NATS Logo by Example

Key-Value Intro in Key-Value

The key-value (KV) capability in NATS is an abstraction over a stream which models message subjects as keys. It uses a standard set of stream configuration to be optimized for KV workloads.

CLI Go Python JavaScript Rust C# C#2 Java Ruby Elixir Crystal C
Jump to the output or the recording
$ nbe run kv/intro/deno
View the source code or learn how to run this example yourself


import the library - in node.js import {connect, etc} from "nats"; or if not doing a module, const {connect, etc} = require("nats");

import { connect } from "";

const servers = Deno.env.get("NATS_URL") || "nats://localhost:4222";

const nc = await connect({
  servers: servers.split(","),

Bucket basics

A key-value (KV) bucket is created by specifying a bucket name.

const js = nc.jetstream();
const kv = await js.views.kv("profiles");

As one would expect, the KeyValue interface provides the standard Put and Get methods. However, unlike most KV stores, a revision number of the entry is tracked.

await kv.put("sue.color", "blue");
let entry = await kv.get("sue.color");
console.log(`${entry?.key} @ ${entry?.revision} -> ${entry?.string()}`);

await kv.put("sue.color", "green");
entry = await kv.get("sue.color");
console.log(`${entry?.key} @ ${entry?.revision} -> ${entry?.string()}`);

A revision number is useful when you need to enforce optimistic concurrency control on a specific key-value entry. In short, if there are multiple actors attempting to put a new value for a key concurrently, we want to prevent the “last writer wins” behavior which is non-deterministic. To guard against this, we can use the kv.Update method and specify the expected revision. Only if this matches on the server, will the value be updated.

await kv.update("sue.color", "red", 1)
  .then(() => {
    console.error("adding at version 1 should have failed!");
  .catch((err) => {

we expect an error:

entry = await kv.get("sue.color");
console.log(`${entry?.key} @ ${entry?.revision} -> ${entry?.string()}`);

Stream abstraction

Before moving on, it is important to understand that a KV bucket is light abstraction over a standard stream. This is by design since it enables some powerful features which we will observe in a minute.

How exactly is a KV bucket modeled as a stream? When one is created, internally, a stream is created using the KV_ prefix as convention. Appropriate stream configuration are used that are optimized for the KV access patterns, so you can ignore the details.

const status = await kv.status();
console.log("KV stream name",;

Since it is a normal stream, we can create a consumer and fetch messages. If we look at the subject, we will notice that first token is a special reserved prefix, the second token is the bucket name, and remaining suffix is the actually key. The bucket name is inherently a namespace for all keys and thus there is no concern for conflict across buckets. This is different from what we need to do for a stream which is to bind a set of public subjects to a stream.

const c = await js.consumers.get("KV_profiles");
let m = await;
console.log(`${m.subject} @ ${} -> '${m.string()}'`);

await kv.put("sue.color", "yellow");
m = await;
console.log(`${m.subject} @ ${} -> '${m.string()}'`);

Unsurprisingly, we get the new updated value as a message. Since it’s KV interface, we should be able to delete a key as well. Does this result in a new message?

await kv.delete("sue.color");
m = await;
console.log(`${m.subject} @ ${} -> '${m.string()}'`);

🤔 That is useful to get a message that something happened to that key, and that this is considered a new revision. However, how do we know if the new value was set to be nil or the key was deleted? To differentiate, delete-based messages contain a header. Notice the KV-Operation: DEL header.

console.log(`kv operation: ${m.headers.get("KV-Operation")}`);

Watching for changes

Although one could subscribe to the stream directly, it is more convenient to use a KeyWatcher which provides a deliberate API and types for tracking changes over time. Notice that we can use a wildcard which we will come back to..

let history = true;
const iter = await{
  key: "sue.*",
  initializedFn: () => {
    history = false;
(async () => {
  for await (const e of iter) {

Values marked with History are existing values - the watcher by default shows the last value for all the keys in the KV

        history ? "History" : "Updated"
      } ${e.key} @ ${e.revision} -> ${e.string()}`,

To finish this short intro, since we know that keys are subjects under the covers, if we put another key, we can observe the change through the watcher. One other detail to call out is notice the revision for this new key is not 1. It relies on the underlying stream’s message sequence number to indicate the revision. The guarantee being that it is always monotonically increasing, but numbers will be shared across keys (like subjects) rather than sequence numbers relative to each key.

await kv.put("", "pizza");

await nc.close();


[?25l[+] Building 0.0s (0/0)                                                                                                                                                                                                             
[?25h[?25l[+] Building 0.0s (0/0)                                                                                                                                                                                                             
[?25hsue.color @ 1 -> blue
sue.color @ 2 -> green
wrong last sequence: 2
sue.color @ 2 -> green
KV stream name KV_profiles
$KV.profiles.sue.color @ 2 -> 'green'
$KV.profiles.sue.color @ 3 -> 'yellow'
$KV.profiles.sue.color @ 4 -> ''
kv operation: DEL
History sue.color @ 4 -> 
Updated @ 5 -> pizza


Note, playback is half speed to make it a bit easier to follow.