stianeikeland / node-etcd

:satellite: Etcd client for nodejs

node-etcd

A nodejs library for ETCD v2, written in coffee-script.

Install

For nodes >= 6.x:

$ npm install node-etcd

For nodes 4.x <= node < 6.x:

$ npm install node-etcd@6

For 0.10.x <= nodejs <= 4.x and iojs:

$ npm install node-etcd@5.1.0

For nodejs == 0.8:

$ npm install node-etcd@3.0.2

Changes

• 7.0.0

  • Fixing vulnerabilities
  • Drop support for nodejs version 4 &4

• 6.0.0

  • Migrate from underscore to lodash for performance / @derektbrown
  • Updated package versions / @derektbrown
  • Update deasync call for newer node versions / @derektbrown
  • Add istanbul coverage reporting / @derektbrown

• 5.1.0

  • Upgrade deasync dep (caused build problems on newer node) #67 / @jefflembeck
  • Upgrade request dep (security vulnerability) #71 / @esatterwhite
  • Sync functions no longer mutate input opts.

• 5.0.3

  • Fix bug #56 (exception when calling mkdir with no options or callback)

• 5.0.2

  • Update deasync dependency, possible fix for #47.

• 5.0.1

  • Was forced to publish 5.0.0 as 5.0.1 because of previous beta tag.

• 5.0.0

  • Supports basic auth, timeout, etc. See options.
  • Breaking: Constructor changes (see below)
  • Breaking: Must provide https url to use https
  • Breaking: Uses new default port 2379.

• 4.2.1

  • Newer deasync fixes issues with iojs 3.3.0 and nodejs 4.0.0.

• 4.1.0

  • Bumps request library version to v2.60.0, this solves an issue with HTTP proxies. HTTP_PROXY and NO_PROXY env variables should now work as expected for all requests. See issue #40

• 4.0.2

  • 307 redirects from etcd 0.4.x clusters when using SSL didn't work properly because of a change in the underlying request library. See issue #39

• 4.0.1

  • Minor fixes for syncronous operations, better handling of server failure.

• 4.0.0

  • Adds support for synchronous operations (@brockwood) - See Synchronous Operations.
  • Adds support for iojs.
  • Drops support for nodejs 0.8 (use v3.x.x).
  • Upgrade dependencies.

• 3.0.2 - Handle cluster leader election periods better (etcd will disconnect us and reject new connections until a new leader is chosen). The client will now retry 3 times with exp backoff if it believes a cluster election is in progress. Retry count is controllable via the { maxRetries: x } option for a request. (npm failed on me and I had to publish as 3.0.2)

• 3.0.0 - Added cluster support, library now accepts a list of servers to connect to, see constructor changes below. All requests now return a cancellation token, meaning you can cancel requests by calling .cancel() or .abort(). This release might break something if you've previously depended on the leaky abstraction to the request library (request object from request library was returned on all api calls - this has been replaced by the cancellation token - the current request is still available under .req on the token if you really need it.).

• 2.1.5 - Watcher: try to resync if etcd reports cleared index

• 2.1.4 - Don't wait before reconnecting if Etcd server times out our watcher.

• 2.1.3 - Etcd sends an empty response on timeout in recent versions. Parsing the empty message caused watcher to emit error. Now it reconnects instead.

• 2.1.2 - Exponential backoff (retry), fix spinning reconnect on error. (@ptte)

• 2.1.1 - Increase request pool.maxSockets to 100

• 2.1.0 - Use proper error objects instead of strings for errors.

• 2.0.10 - Fix error in documentation

• 2.0.9 - Added .post() alias of .create(). Added .compareAndDelete() (for etcd v0.3.0)

• 2.0.8 - Watchers can be canceled. In-order keys using #create(). Raw requests using #raw().

• 2.0.7 - Avoid calling callback if callback not given.

• 2.0.6 - Refactoring, fix responsehandler error.

• 2.0.5 - Undo use of 'x-etcd-index', this refers to global state.

• 2.0.4 - Use 'x-etcd-index' for index when watching a key.

• 2.0.3 - Watcher supports options. Watcher emits etcd action type.

• 2.0.2 - Mkdir and rmdir. Fix watcher for v2 api.

• 2.0.1 - Watch, delete and stats now use new v2 api. Added testAndSet convenience method.

• 2.0.0 - Basic support for etcd protocol v2. set, get, del now supports options.

• 0.6.1 - Fixes issue #10, missing response caused error when server connection failed / server responded incorrectly.

• 0.6.0 - Watcher now emits 'error' on invalid responses.

Basic usage

var Etcd = require('node-etcd');
var etcd = new Etcd();
etcd.set("key", "value");
etcd.get("key", console.log);

Callbacks follows the default (error, result) nodejs convention:

function callback(err, res) {
    console.log("Error: ", err);
    console.log("Return: ", res);
}
etcd.get("key", callback);
// Error: null
// Return: { action: 'get', node: { key: '/key', value: 'value', modifiedIndex: 4, createdIndex: 4 } }

Methods

Etcd(hosts = ['127.0.0.1:2379'], [options])

Create a new etcd client for a single host etcd setup

etcd = new Etcd();
etcd = new Etcd("127.0.0.1:2379");
etcd = new Etcd("http://127.0.0.1:2379");
etcd = new Etcd("https://127.0.0.1:2379");
etcd = new Etcd(["http://127.0.0.1:2379"]);

Etcd(hosts, [options])

Create a new etcd client for a clustered etcd setup. Client will connect to servers in random order. On failure it will try the next server. When all servers have failed it will callback with error. If it suspects the cluster is in leader election mode it will retry up to 3 times with exp backoff. Number of retries can be controlled by adding { maxRetries: x } as an option to requests.

etcd = new Etcd(['127.0.0.1:2379','192.168.1.1:2379']);
etcd = new Etcd(['http://127.0.0.1:2379','http://192.168.1.1:2379']);

.set(key, value = null, [options], [callback])

Set key to value, or create key/directory.

etcd.set("key");
etcd.set("key", "value");
etcd.set("key", "value", console.log);
etcd.set("key", "value", { ttl: 60 }, console.log);
etcd.set("key", "value", { maxRetries: 3 }, console.log);

Available options include:

• ttl (time to live in seconds)
• prevValue (previous value, for compare and swap)
• prevExist (existance test, for compare and swap)
• prevIndex (previous index, for compare and swap)

Will create a directory when used without value (value=null): etcd.set("directory/");

.compareAndSwap(key, value, oldvalue, [options], [callback])

Convenience method for test and set (set with {prevValue: oldvalue})

etcd.compareAndSwap("key", "newvalue", "oldvalue");
etcd.compareAndSwap("key", "newValue", "oldValue", options, console.log);

Alias: .testAndSet()

.get(key, [options], [callback])

Get a key or path.

etcd.get("key", console.log);
etcd.get("key", { recursive: true }, console.log);

Available options include:

• recursive (bool, list all values in directory recursively)
• wait (bool, wait for changes to key)
• waitIndex (wait for changes after given index)

.del(key, [options], [callback])

Delete a key or path

etcd.del("key");
etcd.del("key", console.log);
etcd.del("key/", { recursive: true }, console.log);

Available options include:

• recursive (bool, delete recursively)

Alias: .delete()

.compareAndDelete(key, oldvalue, [options], [callback])

Convenience method for test and delete (delete with {prevValue: oldvalue})

etcd.compareAndDelete("key", "oldvalue");
etcd.compareAndDelete("key", "oldValue", options, console.log);

Alias: .testAndDelete()

.mkdir(dir, [options], [callback])

Create a directory

etcd.mkdir("dir");
etcd.mkdir("dir", console.log);
etcd.mkdir("dir/", options, console.log);

.rmdir(dir, [options], [callback])

Remove a directory

etcd.rmdir("dir");
etcd.rmdir("dir", console.log);
etcd.rmdir("dir/", { recursive: true }, console.log);

Available options include:

• recursive (bool, delete recursively)

.create(path, value, [options], [callback])

Atomically create in-order keys.

etcd.create("queue", "first")
etcd.create("queue", "next", console.log)

Alias: .post()

.watch(key, [options], [callback])

This is a convenience method for get with {wait: true}.

etcd.watch("key");
etcd.watch("key", console.log);

The watch command is pretty low level, it does not handle reconnects or timeouts (Etcd will disconnect you after 5 minutes). Use the .watcher() below if you do not wish to handle this yourself.

.watchIndex(key, index, [options], callback)

This is a convenience method for get with {wait: true, waitIndex: index}.

etcd.watchIndex("key", 7, console.log);

See .watch() above.

.watcher(key, [index], [options])

Returns an eventemitter for watching for changes on a key

watcher = etcd.watcher("key");
watcher.on("change", console.log); // Triggers on all changes
watcher.on("set", console.log);    // Triggers on specific changes (set ops)
watcher.on("delete", console.log); // Triggers on delete.
watcher2 = etcd.watcher("key", null, {recursive: true});
watcher2.on("error", console.log);

You can cancel a watcher by calling .stop().

Signals:

• change - emitted on value change
• reconnect - emitted on reconnect
• error - emitted on invalid content
• <etcd action> - the etcd action that triggered the watcher (ex: set, delete).
• stop - watcher was canceled.
• resync - watcher lost sync (server cleared and outdated the index).

It will handle reconnects and timeouts for you, it will also resync (best effort) if it loses sync with Etcd (Etcd only keeps 1000 items in its event history - for high frequency setups it's possible to fall behind).

Use the .watch() command in you need more direct control.

.raw(method, key, value, options, callback)

Bypass the API and do raw queries. Method must be one of: PUT, GET, POST, PATCH, DELETE

etcd.raw("GET", "v2/stats/leader", null, {}, callback)
etcd.raw("PUT", "v2/keys/key", "value", {}, callback)

Remember to provide the full path, without any leading '/'

.machines(callback)

Returns information about etcd nodes in the cluster

etcd.machines(console.log);

.leader(callback)

Return the leader in the cluster

etcd.leader(console.log);

.leaderStats(callback)

Return statistics about cluster leader

etcd.leaderStats(console.log);

.selfStats(callback)

Return statistics about connected etcd node

etcd.selfStats(console.log);

Synchronous operations

The library supports a set of basic synchronous/blocking operations that can be useful during program startup (used like fs.readFileSync).

Synchronous functions perform the etcd request immediately (blocking) and return the following:

{
  err // Error message or null if request completed successfully
  body // Body of the message or null if error
  headers // Headers from the response
}

.setSync(key, value = null, [options])

Synchronously set key to value, or create key/directory.

etcd.setSync("key");
etcd.setSync("key", "value");
etcd.setSync("key", "value", { ttl: 60 });
etcd.setSync("key", "value", { maxRetries: 3 });

Same options and function as .set().

.getSync(key, [options])

Get a key or path.

etcd.getSync("key");
etcd.getSync("key", { recursive: true });

.delSync(key, [options])

Synchronously delete a key or path

etcd.delSync("key");
etcd.delSync("key/", { recursive: true });

The available options are the same as .del() above.

.mkdirSync(dir, [options])

Synchronously create a directory

etcd.mkdirSync("dir");
etcd.mkdirSync("dir/", options);

.rmdirSync(dir, [options])

Synchronously remove a directory

etcd.rmdirSync("dir");
etcd.rmdirSync("dir/", { recursive: true });

The available options are the same as .rmdir() above.

Aborting a request

All async requests will return a cancellation token, to abort a request, do the following:

var token = etcd.get("key", console.log);
token.abort() // also aliased as token.cancel()

console.log("Request is cancelled: ", token.isAborted());

Note that there are no guarantees that aborted write operations won't have affected server state before they were aborted. They only guarantee here is that you won't get any callbacks from the request after calling .abort().

SSL support

Provide ca, cert, key as options. Remember to use https-url.

var fs = require('fs');

var options = {
    ca:   fs.readFileSync('ca.pem'),
    cert: fs.readFileSync('cert.pem'),
    key:  fs.readFileSync('key.pem')
};

var etcdssl = new Etcd("https://localhost:2379", options);

Basic auth

Pass a hash containing username and password as auth option to use basic auth.

var auth = {
    user: "username",
    pass: "password"
};

var etcd = new Etcd("localhost:2379", { auth: auth });

Constructor options

Pass in a hash after server in the constructor to set options. Some useful constructor options include:

• timeout - Integer request timeout in milliseconds to wait for server response.
• ca - Ca certificate
• cert - Client certificate
• key - Client key
• passphrase - Client key passphrase
• auth - A hash containing {user: "username", pass: "password"} for basic auth.

var etcd = new Etcd("127.0.0.1:2379", { timeout: 1000, ... });'

Debugging

Nodejs console.log/console.dir truncates output to a couple of levels - often hiding nested errors. Use util.inspect to show inner errors.

etcd.get('key', function(err, val) {
    console.log(require('util').inspect(err, true, 10));
});

//{ [Error: All servers returned error]
//  [stack]: [Getter/Setter],
//  [message]: 'All servers returned error',
//  errors:
//   [ { server: 'https://localhost:2379',
//       httperror:
//        { [Error: Hostname/IP doesn't match certificate's altnames: "Host: localhost. is not cert's CN: undefined"]
//          [stack]: [Getter/Setter],
//          [message]: 'Hostname/IP doesn\'t match certificate\'s altnames: "Host: localhost. is not cert\'s CN: undefined"',
//          reason: 'Host: localhost. is not cert\'s CN: undefined',
//          host: 'localhost.',
//          cert:
//           { subject: { C: 'USA', O: 'etcd-ca', OU: 'CA' },
//             issuer: { C: 'USA', O: 'etcd-ca', OU: 'CA' } } },
//       httpstatus: undefined,
//       httpbody: undefined,
//       response: undefined,
//       timestamp: Sun Dec 27 2015 23:05:15 GMT+0100 (CET) },
//     [length]: 1 ],
//  retries: 0 }

FAQ:

• Are there any order of execution guarantees when doing multiple requests without using callbacks?
  • No, order of execution is up to NodeJS and the network. Requests run from a connection pool, meaning that if one request is delayed for some reason they'll arrive at the server out of order. Use callbacks (and maybe even a nice async callback handling library for convenient syntax) if ordering is important to prevent race conditions.