ari-proxy

Proxy for the Asterisk REST interface (ARI).

The ARI proxy facilitates scaling of both applications and Asterisk, independently and with minimal coordination. Each Asterisk instance and ARI application pair runs an ari-proxy server instance, which talks to a common NATS or RabbitMQ cluster. Each client application talks to the same message bus. The clients automatically and continuously discover new Asterisk instances, so the only coordination needed is the common location of the message bus.

The ARI proxy allows for:

Any number of applications running the ARI client
Any number of ari-proxy services running on any number of Asterisk instances
Simple call control throughout the cluster, regardless of which Asterisk instance is controlling the call
Simple call distribution regardless of the number of potential application services. (New calls are automatically sent to a single recipient application.)
Simple call event reception by any number of application clients. (No single-app lockout)

Supported message buses:

Proxy server

Docker images are kept up to date with releases and are tagged accordingly. The ari-proxy does not expose any services, so no ports need to be opened for it. However, it does need to know how to connect to both Asterisk and the message bus.

   docker run \
     -e ARI_APPLICATION="my_test_app" \
     -e ARI_USERNAME="demo-user" \
     -e ARI_PASSWORD="supersecret" \
     -e ARI_HTTP_URL="http://asterisk:8088/ari" \
     -e ARI_WEBSOCKET_URL="ws://asterisk:8088/ari/events" \
     -e MESSAGEBUS_URL="nats://nats:4222" \
     cycoresystems/ari-proxy

Binary releases are available on the releases page.

You can also install the server manually:

   go install github.com/CyCoreSystems/ari-proxy/v5

Client library

ari-proxy uses semantic versioning and standard Go modules. To use it in your own Go package, simply reference the github.com/CyCoreSystems/ari-proxy/client/v5 package, and your dependency management tool should be able to manage it.

Usage

Connecting the client to NATS is simple:

import (
   "github.com/CyCoreSystems/ari/v5"
   "github.com/CyCoreSystems/ari-proxy/v5/client"
)

func connect(ctx context.Context, appName string) (ari.Client,error) {
   c, err := client.New(ctx,
      client.WithApplication(appName),
      client.WithURI("nats://natshost:4222"),
   )
}

Connecting the client to RabbitMQ is like:

import (
   "github.com/CyCoreSystems/ari/v5"
   "github.com/CyCoreSystems/ari-proxy/v5/client"
)

func connect(ctx context.Context, appName string) (ari.Client,error) {
   c, err := client.New(ctx,
      client.WithApplication(appName),
      client.WithURI("amqp://user:password@rabbitmqhost:5679/"),
   )
}

Configuration of the client can also be done with environment variables. ARI_APPLICATION can be used to set the ARI application name, and MESSAGEBUS_URL can be used to set the message bus URL. Doing so allows you to get a client connection simply with client.New(ctx).

Once an ari.Client is obtained, the client functions exactly as the native ari client.

Context

Note the use of the context.Context parameter. This can be useful to properly shutdown the client by a controlling context. This shutdown will also close any open subscriptions on the client.

Layers of clients can be used efficiently with different contexts using the New(context.Context) function of each client instance. Subtended clients will be closed with their parents, use a common internal message bus connection, and can be severally closed by their individual contexts. This makes managing many active channels easy and efficient.

Lifecycle

There are two levels of client in use. The first is a connection, which is a long-lived network connection to the message bus. In general, the end user should not close this connection. However, it is available, if necessary, as DefaultConn and offers a Close() function for itself.

The second level is the ARI client. Any number of ARI clients may make use of the same underlying connection, but each client maintains its own separate bus and subscription implementation. Thus, when a user closes its client, the connection is maintained, but all subscriptions are released. Users should always Close() their clients when done with them to avoid accumulating stale subscriptions.

Clustering

The ARI proxy works in a cluster setting by utilizing two coordinates:

The Asterisk ID
The ARI Application

Between the two of these, we can uniquely address each ARI resource, regardless of where the client is located. These pieces of information are handled transparently and internally by the ARI proxy and the ARI proxy client to route commands and events where they should be sent.

Message bus protocol details

The protocol details described below are only necessary to know if you do not use the provided client and/or server. By using both components in this repository, the protocol details below are transparently handled for you.

Subject structure

The message bus subject prefix defaults to ari., and all messages used by this proxy will be prefixed by that term.

Next is added one of four resource classifications:

event - Messages from Asterisk to clients
get - Read-only requests from clients to Asterisk
command - Non-creation operational requests from clients to Asterisk
create - Creation-related requests from clients to Asterisk

After the resource, the ARI application is appended.

Finally, the Asterisk ID will be added to the end. Thus, the subject for an event for the ARI application "test" from the Asterisk box with ID "00:01:02:03:04:05" would look like:

ari.event.test.00:01:02:03:04:05

For a channel creation command to the same app and node:

ari.create.test.00:01:02:03:04:05

The Asterisk ID component of the subject is optional for commands. If a command does not include an Asterisk ID, any ARI proxy running the provided ARI application may respond to the request. (Thus, implicitly, each ARI proxy service listens to both its Asterisk ID-specific command subject and its generic ARI application command subject. In fact, each ARI proxy listens to each of the three levels. A request to ari.command will result in all ARI proxies responding.)

This setup allows for a variable generalization in the listeners by using message bus wildcard subscriptions. For instance, if you want to receive all events for the "test" application regardless from which Asterisk machine they come, you would subscribe to:

ari.event.test.> //NATS ari.event.test.# //RabbitMQ

Dialogs

Events may be further classified by the arbitrary "dialog" ID. If any command specifies a Dialog ID in its metadata, the ARI proxy will further classify events related to that dialog. Relationships are defined by the entity type on which the Dialog-infused command operates.

Dialog-related events are published on their own message bus subject tree, dialogevent. Thus dialogs abstract ARI application and Asterisk ID. An event for dialog "testme123" would be published to:

ari.dialogevent.testme123

Keep in mind that regardless of dialog associations, all events are also published to their appropriate canonical message bus subjects. Dialogs are intended as a mechanism to:

reduce client message traffic load
transcend ARI Applications and/or Asterisk nodes while maintaining logical separation of events

Message delivery

The means of a delivery for a generically-routed message depends on the type of message it is.

Events are always delivered to all listeners.
Read-only commands are delivered to all listeners.
Non-creation operation commands are delivered to all listeners.
Creation-related commands are delivered to only one listener.

Thus, for efficiency, it is always recommended to use as precise a subject line as possible.

Node discovery

Each ARI proxy sends out a periodic ping announcing itself in the cluster. Clients may aggregate these pings to construct an expected map of machines in the cluster. Knowing this map allows the client to optimize its all-listener commands by cancelling the wait period if it receives responses from all nodes before the timeout has elapsed.

ARI proxies listen to ari.ping and send announcements on ari.announce. The structure of the announcement is thus:

{
   "asterisk": "00:10:20:30:40:50",
   "application": "test"
}

Payload structure

For most requests, payloads exactly match their ARI library values. However, treatment of handlers is slightly different.

Instead of a handler, an Entity or array of Entitys is returned. This response type contains the Metadata for the entity (ARI application, Asterisk ID, and optionally Dialog) as well as the unique ID of the entity.