NATS.deno - A NATS client for Deno

A Deno client for the NATS messaging system.

Installation

You can get the latest release version like this:

import * as nats from "https://deno.land/x/nats/src/mod.ts";

To specify a specific released version, simply replace nats with nats@versionTag.

You can get the current development version by:

import * as nats from "https://raw.githubusercontent.com/nats-io/nats.deno/main/src/mod.ts";

Documentation

While the best documentation is looking at code examples, you may want to browse the JSDoc documentation. The best entry point into the JS Doc is the NatsConnection all functionality starts with a connection.

Check out NATS by example - An evolving collection of runnable, cross-client reference examples for NATS.

Basics

Connecting to a nats-server

To connect to a server you use the connect() function. It returns a connection that you can use to interact with the server. You can customize the behavior of the client by specifying many ConnectionOptions.

By default, a connection will attempt a connection on 127.0.0.1:4222. If the connection is dropped, the client will attempt to reconnect. You can customize the server you want to connect to by specifying port (for local connections), or full host port on the servers option. Note that the servers option can be a single hostport (a string) or an array of hostports.

The example below will attempt to connect to different servers by specifying different ConnectionOptions. At least two of them should work if your internet is working.

// import the connect function
import { connect } from "../../src/mod.ts";

const servers = [
  {},
  { servers: ["demo.nats.io:4442", "demo.nats.io:4222"] },
  { servers: "demo.nats.io:4443" },
  { port: 4222 },
  { servers: "localhost" },
];
await servers.forEach(async (v) => {
  try {
    const nc = await connect(v);
    console.log(`connected to ${nc.getServer()}`);
    // this promise indicates the client closed
    const done = nc.closed();
    // do something with the connection

    // close the connection
    await nc.close();
    // check if the close was OK
    const err = await done;
    if (err) {
      console.log(`error closing:`, err);
    }
  } catch (err) {
    console.log(`error connecting to ${JSON.stringify(v)}`);
  }
});

To disconnect from the nats-server, call close() on the connection. A connection can also be terminated when an unexpected error happens. For example, the server returns a run-time error. In those cases, the client will re-initiate a connection.

By default, the client will always attempt to reconnect if the connection is closed for a reason other than calling close(). To get notified when the connection is closed for some reason, await the resolution of the Promise returned by closed(). If closed resolves to a value, the value is a NatsError indicating why the connection closed.

Publish and Subscribe

The basic client operations are publish to send messages and subscribe to receive messages.

Messages are published to a subject. A subject is like an URL with the exception that it doesn't specify an actual endpoint. All recipients that have expressed interest in a subject will receive messages addressed to that subject (provided they have access and permissions to get it). To express interest in a subject, you create a subscription.

In JavaScript clients (websocket, Deno, or Node) subscriptions work as an async iterator - clients simply loop to process messages as they become available.

NATS messages are payload agnostic. Payloads are Uint8Arrays. You can easily convert to and from JSON or strings by using JSONCodec or StringCodec, or a custom Codec.

To cancel a subscription and terminate your interest, you call unsubscribe() or drain() on a subscription. Unsubscribe will typically terminate regardless of whether there are messages in flight for the client. Drain ensures that all messages that are inflight are processed before canceling the subscription. Connections can also be drained as well. Draining a connection closes it, after all subscriptions have been drained and all outbound messages have been sent to the server.

// import the connect function
import { connect, StringCodec } from "../../src/mod.ts";

// to create a connection to a nats-server:
const nc = await connect({ servers: "demo.nats.io:4222" });

// create a codec
const sc = StringCodec();
// create a simple subscriber and iterate over messages
// matching the subscription
const sub = nc.subscribe("hello");
(async () => {
  for await (const m of sub) {
    console.log(`[${sub.getProcessed()}]: ${sc.decode(m.data)}`);
  }
  console.log("subscription closed");
})();

nc.publish("hello", sc.encode("world"));
nc.publish("hello", sc.encode("again"));

// we want to insure that messages that are in flight
// get processed, so we are going to drain the
// connection. Drain is the same as close, but makes
// sure that all messages in flight get seen
// by the iterator. After calling drain on the connection
// the connection closes.
await nc.drain();

JSONCodec allows you to encode and decode JSON and if you are using typescript, you can even use generics to have the compiler help you:

// create a codec
const sc = JSONCodec<Person>();

// create a simple subscriber and iterate over messages
// matching the subscription
const sub = nc.subscribe("people");
(async () => {
  for await (const m of sub) {
    // typescript will see this as a Person
    const p = sc.decode(m.data);
    console.log(`[${sub.getProcessed()}]: ${p.name}`);
  }
})();

// if you made a typo or added other properties
// the compiler will get angry
const p = { name: "Memo" } as Person;
nc.publish("people", sc.encode(p));

Wildcard Subscriptions

Subjects can be used to organize messages into hierarchies. For example, a subject may contain additional information that can be useful in providing a context to the message, such as the ID of the client that sent the message, or the region where a message originated.

Instead of subscribing to each specific subject, you can create subscriptions that have subjects with wildcards. Wildcards match one or more tokens in a subject. A token is a string following a period.

All subscriptions are independent. If two different subscriptions match a subject, both will get to process the message:

import { connect, StringCodec, Subscription } from "../../src/mod.ts";
const nc = await connect({ servers: "demo.nats.io:4222" });
const sc = StringCodec();

// subscriptions can have wildcard subjects
// the '*' matches any string in the specified token position
const s1 = nc.subscribe("help.*.system");
const s2 = nc.subscribe("help.me.*");
// the '>' matches any tokens in that position or following
// '>' can only be specified at the end of the subject
const s3 = nc.subscribe("help.>");

async function printMsgs(s: Subscription) {
  let subj = s.getSubject();
  console.log(`listening for ${subj}`);
  const c = (13 - subj.length);
  const pad = "".padEnd(c);
  for await (const m of s) {
    console.log(
      `[${subj}]${pad} #${s.getProcessed()} - ${m.subject} ${
        m.data ? " " + sc.decode(m.data) : ""
      }`,
    );
  }
}

printMsgs(s1);
printMsgs(s2);
printMsgs(s3);

// don't exit until the client closes
await nc.closed();

Services: Request/Reply

Request/Reply is NATS equivalent to an HTTP request. To make requests you publish messages as you did before, but also specify a reply subject. The reply subject is where a service will publish your response.

NATS provides syntactic sugar, for publishing requests. The request() API will generate a reply subject and manage the creation of a subscription under the covers. It will also start a timer to ensure that if a response is not received within your alloted time, the request fails. The example also illustrates a graceful shutdown.

Services

Here's an example of a service. It is a bit more complicated than expected simply to illustrate not only how to create responses, but how the subject itself is used to dispatch different behaviors.

import { connect, StringCodec, Subscription } from "../../src/mod.ts";

// create a connection
const nc = await connect({ servers: "demo.nats.io" });

// create a codec
const sc = StringCodec();

// this subscription listens for `time` requests and returns the current time
const sub = nc.subscribe("time");
(async (sub: Subscription) => {
  console.log(`listening for ${sub.getSubject()} requests...`);
  for await (const m of sub) {
    if (m.respond(sc.encode(new Date().toISOString()))) {
      console.info(`[time] handled #${sub.getProcessed()}`);
    } else {
      console.log(`[time] #${sub.getProcessed()} ignored - no reply subject`);
    }
  }
  console.log(`subscription ${sub.getSubject()} drained.`);
})(sub);

// this subscription listens for admin.uptime and admin.stop
// requests to admin.uptime returns how long the service has been running
// requests to admin.stop gracefully stop the client by draining
// the connection
const started = Date.now();
const msub = nc.subscribe("admin.*");
(async (sub: Subscription) => {
  console.log(`listening for ${sub.getSubject()} requests [uptime | stop]`);
  // it would be very good to verify the origin of the request
  // before implementing something that allows your service to be managed.
  // NATS can limit which client can send or receive on what subjects.
  for await (const m of sub) {
    const chunks = m.subject.split(".");
    console.info(`[admin] #${sub.getProcessed()} handling ${chunks[1]}`);
    switch (chunks[1]) {
      case "uptime":
        // send the number of millis since up
        m.respond(sc.encode(`${Date.now() - started}`));
        break;
      case "stop": {
        m.respond(sc.encode(`[admin] #${sub.getProcessed()} stopping....`));
        // gracefully shutdown
        nc.drain()
          .catch((err) => {
            console.log("error draining", err);
          });
        break;
      }
      default:
        console.log(
          `[admin] #${sub.getProcessed()} ignoring request for ${m.subject}`,
        );
    }
  }
  console.log(`subscription ${sub.getSubject()} drained.`);
})(msub);

// wait for the client to close here.
await nc.closed().then((err?: void | Error) => {
  let m = `connection to ${nc.getServer()} closed`;
  if (err) {
    m = `${m} with an error: ${err.message}`;
  }
  console.log(m);
});

Making Requests

Here's a simple example of a client making a simple request from the service above:

import { connect, Empty, StringCodec } from "../../src/mod.ts";

// create a connection
const nc = await connect({ servers: "demo.nats.io:4222" });

// create an encoder
const sc = StringCodec();

// the client makes a request and receives a promise for a message
// by default the request times out after 1s (1000 millis) and has
// no payload.
await nc.request("time", Empty, { timeout: 1000 })
  .then((m) => {
    console.log(`got response: ${sc.decode(m.data)}`);
  })
  .catch((err) => {
    console.log(`problem with request: ${err.message}`);
  });

await nc.close();

Queue Groups

Queue groups allow scaling of services horizontally. Subscriptions for members of a queue group are treated as a single service. When you send a message to a queue group subscription, only a single client in a queue group will receive it.

There can be any number of queue groups. Each group is treated as its own independent unit. Note that non-queue subscriptions are also independent of subscriptions in a queue group.

import {
  connect,
  NatsConnection,
  StringCodec,
  Subscription,
} from "../../src/mod.ts";

// this is the definition of a service with `count` members in them.
// if the queue is specified, the they will be part of a queue
async function createService(
  name: string,
  count: number = 1,
  queue: string = "",
): Promise<NatsConnection[]> {
  const conns: NatsConnection[] = [];
  for (let i = 1; i <= count; i++) {
    const n = queue ? `${name}-${i}` : name;
    const nc = await connect(
      { servers: "demo.nats.io:4222", name: `${n}` },
    );
    nc.closed()
      .then((err) => {
        if (err) {
          console.error(
            `service ${n} exited because of error: ${err.message}`,
          );
        }
      });
    // create a subscription - note the option for a queue, if set
    // any client with the same queue will be a member of the group.
    const sub = nc.subscribe("echo", { queue: queue });
    const _ = handleRequest(n, sub);
    console.log(`${nc.options.name} is listening for 'echo' requests...`);
    conns.push(nc);
  }
  return conns;
}

const sc = StringCodec();

// simple handler for service requests
async function handleRequest(name: string, s: Subscription) {
  const p = 12 - name.length;
  const pad = "".padEnd(p);
  for await (const m of s) {
    // respond returns true if the message had a reply subject, thus it could respond
    if (m.respond(m.data)) {
      console.log(
        `[${name}]:${pad} #${s.getProcessed()} echoed ${sc.decode(m.data)}`,
      );
    } else {
      console.log(
        `[${name}]:${pad} #${s.getProcessed()} ignoring request - no reply subject`,
      );
    }
  }
}

// let's create two queue groups and a standalone subscriber
const conns: NatsConnection[] = [];
conns.push(...await createService("echo", 3, "echo"));
conns.push(...await createService("other-echo", 2, "other-echo"));
conns.push(...await createService("standalone"));

const a: Promise<void | Error>[] = [];
conns.forEach((c) => {
  a.push(c.closed());
});
await Promise.all(a);

Run it and publish a request to the subject echo to see what happens.

Advanced Usage

Headers

NATS headers are similar to HTTP headers. Headers are enabled automatically if the server supports them. Note that if you publish a message using headers but the server doesn't support them, an Error is thrown. Also note that even if you are publishing a message with a header, it is possible for the recipient to not support them.

import { connect, createInbox, Empty, headers } from "../../src/mod.ts";
import { nuid } from "../../nats-base-client/nuid.ts";

const nc = await connect(
  {
    servers: `demo.nats.io`,
  },
);

const subj = createInbox();
const sub = nc.subscribe(subj);
(async () => {
  for await (const m of sub) {
    if (m.headers) {
      for (const [key, value] of m.headers) {
        console.log(`${key}=${value}`);
      }
      // reading a header is not case sensitive
      console.log("id", m.headers.get("id"));
    }
  }
})().then();

// header names can be any printable ASCII character with the  exception of `:`.
// header values can be any ASCII character except `\r` or `\n`.
// see https://www.ietf.org/rfc/rfc822.txt
const h = headers();
h.append("id", nuid.next());
h.append("unix_time", Date.now().toString());
nc.publish(subj, Empty, { headers: h });

await nc.flush();
await nc.close();

No Responders

Requests can fail for many reasons. A common reason for a failure is the lack of interest in the subject. Typically these surface as a timeout error. If the server is enabled to use headers, it will also enable a no responders feature. If you send a request for which there's no interest, the request will be immediately rejected:

const nc = await connect({
  servers: `demo.nats.io`,
});

try {
  const m = await nc.request("hello.world");
  console.log(m.data);
} catch (err) {
  const nerr = err as NatsError;
  switch (nerr.code) {
    case ErrorCode.NoResponders:
      console.log("no one is listening to 'hello.world'");
      break;
    case ErrorCode.Timeout:
      console.log("someone is listening but didn't respond");
      break;
    default:
      console.log("request failed", err);
  }
}

await nc.close();

Authentication

NATS supports many different forms of credentials:

• username/password
• token
• NKEYS
• client certificates
• JWTs

For user/password and token authentication, you can simply provide them as ConnectionOptions - see user, pass, token. Internally these mechanisms are implemented as an Authenticator. An Authenticator is simply a function that handles the type of authentication specified.

Setting the user/pass or token options, simply initializes an Authenticator and sets the username/password.

// if the connection requires authentication, provide `user` and `pass` or
// `token` options in the NatsConnectionOptions
import { connect } from "src/mod.ts";

const nc1 = await connect({
  servers: "127.0.0.1:4222",
  user: "jenny",
  pass: "867-5309",
});
const nc2 = await connect({ port: 4222, token: "t0pS3cret!" });

Authenticators

NKEYs and JWT authentication are more complex, as they cryptographically respond to a server challenge.

Because NKEY and JWT authentication may require reading data from a file or an HTTP cookie, these forms of authentication will require a bit more from the developer to activate them. However, the work is related to accessing these resources varies depending on the platform.

After the credential artifacts are read, you can use one of these functions to create the authenticator. You then simply assign it to the authenticator property of the ConnectionOptions:

• nkeyAuthenticator(seed?: Uint8Array | (() => Uint8Array)): Authenticator
• jwtAuthenticator(jwt: string | (() => string), seed?: Uint8Array | (()=> Uint8Array)): Authenticator
• credsAuthenticator(creds: Uint8Array): Authenticator

The first two options provide the ability to specify functions that return the desired value. This enables dynamic environments such as a browser where values accessed by fetching a value from a cookie.

Here's an example:

// read the creds file as necessary, in the case it
// is part of the code for illustration purposes
const creds = `-----BEGIN NATS USER JWT-----
    eyJ0eXAiOiJqdSDJB....
  ------END NATS USER JWT------

************************* IMPORTANT *************************
  NKEY Seed printed below can be used sign and prove identity.
  NKEYs are sensitive and should be treated as secrets.

  -----BEGIN USER NKEY SEED-----
    SUAIBDPBAUTW....
  ------END USER NKEY SEED------
`;

const nc = await connect(
  {
    port: 4222,
    authenticator: credsAuthenticator(new TextEncoder().encode(creds)),
  },
);

Flush

Flush sends a PING to the server. When the server responds with PONG you are guaranteed that all pending data was sent and received by the server. Note ping() effectively adds a server round-trip. All NATS clients handle their buffering optimally, so ping(): Promise<void> shouldn't be used except in cases where you are writing some sort of test.

nc.publish("foo");
nc.publish("bar");
await nc.flush();

PublishOptions

When you publish a message you can specify some options:

• reply - this is a subject to receive a reply (you must setup a subscription) before you publish.
• headers - a set of headers to decorate the message.

SubscriptionOptions

You can specify several options when creating a subscription:

• max: maximum number of messages to receive - auto unsubscribe
• timeout: how long to wait for the first message
• queue: the queue group name the subscriber belongs to
• callback: a function with the signature (err: NatsError|null, msg: Msg) => void; that should be used for handling the message. Subscriptions with callbacks are NOT iterators.

Auto Unsubscribe

// subscriptions can auto unsubscribe after a certain number of messages
nc.subscribe("foo", { max: 10 });

Timeout Subscriptions

// create subscription with a timeout, if no message arrives
// within the timeout, the function running the iterator with
// reject - depending on how you code it, you may need a
// try/catch block.
const sub = nc.subscribe("hello", { timeout: 1000 });
(async () => {
  for await (const m of sub) {
  }
})().catch((err) => {
  if (err.code === ErrorCode.Timeout) {
    console.log(`sub timed out!`);
  } else {
    console.log(`sub iterator got an error!`);
  }
});

RequestOptions

When making a request, there are several options you can pass:

• timeout: how long to wait for the response
• headers: optional headers to include with the message
• noMux: create a new subscription to handle the request. Normally a shared subscription is used to receive response messages.
• reply: optional subject where the reply should be sent.

noMux and reply

Under the hood, the request API simply uses a wildcard subscription to handle all requests you send.

In some cases, the default subscription strategy doesn't work correctly. For example, a client may be constrained by the subjects where it can receive replies.

When noMux is set to true, the client will create a normal subscription for receiving the response to a generated inbox subject before the request is published. The reply option can be used to override the generated inbox subject with an application provided one. Note that setting reply requires noMux to be true:

const m = await nc.request(
  "q",
  Empty,
  { reply: "bar", noMux: true, timeout: 1000 },
);

Draining Connections and Subscriptions

Draining provides for a graceful way to unsubscribe or close a connection without losing messages that have already been dispatched to the client.

You can drain a subscription or all subscriptions in a connection.

When you drain a subscription, the client sends an unsubscribe protocol message to the server followed by a flush. The subscription handler is only removed after the server responds. Thus all pending messages for the subscription have been processed.

Draining a connection, drains all subscriptions. However when you drain the connection it becomes impossible to make new subscriptions or send new requests. After the last subscription is drained it also becomes impossible to publish a message. These restrictions do not exist when just draining a subscription.

Lifecycle/Informational Events

Clients can get notification on various event types:

• Events.DISCONNECT
• Events.RECONNECT
• Events.UPDATE
• Events.LDM
• Events.ERROR

The first two fire when a client disconnects and reconnects respectively. The payload will be the server where the event took place.

The UPDATE event notifies whenever the client receives a cluster configuration update. The ServersChanged interface provides two arrays: added and deleted listing the servers that were added or removed.

The LDM event notifies that the current server has signaled that it is running in Lame Duck Mode and will evict clients. Depending on the server configuration policy, the client may want to initiate an ordered shutdown, and initiate a new connection to a different server in the cluster.

The ERROR event notifies you of async errors that couldn't be routed in a more precise way to your client. For example, permission errors for a subscription or request, will properly be reported by the subscription or request. However, permission errors on publish will be reported via the status mechanism.

const nc = await connect(opts);
(async () => {
  console.info(`connected ${nc.getServer()}`);
  for await (const s of nc.status()) {
    console.info(`${s.type}: ${s.data}`);
  }
})().then();

nc.closed()
  .then((err) => {
    console.log(
      `connection closed ${err ? " with error: " + err.message : ""}`,
    );
  });

Be aware that when a client closes, you will need to wait for the closed() promise to resolve. When it resolves, the client is done and will not reconnect.

Async vs. Callbacks

Previous versions of the JavaScript NATS clients specified callbacks for message processing. This required complex handling logic when a service required coordination of operations. Callbacks are an inversion of control anti-pattern.

The async APIs trivialize complex coordination and makes your code easier to maintain. With that said, there are some implications:

• Async subscriptions buffer inbound messages.
• Subscription processing delays until the runtime executes the promise related microtasks at the end of an event loop.

In a traditional callback-based library, I/O happens after all data yielded by a read in the current event loop completes processing. This means that callbacks are invoked as part of processing. With async, the processing is queued in a microtask queue. At the end of the event loop, the runtime processes the microtasks, which in turn resumes your functions. As expected, this increases latency, but also provides additional liveliness.

To reduce async latency, the NATS client allows processing a subscription in the same event loop that dispatched the message. Simply specify a callback in the subscription options. The signature for a callback is (err: (NatsError|null), msg: Msg) => void. When specified, the subscription iterator will never yield a message, as the callback will intercept all messages.

Note that callback likely shouldn't even be documented, as likely it is a workaround to an underlying application problem where you should be considering a different strategy to horizontally scale your application, or reduce pressure on the clients, such as using queue workers, or more explicitly targeting messages. With that said, there are many situations where using callbacks can be more performant or appropriate.

Connection Options

The following is the list of connection options and default values.

TlsOptions

In some Node and Deno clients, having the option set to an empty option, requires the client have a secured connection.

Jitter

The settings reconnectTimeWait, reconnectJitter, reconnectJitterTLS, reconnectDelayHandler are all related. They control how long before the NATS client attempts to reconnect to a server it has previously connected.

The intention of the settings is to spread out the number of clients attempting to reconnect to a server over a period of time, and thus preventing a "Thundering Herd".

The relationship between these is:

• If reconnectDelayHandler is specified, the client will wait the value returned by this function. No other value will be taken into account.
• If the client specified TLS options, the client will generate a number between 0 and reconnectJitterTLS and add it to reconnectTimeWait.
• If the client didn't specify TLS options, the client will generate a number between 0 and reconnectJitter and add it to reconnectTimeWait.

JetStream

JetStream is the NATS persistence engine providing streaming, message, and worker queues with At-Least-Once semantics. Support for JetStream is built-in.

Service API

The service API allows you to easily build NATS services The services API is currently in beta functionality.