Seneca transport

seneca-transport - a Seneca plugin

This plugin provides the HTTP and TCP transport channels for micro-service messages. It's a built-in dependency of the Seneca module, so you don't need to include it manually. You use this plugin to wire up your micro-services so that they can talk to each other.

Current Version: 0.6.4

Tested on: Seneca 0.6.1, Node 0.10.36

For a gentle introduction to Seneca itself, see the site.

Annotated Source

If you're using this plugin module, feel free to contact me on twitter if you have any questions! :) @rjrodger

This plugin module is included in the main Seneca module:

npm install seneca

To install separately, use:

npm install seneca-transport

Let's do everything in one script to begin with. You'll define a simple Seneca plugin that returns the hex value of color words. In fact, all it can handle is the color red!

You define the action pattern color:red, which aways returns the result {hex:'#FF0000'}. You're also using the name of the function color to define the name of the plugin (see How to write a Seneca plugin).

function color() {
  this.add( 'color:red', function(args,done){
    done(null, {hex:'#FF0000'});

Now, let's create a server and client. The server Seneca instance will load the color plugin and start a web server to listen for inbound messages. The client Seneca instance will submit a color:red message to the server.

var seneca = require('seneca')

You can create multiple instances of Seneca inside the same Node.js process. They won't interfere with each other, but they will share external options from configuration files or the command line.

If you run the full script (full source is in readme-color.js), you'll see the standard Seneca startup log messages, but you won't see anything that tells you what the color plugin is doing since this code doesn't bother printing the result of the action. Let's use a filtered log to output the inbound and outbound action messages from each Seneca instance so we can see what's going on. Run the script with:

node readme-color.js --seneca.log=type:act,regex:color:red

_NOTE: when running the examples in this documentation, you'll find that most of the Node.js processes do not exit. This because they running in server mode. You'll need to kill all the Node.js processes between execution runs. The quickest way to do this is:

$ killall node


This log filter restricts printed log entries to those that report inbound and outbound actions, and further, to those log lines that match the regular expression /color:red/. Here's what you'll see:

[TIME] vy../..15/- DEBUG act -     - IN  485n.. color:red {color=red}   CLIENT 
[TIME] ly../..80/- DEBUG act color - IN  485n.. color:red {color=red}   f2rv..
[TIME] ly../..80/- DEBUG act color - OUT 485n.. color:red {hex=#FF0000} f2rv..
[TIME] vy../..15/- DEBUG act -     - OUT 485n.. color:red {hex=#FF0000} CLIENT

The second field is the identifier of the Seneca instance. You can see that first the client (with an identifier of vy../..15/-) sends the message {color=red}. The message is sent over HTTP to the server (which has an identifier of ly../..80/-). The server performs the action, generating the result {hex=#FF0000}, and sending it back.

The third field, DEBUG, indicates the log level. The next field, act indicates the type of the log entry. Since you specified type:act in the log filter, you've got a match!

The next two fields indicate the plugin name and tag, in this case color -. The plugin is only known on the server side, so the client just indicates a blank entry with -. For more details on plugin names and tags, see How to write a Seneca plugin.

The next field (also known as the case) is either IN or OUT, and indicates the direction of the message. If you follow the flow, you can see that the message is first inbound to the client, and then inbound to the server (the client sends it onwards). The response is outbound from the server, and then outbound from the client (back to your own code). The field after that, 485n.., is the message identifier. You can see that it remains the same over multiple Seneca instances. This helps you to debug message flow.

The next two fields show the action pattern of the message color:red, followed by the actual data of the request message (when inbound), or the response message (when outbound).

The last field f2rv.. is the internal identifier of the action function that acts on the message. On the client side, there is no action function, and this is indicated by the CLIENT marker. If you'd like to match up the action function identifier to message executions, add a log filter to see them:

node readme-color.js --seneca.log=type:act,regex:color:red \
[TIME] ly../..80/- DEBUG plugin color - ADD f2rv.. color:red
[TIME] vy../..15/- DEBUG act    -     - IN  485n.. color:red {color=red}   CLIENT 
[TIME] ly../..80/- DEBUG act    color - IN  485n.. color:red {color=red}   f2rv..
[TIME] ly../..80/- DEBUG act    color - OUT 485n.. color:red {hex=#FF0000} f2rv..
[TIME] vy../..15/- DEBUG act    -     - OUT 485n.. color:red {hex=#FF0000} CLIENT

The filter plugin:color,case:ADD picks out log entries of type plugin, where the plugin has the name color, and where the case is ADD. These entries indicate the action patterns that a plugin has registered. In this case, there's only one, color:red.

You've run this example in a single Node.js process up to now. Of course, the whole point is to run it a separate processes! Let's do that. First, here's the server:

function color() {
  this.add( 'color:red', function(args,done){
    done(null, {hex:'#FF0000'});
var seneca = require('seneca')

Run this in one terminal window with:

$ node readme-color-service.js --seneca.log=type:act,regex:color:red

And on the client side:

var seneca = require('seneca')

And run with:

$ node readme-color-client.js --seneca.log=type:act,regex:color:red

You'll see the same log lines as before, just split over the two processes. The full source code is the test folder.

The default transport mechanism for messages is HTTP. This means you can communicate easily with a Seneca micro-service from other platforms. By default, the listen method starts a web server on port 10101, listening on all interfaces. If you run the readme-color-service.js script again (as above), you can talk to it by POSTing JSON data to the /act path. Here's an example using the command line curl utility.

$ curl -d '{"color":"red"}' http://localhost:10101/act

If you dump the response headers, you'll see some additional headers that give you contextual information. Let's use the -v option of curl to see them:

$ curl -d '{"color":"red"}' -v http://localhost:10101/act
* Connected to localhost ( port 10101 (#0)
> POST /act HTTP/1.1
> User-Agent: curl/7.30.0
> Host: localhost:10101
> Accept: */*
> Content-Length: 15
> Content-Type: application/x-www-form-urlencoded
* upload completely sent off: 15 out of 15 bytes
< HTTP/1.1 200 OK
< Content-Type: application/json
< Cache-Control: private, max-age=0, no-cache, no-store
< Content-Length: 17
< seneca-id: 9wu80xdsn1nu
< seneca-kind: res
< seneca-origin: curl/7.30.0
< seneca-accept: sk5mjwcxxpvh/1409222334824/-
< seneca-time-client-sent: 1409222493910
< seneca-time-listen-recv: 1409222493910
< seneca-time-listen-sent: 1409222493910
< Date: Thu, 28 Aug 2014 10:41:33 GMT
< Connection: keep-alive
* Connection #0 to host localhost left intact 

You can get the message identifier from the seneca-id header, and the identifier of the Seneca instance from seneca-accept.

There are two structures that the submitted JSON document can take:

  • Vanilla JSON containing your request message, plain and simple, as per the example above,
  • OR: A JSON wrapper containing the client details along with the message data.

The JSON wrapper follows the standard form of Seneca messages used in other contexts, such as message queue transports. However, the simple vanilla format is perfectly valid and provided explicitly for integration. The wrapper format is described below.

If you need Seneca to listen on a particular port or host, you can specify these as options to the listen method. Both are optional.

  .listen( { host:'', port:80 } )

On the client side, either with your own code, or the Seneca client, you'll need to use matching host and port options.

$ curl -d '{"color":"red"}'
  .client( { host:'', port:80 } )

You can also set the host and port via the Seneca options facility. When using the options facility, you are setting the default options for all message transports. These can be overridden by arguments to individual listen and client calls.

Let's run the color example again, but with a different port. On the server-side:

$ node readme-color-service.js --seneca.log=type:act,regex:color:red \

And the client-side:

curl -d '{"color":"red"}' -v http://localhost:8888/act


$ node readme-color-client.js --seneca.log=type:act,regex:color:red \

Also included in this plugin is a TCP transport mechanism. The HTTP mechanism offers easy integration, but it is necessarily slower. The TCP transport opens a direct TCP connection to the server. The connection remains open, avoiding connection overhead for each message. The client side of the TCP transport will also attempt to reconnect if the connection breaks, providing fault tolerance for server restarts.

To use the TCP transport, specify a type property to the listen and client methods, and give it the value tcp. Here's the single script example again:


The full source code is in the readme-color-tcp.js file. When you run this script it would be great to verify that the right transport channels are being created. You'd like to see the configuration, and any connections that occur. By default, this information is printed with a log level of INFO, so you will see it if you don't use any log filters.

Of course, we are using a log filter. So let's add another one to print the connection details so we can sanity check the system. We want to print any log entries with a log level of INFO. Here's the command:

$ node readme-color-tcp.js --seneca.log=level:INFO \

This produces the log output:

[TIME] 6g../..49/- INFO  hello  Seneca/0.5.20/6g../..49/-
[TIME] f1../..79/- INFO  hello  Seneca/0.5.20/f1../..79/-
[TIME] f1../..79/- DEBUG act    -         - IN  wdfw.. color:red {color=red} CLIENT 
[TIME] 6g../..49/- INFO  plugin transport - ACT b01d.. listen open {type=tcp,host=,port=10201,...}
[TIME] f1../..79/- INFO  plugin transport - ACT nid1.. client {type=tcp,host=,port=10201,...} any
[TIME] 6g../..49/- INFO  plugin transport - ACT b01d.. listen connection {type=tcp,host=,port=10201,...} remote 52938
[TIME] 6g../..49/- DEBUG act    color     - IN  bpwi.. color:red {color=red} mcx8i4slu68z UNGATE
[TIME] 6g../..49/- DEBUG act    color     - OUT bpwi.. color:red {hex=#FF0000} mcx8i4slu68z
[TIME] f1../..79/- DEBUG act    -         - OUT wdfw.. color:red {hex=#FF0000} CLIENT

The inbound and outbound log entries are as before. In addition, you can see the INFO level entries. At startup, Seneca logs a "hello" entry with the identifier of the current instance execution. This identifier has the form: Seneca/[version]/[12-random-chars]/[timestamp]/[tag]. This identifier can be used for debugging multi-process message flows. The second part is a local timestamp. The third is an optional tag, which you could provide with seneca({tag:'foo'}), although we don't use tags in this example.

There are three INFO level entries of interest. On the server-side, the listen facility logs the fact that it has opened a TCP port, and is now listening for connections. Then the client-side logs that it has opened a connection to the server. And finally the server logs the same thing.

As with the HTTP transport example above, you can split this code into two processes by separating the client and server code. Here's the server:

function color() {
  this.add( 'color:red', function(args,done){
    done(null, {hex:'#FF0000'});
var seneca = require('seneca')

And here's the client:


You can cheat by running the HTTP examples with the additional command line option: --seneca.options.transport.type=tcp.

HTTP and TCP are not the only transport mechanisms available. Of course, in true Seneca-style, the other mechanisms are available as plugins. Here's the list.

If you're written your own transport plugin (see below for instructions), and want to have it listed here, please submit a pull request.

You can use multiple listen and client definitions on the same Seneca instance, in any order. By default, a single client definition will send all unrecognized action patterns over the network. When you have multiple client definitions, it's becuase you want to send some action patterns to one micro-service, and other patterns to other micro-services. To do this, you need to specify the patterns you are interested in. In Seneca, this is done with a pin.

A Seneca pin is a pattern for action patterns. You provide a list of property names and values that must match. Unlike ordinary action patterns, where the values are fixed, with a pin, you can use globs to match more than one value. For example, let's say you have the patterns:

  • foo:1,bar:zed-aaa
  • foo:1,bar:zed-bbb
  • foo:1,bar:zed-ccc

Then you can use these pins to pick out the patterns you want:

  • foo:1 matches foo:1,bar:zed-aaa; foo:1,bar:zed-bbb; foo:1,bar:zed-ccc
  • foo:1, bar:* also matches foo:1,bar:zed-aaa; foo:1,bar:zed-bbb; foo:1,bar:zed-ccc
  • foo:1, bar:*-aaa matches only foo:1,bar:zed-aaa

Let's extend the color service example. You'll have three separate services, all running in separate processes. They will listen on ports 8081, 8082, and 8083 respectively. You'll use command line arguments for settings. Here's the service code (see readme-many-colors-server.js):

var color  = process.argv[2]
var hexval = process.argv[3]
var port   = process.argv[4]
var seneca = require('seneca')
  .add( 'color:'+color, function(args,done){
    done(null, {hex:'#'+hexval});
  .listen( port )'color',color,hexval,port)

This service takes in a color name, a color hexadecimal value, and a port number from the command line. You can also see how the listen method can take a single argument, the port number. To offer the color:red service, run this script with:

$ node readme-many-colors-server.js red FF0000 8081

And you can test with:

$ curl -d '{"color":"red"}' http://localhost:8081/act

Of course, you need to use some log filters to pick out the activity you're interested in. In this case, you've used a call to dump out settings. You'll also want to see the actions as the occur. Try this:

node readme-many-colors-server.js red FF0000 8081 --seneca.log=level:info \

And you'll get:

[TIME] mi../..66/- INFO  hello  Seneca/0.5.20/mi../..66/-
[TIME] mi../..66/- INFO  color  red       FF0000 8081
[TIME] mi../..66/- INFO  plugin transport -      ACT 7j.. listen {type=web,port=8081,host=,path=/act,protocol=http,timeout=32778,msgprefix=seneca_,callmax=111111,msgidlen=12,role=transport,hook=listen}
[TIME] mi../..66/- DEBUG act    -         -      IN  ux.. color:red {color=red} 9l..
[TIME] mi../..66/- DEBUG act    -         -      OUT ux.. color:red {hex=#FF0000} 9l..

You can see the custom INFO log entry at the top, and also the transport settings after that.

Let's run three of these servers, one each for red, green and blue. Let's also run a client to connect to them.

Let's make it interesting. The client will listen so that it can handle incoming actions, and pass them on to the appropriate server by using a pin. The client will also define a new action that can aggregate color lookups.

var seneca = require('seneca')
  // send matching actions out over the network 
  .client({ port:8081, pin:'color:red' })
  .client({ port:8082, pin:'color:green' })
  .client({ port:8083, pin:'color:blue' })
  // an aggregration action that calls other actions 
  .add( 'list:colors', function( argsdone ){
    var seneca = this
    var colors = {}
    args.names.forEach(function( name ){
      seneca.act({color:name}, function(errresult){
        if( err ) return done(err);
        colors[name] = result.hex
        if( Object.keys(colors).length == args.names.length ) {
          return done(null,colors)
  // this is a sanity check 

This code calls the client method three times. Each time, it specifies an action pattern pin, and a destination port. And action submitted to this Seneca instance via the act method will be matched against these pin patterns. If there is a match, they will not be processed locally. Instead they will be sent out over the network to the micro-service that deals with them.

In this code, you are using the default HTTP transport, and just changing the port number to connect to. This reflects the fact that each color micro-service runs on a separate port.

The listen call at the bottom makes this "client" also listen for inbound messages. So if you run, say the color:red service, and also run the client, then you can send color:red messages to the client.

You need to run four processes:

node readme-many-colors-server.js red FF0000 8081 --seneca.log=level:info --seneca.log=type:act,regex:color &
node readme-many-colors-server.js green 00FF00 8082 --seneca.log=level:info --seneca.log=type:act,regex:color &
node readme-many-colors-server.js blue 0000FF 8083 --seneca.log=level:info --seneca.log=type:act,regex:color &
node readme-many-colors-client.js --seneca.log=type:act,regex:CLIENT &

And then you can test with:

$ curl -d '{"color":"red"}' http://localhost:10101/act
$ curl -d '{"color":"green"}' http://localhost:10101/act
$ curl -d '{"color":"blue"}' http://localhost:10101/act

These commands are all going via the client, which is listening on port 10101.

The client code also includes an aggregation action, list:colors. This lets you call multiple color actions and return one result. This is a common micro-service pattern.

The script wraps all this up into one place for you so that it is easy to run.

Seneca does not require you to use message transports. You can run everything in one process. But when the time comes, and you need to scale, or you need to break out micro-services, you have the option to do so.

There is no message protocol as such, as the data representation of the underlying message transport is used. However, the plain text message representation is JSON in all known transports.

For the HTTP transport, message data is encoded as per the HTTP protocol. For the TCP transport, UTF8 JSON is used, with one well-formed JSON object per line (with a single "\n" as line terminator).

For other transports, please see the documentation for the underlying protocol. In general the transport plugins, such as seneca-redis-transport will handle this for you so that you only have to think in terms of JavaScript objects

The JSON object is a wrapper for the message data. The wrapper contains some tracking fields to make debugging easier, these are:

  • id: action identifier (appears in Seneca logs after IN/OUT)
  • kind: 'act' for inbound actions, 'res' for outbound responses
  • origin: identifier of orginating Seneca instance, where action is submitted
  • accept: identifier of accepting Seneca instance, where action is performed
  • time:
    • client_sent: client timestamp when message sent
    • listen_recv: server timestamp when message received
    • listen_sent: server timestamp when response sent
    • client_recv: client timestamp when response received
  • act: action message data, as submitted to Seneca
  • res: response message data, as provided by Seneca
  • error: error message, if any
  • input: input generating error, if any

To write your own transport, the best approach is to copy one of the existing ones:

Choose a type for your transport, say "foo". You will need to implement two patterns:

  • role:transport, hook:listen, type:foo
  • role:transport, hook:client, type:foo

To implement the client, use the template:

var transport_utils = seneca.export('transport/utils')
function hook_client_redis( argsclientdone ) {
  var seneca         = this
  var type           = args.type
  // get your transport type default options 
  var client_options = seneca.util.clean(_.extend({},options[type],args))
  transport_utils.make_client( make_send, client_options, clientdone )
  // implement your transport here 
  function make_send( spectopicsend_done ) {
    // see an existing transport for full details 

To implement the server, use the template:

var transport_utils = seneca.export('transport/utils')
function hook_listen_redis( argsdone ) {
  var seneca         = this
  var type           = args.type
  // get your transport type default options 
  var listen_options = seneca.util.clean(_.extend({},options[type],args))
  // get inbound data, and... 
  transport_utils.handle_request( seneca, data, listen_options, function(out){
    // see an existing transport for full details 

Message transport code should be written very carefully as it will be subject to high load and many error conditions - have fun!

The transport plugin family uses an extension to the normal Seneca options facility. As well as supporting the standard method for defining options (see How to Write a Plugin), you can also supply options via arguments to the client or listen methods, and via the type name of the transport under the top-level transport property.

The primary options are:

  • msgprefix: a string to prefix to topic names so that they are namespaced
  • callmax: the maximum number of in-flight request/response messages to cache
  • msgidlen: length of the message indentifier string

These can be set within the top-level transport property of the main Seneca options tree:

var seneca = require('seneca')

Each transport type forms a sub-level within the transport option. The recognized types depend on the transport plugins you have loaded. By default, web and tcp are available. To use redis, for example, you need to do this:

var seneca = require('seneca')
  // assumes npm install seneca-redis-transport 

You can set transport-level options inside the type property:

var seneca = require('seneca')

The transport-level options vary by transport. Here are the default ones for HTTP:

  • type: type name; constant: 'web'
  • port: port number; default: 10101
  • host: hostname; default: '' (all interfaces)
  • path: URL path to submit messages; default: '/act'
  • protocol: HTTP protocol; default 'http'
  • timeout: timeout in milliseconds; default: 5555

And for TCP:

  • type: type name; constant: 'tcp'
  • port: port number; default: 10201
  • host: hostname; default: '' (all interfaces)
  • timeout: timeout in milliseconds; default: 5555

The client and listen methods accept an options object as the primary way to specify options:

var seneca = require('seneca')

As a convenience, you can specify the port and host as optional arguments:

var seneca = require('seneca')
  .client( 8080 )
  .listen( 9090, 'localhost')

To see the options actually in use at any time, you can call the seneca.options() method. Or try

$ node seneca-script.js --seneca.log=type:options

This module itself does not contain any direct reference to Seneca, as it is a Seneca dependency. However, Seneca is needed to test it, so the test script will perform an npm install seneca (if needed). This is not saved to package.json.

npm test
  • 0.2.6: fixed error transmit bug