses(UrlBase, lists:seq(1,?COUNT), []).
connect_buses(UrlBase, [1=Node|Nodes], Buses) ->
Url = make_url(UrlBase, Node),
{ok,Bus} = enm:bus([{bind,Url},{active,false}]),
{ok,_} = enm:connect(Bus, make_url(UrlBase, 2)),
{ok,_} = enm:connect(Bus, make_url(UrlBase, 3)),
connect_buses(UrlBase, Nodes, [{Bus,Node}|Buses]);
connect_buses(UrlBase, [?COUNT=Node|Nodes], Buses) ->
Url = make_url(UrlBase, Node),
{ok,Bus} = enm:bus([{bind,Url},{active,false}]),
{ok,_} = enm:connect(Bus, make_url(UrlBase, 1)),
connect_buses(UrlBase, Nodes, [{Bus,Node}|Buses]);
connect_buses(UrlBase, [Node|Nodes], Buses) ->
Url = make_url(UrlBase, Node),
{ok,Bus} = enm:bus([{bind,Url},{active,false}]),
Urls = [make_url(UrlBase,N) || N <- lists:seq(Node+1,?COUNT)],
[{ok,_} = enm:connect(Bus,U) || U <- Urls],
connect_buses(UrlBase, Nodes, [{Bus,Node}|Buses]);
connect_buses(_, [], Buses) ->
Buses.
send_and_receive(Buses, Parent) ->
send_and_receive(Buses, Parent, []).
send_and_receive([{Bus,Id}|Buses], Parent, Acc) ->
Pid = spawn_link(fun() -> bus(Bus, Id, Parent) end),
send_and_receive(Buses, Parent, [Pid|Acc]);
send_and_receive([], _, Acc) ->
Acc.
bus(Bus, Id, Parent) ->
Name = "node"++integer_to_list(Id),
io:format("node ~w sending \"~s\"~n", [Id, Name]),
ok = enm:send(Bus, Name),
collect(Bus, Id, Parent).
collect(Bus, Id, Parent) ->
case enm:recv(Bus, 1000) of
{ok,Data} ->
io:format("node ~w received \"~s\"~n", [Id, Data]),
collect(Bus, Id, Parent);
{error,etimedout} ->
Parent ! {done, self(), Bus}
end.
wait_for_pids([Pid|Pids]) ->
receive
{done,Pid,Bus} ->
enm:close(Bus),
wait_for_pids(Pids)
end;
wait_for_pids([]) ->
ok.
make_url(Base,N) ->
Base++integer_to_list(N).
In this example consisting of four nodes, each node is connected such that
it receives one message from each of the other nodes. Each node binds to
one bus address and connects to one or more of the other bus addresses
— for example, node 1 connects to nodes 2 and 3, and node 4 connects
only to node 1. This example uses {active,false}
mode since the Erlang
processes calling enm:recv/2
are not the controlling processes for the
receiving sockets.
Bus Results
1> c("examples/bus", [{o,"examples"}]).
{ok,bus}
2> bus:start().
node 4 sending "node4"
node 3 sending "node3"
node 2 sending "node2"
node 1 sending "node1"
node 3 received "node4"
node 2 received "node4"
node 1 received "node4"
node 4 received "node3"
node 3 received "node2"
node 2 received "node3"
node 1 received "node3"
node 4 received "node2"
node 3 received "node1"
node 2 received "node1"
node 1 received "node2"
node 4 received "node1"
ok