etcd

README version 0.2.0

A highly-available key value store for shared configuration and service discovery. etcd is inspired by zookeeper and doozer, with a focus on:

• Simple: curl'able user facing API (HTTP+JSON)
• Secure: optional SSL client cert authentication
• Fast: benchmarked 1000s of writes/s per instance
• Reliable: Properly distributed using Raft

Etcd is written in Go and uses the raft consensus algorithm to manage a highly-available replicated log.

See etcdctl for a simple command line client. Or feel free to just use curl, as in the examples below.

Getting Started

Getting etcd

The latest release is available as a binary at Github.

Building

You can build etcd from source:

git clone https://github.com/coreos/etcd
cd etcd
./build

This will generate a binary in the base directory called ./etcd.

NOTE: you need go 1.1+. Please check your installation with

go version

Running a single node

These examples will use a single node cluster to show you the basics of the etcd REST API. Lets start etcd:

./etcd -d node0 -n node0

This will bring up an etcd node listening on port 4001 for client communication and on port 7001 for server-to-server communication. The -d node0 argument tells etcd to write node configuration, logs and snapshots to the ./node0/ directory. The -n node0 tells the rest of the cluster that this node is named node0.

Usage

Setting the value to a key

Let’s set the first key-value pair to the node. In this case the key is /message and the value is Hello world.

curl -L http://127.0.0.1:4001/v2/keys/message -XPUT -d value="Hello world"

{"action":"set","key":"/message","value":"Hello world","index":3,"term":0}

This response contains five fields. We will introduce three more fields as we try more commands.

1. The action of the request; we set the value via a PUT request, thus the action is set.

2. The key of the request; we set /message to Hello world!, so the key field is /message. Notice we use a file system like structure to represent the key-value pairs. So each key starts with /.

3. The current value of the key; we set the value toHello world.

4. Index is the unique internal log index of the set request. Requests that change the log index include set, delete, update, create and compareAndSwap. The get and watch commands do not change state in the store and so they do not change the index. You may notice that in this example the index is 3, although it is the first request you sent to the server. This is because there are internal commands that also change the state like adding and syncing servers.

Get the value of a key

Get the value that we just set in /message by issuing a GET:

curl -L http://127.0.0.1:4001/v2/keys/message

{"action":"get","key":"/message","value":"Hello world","index":3,"term":0}

Changing the value of a key

Change the value of /message from Hello world to Hello etcd with another PUT request to the key:

curl -L http://127.0.0.1:4001/v1/keys/message -XPUT -d value="Hello etcd"

{"action":"set","key":"/message","prevValue":"Hello world","value":"Hello etcd","index":4,"term":0}

Notice that the prevValue is set to the previous value of the key - Hello world. It is useful when you want to atomically set a value to a key and get its old value.

Deleting a key

Remove the /message key with a DELETE:

curl -L http://127.0.0.1:4001/v2/keys/message -XDELETE

{"action":"delete","key":"/message","prevValue":"Hello etcd","index":5,"term":0}

Using key TTL

Keys in etcd can be set to expire after a specified number of seconds. That is done by setting a TTL (time to live) on the key when you POST:

curl -L http://127.0.0.1:4001/v2/keys/foo -XPUT -d value=bar -d ttl=5

{"action":"set","key":"/foo","value":"bar","expiration":"2013-10-19T18:44:04.528757176-07:00","ttl":5,"index":6,"term":0}

Note the last two new fields in response:

1. The expiration is the time that this key will expire and be deleted.

2. The ttl is the time to live of the key.

Now you can try to get the key by sending:

curl -L http://127.0.0.1:4001/v2/keys/foo

If the TTL has expired, the key will be deleted, and you will be returned a 100.

{"errorCode":100,"message":"Key Not Found","cause":"/foo","index":6,"term":0}

Waiting for a change

We can watch for a change and get a notification at a given path or any keys underneath it.

In one terminal, we send a get request with wait=true :

curl -L http://127.0.0.1:4001/v2/keys/foo?wait=true

Now, we are waiting for any changes at path /foo.

In another terminal, we set a key /foo with value bar:

curl -L http://127.0.0.1:4001/v2/keys/foo -XPUT -d value=bar

The first terminal should get the notification and return with the same response as the set request.

{"action":"set","key":"/foo","value":"bar","index":7,"term":0}

However, the watch command can do more than this. Using the the index we can watch for commands that has happened in the past. This is useful for ensuring you don't miss events between watch commands.

Let's try to watch for the set command of index 7 again:

curl -L http://127.0.0.1:4001/v2/keys/foo?wait=true\&waitIndex=7

The watch command returns immediately with the same response as previous.

Atomic Compare and Swap

Etcd can be used as a centralized coordination service in a cluster and CompareAndSwap is the most basic operation to build distributed lock service.

This command will set the value to the key only if the client provided conditions are equal to the current conditions.

The current comparable conditions are:

1. prevValue previous value of the key:

2. prevIndex previous index of the key

3. prevExist previous existence of the key: if prevExist is true, it is a update request; if prevExist is false, it is a create request.

Here is a simple example. Let's create a key-value pair first: foo=one.

curl -L http://127.0.0.1:4001/v1/keys/foo -XPUT -d value=one

Let's try an invalid CompareAndSwap command. We can give another parameter prevValue to set command to make it a CompareAndSwap command.

curl -L http://127.0.0.1:4001/v1/keys/foo?prevValue=two -XPUT -d value=three

This will try to compare the previous value of the key and the previous value we provided. If they are equal, the value of the key will change to three.

{"errorCode":101,"message":"Test Failed","cause":"[two != one] [0 != 8]","index":9,"term":0}

which means compareAndSwap failed.

Let us try a valid one.

curl -L http://127.0.0.1:4001/v2/keys/foo?prevValue=one -XPUT -d value=two

The response should be

{"action":"compareAndSwap","key":"/foo","prevValue":"one","value":"two","index":10,"term":0}

We successfully changed the value from “one” to “two”, since we give the correct previous value.

Listing a directory

Let us create some keys first.

We already have /foo=two

We create another one /foo_dir/foo=bar

curl -L http://127.0.0.1:4001/v2/keys/foo_dir/foo -XPUT -d value=bar

{"action":"set","key":"/foo_dir/foo","value":"bar","index":11,"term":0}

Now list the keys under root /

curl -L http://127.0.0.1:4001/v2/keys/

We should see the response as an array of items

{"action":"get","key":"/","dir":true,"kvs":[{"key":"/foo","value":"two"},{"key":"/foo_dir","dir":true}],"index":11,"term":0}

which meas /foo=two is a key-value pair under / and /foo_dir` is a directory.

Also we can recursively get all the content under a directory by add recursive=true.

curl -L http://127.0.0.1:4001/v2/keys/?recursive=true

{"action":"get","key":"/","dir":true,"kvs":[{"key":"/foo","value":"two"},{"key":"/foo_dir","dir":true,"kvs":[{"key":"/foo_dir/foo","value":"bar"}]}],"index":11,"term":0}

Deleting a directory

Let try to delete the directory /foo_dir.

To delete a directory, we must add recursive=true.

curl -L http://127.0.0.1:4001/v2/keys/foo_dir?recursive=true -XDELETE

{"action":"delete","key":"/foo_dir","dir":true,"index":12,"term":0}

Creating a hidden node

We can create a hidden key-value pair or directory by add _ prefix. The hidden item will not be list when using get for a directory.

curl -L http://127.0.0.1:4001/v2/keys/_message -XPUT -d value="Hello hidden world"

{"action":"set","key":"/_message","value":"Hello hidden world","index":13,"term":0}

curl -L http://127.0.0.1:4001/v2/keys/message -XPUT -d value="Hello world"

{"action":"set","key":"/message","value":"Hello world","index":14,"term":0}

Let us try to get the root /

curl -L http://127.0.0.1:4001/v2/keys/

{"action":"get","key":"/","dir":true,"kvs":[{"key":"/foo","value":"two"},{"key":"/message","value":"Hello world"}],"index":15,"term":0}

We can only get /message, but cannot get /_message.

Advanced Usage

Transport security with HTTPS

Etcd supports SSL/TLS and client cert authentication for clients to server, as well as server to server communication

First, you need to have a CA cert clientCA.crt and signed key pair client.crt, client.key. This site has a good reference for how to generate self-signed key pairs: http://www.g-loaded.eu/2005/11/10/be-your-own-ca/

For testing you can use the certificates in the fixtures/ca directory.

Next, lets configure etcd to use this keypair:

./etcd -n node0 -d node0 -clientCert=./fixtures/ca/server.crt -clientKey=./fixtures/ca/server.key.insecure -f

-f forces new node configuration if existing configuration is found (WARNING: data loss!) -clientCert and -clientKey are the key and cert for transport layer security between client and server

You can now test the configuration using https:

curl --cacert fixtures/ca/ca.crt https://127.0.0.1:4001/v2/keys/foo -XPUT -d value=bar -v

You should be able to see the handshake succeed.

...
SSLv3, TLS handshake, Finished (20):
...

And also the response from the etcd server.

{"action":"set","key":"/foo","value":"bar","index":3, "term: 0"}

Authentication with HTTPS client certificates

We can also do authentication using CA certs. The clients will provide their cert to the server and the server will check whether the cert is signed by the CA and decide whether to serve the request.

./etcd -n node0 -d node0 -clientCAFile=./fixtures/ca/ca.crt -clientCert=./fixtures/ca/server.crt -clientKey=./fixtures/ca/server.key.insecure -f

-clientCAFile is the path to the CA cert.

Try the same request to this server:

curl --cacert fixtures/ca/ca.crt https://127.0.0.1:4001/v2/keys/foo -XPUT -d value=bar -v

The request should be rejected by the server.

...
routines:SSL3_READ_BYTES:sslv3 alert bad certificate
...

We need to give the CA signed cert to the server.

curl -L https://127.0.0.1:4001/v1/keys/foo -XPUT -d value=bar -v --key myclient.key --cert myclient.crt -cacert clientCA.crt

You should able to see

...
SSLv3, TLS handshake, CERT verify (15):
...
TLS handshake, Finished (20)

And also the response from the server:

{"action":"set","key":"/foo","value":"bar","index":3,"term:0"}

Clustering

Example cluster of three machines

Let's explore the use of etcd clustering. We use go-raft as the underlying distributed protocol which provides consistency and persistence of the data across all of the etcd instances.

Let start by creating 3 new etcd instances.

We use -s to specify server port and -c to specify client port and -d to specify the directory to store the log and info of the node in the cluster

./etcd -s 127.0.0.1:7001 -c 127.0.0.1:4001 -d nodes/node1 -n node1

Note: If you want to run etcd on external IP address and still have access locally you need to add -cl 0.0.0.0 so that it will listen on both external and localhost addresses. A similar argument -sl is used to setup the listening address for the server port.

Let the join two more nodes to this cluster using the -C argument:

./etcd -c 127.0.0.1:4002 -s 127.0.0.1:7002 -C 127.0.0.1:7001 -d nodes/node2 -n node2
./etcd -c 127.0.0.1:4003 -s 127.0.0.1:7003 -C 127.0.0.1:7001 -d nodes/node3 -n node3

Get the machines in the cluster:

curl -L http://127.0.0.1:4001/v1/machines

We should see there are three nodes in the cluster

http://127.0.0.1:4001, http://127.0.0.1:4002, http://127.0.0.1:4003

The machine list is also available via this API:

curl -L http://127.0.0.1:4001/v1/keys/_etcd/machines

[{"action":"get","key":"/_etcd/machines/node1","value":"raft=http://127.0.0.1:7001&etcd=http://127.0.0.1:4001&raftVersion=v0.1.1-311-g91cad59","index":4},{"action":"get","key":"/_etcd/machines/node2","value":"raft=http://127.0.0.1:7002&etcd=http://127.0.0.1:4002&raftVersion=v0.1.1-311-g91cad59","index":4},{"action":"get","key":"/_etcd/machines/node3","value":"raft=http://127.0.0.1:7003&etcd=http://127.0.0.1:4003&raftVersion=v0.1.1-311-g91cad59","index":4}]

The key of the machine is based on the commit index when it was added. The value of the machine is hostname, raft port and client port.

Also try to get the current leader in the cluster

curl -L http://127.0.0.1:4001/v2/leader

The first server we set up should be the leader, if it has not died during these commands.

http://127.0.0.1:7001

Now we can do normal SET and GET operations on keys as we explored earlier.

curl -L http://127.0.0.1:4001/v2/keys/foo -XPUT -d value=bar

{"action":"set","key":"/foo","value":"bar","index":5,"term:0"}

Killing Nodes in the Cluster

Let's kill the leader of the cluster and get the value from the other machine:

curl -L http://127.0.0.1:4002/v1/keys/foo

A new leader should have been elected.

curl -L http://127.0.0.1:4001/v1/leader

http://127.0.0.1:7002

or

http://127.0.0.1:7003

You should be able to see this:

{"action":"get","key":"/foo","value":"bar","index":5,"term:1"}

It succeeded!

Testing Persistence

OK. Next let us kill all the nodes to test persistence. And restart all the nodes use the same command as before.

Your request for the foo key will return the correct value:

curl -L http://127.0.0.1:4002/v1/keys/foo

{"action":"GET","key":"/foo","value":"bar","index":5}

Using HTTPS between servers

In the previous example we showed how to use SSL client certs for client to server communication. Etcd can also do internal server to server communication using SSL client certs. To do this just change the -client* flags to -server*.

If you are using SSL for server to server communication, you must use it on all instances of etcd.

Contributing

See CONTRIBUTING for details on submitting patches and contacting developers via IRC and mailing lists.

Libraries and Tools

Tools

• etcdctl - A command line client for etcd

Go libraries

• go-etcd

Java libraries

• justinsb/jetcd
• diwakergupta/jetcd

Python libraries

• transitorykris/etcd-py
• jplana/python-etcd
• russellhaering/txetcd - a Twisted Python library

Node libraries

• stianeikeland/node-etcd

Ruby libraries

• iconara/etcd-rb
• jpfuentes2/etcd-ruby
• ranjib/etcd-ruby

C libraries

• jdarcy/etcd-api

Chef Integration

• coderanger/etcd-chef

Chef Cookbook

• spheromak/etcd-cookbook

Projects using etcd

• binocarlos/yoda - etcd + ZeroMQ
• calavera/active-proxy - HTTP Proxy configured with etcd
• derekchiang/etcdplus - A set of distributed synchronization primitives built upon etcd
• gleicon/goreman - Branch of the Go Foreman clone with etcd support
• garethr/hiera-etcd - Puppet hiera backend using etcd
• mattn/etcd-vim - SET and GET keys from inside vim
• mattn/etcdenv - "env" shebang with etcd integration

FAQ

What size cluster should I use?

Every command the client sends to the master is broadcast to all of the followers. But, the command is not committed until the majority of the cluster machines receive that command.

Because of this majority voting property the ideal cluster should be kept small to keep speed up and be made up of an odd number of machines.

Odd numbers are good because if you have 8 machines the majority will be 5 and if you have 9 machines the majority with be 5. The result is that an 8 machine cluster can tolerate 3 machine failures and a 9 machine cluster can tolerate 4 nodes failures. And in the best case when all 9 machines are responding the cluster will perform at the speed of the fastest 5 nodes.

Project Details

Versioning

etcd uses semantic versioning. When we release v1.0.0 of etcd we will promise not to break the "v1" REST API. New minor versions may add additional features to the API however.

You can get the version of etcd by issuing a request to /version:

curl -L http://127.0.0.1:4001/version

During the v0 series of releases we may break the API as we fix bugs and get feedback.

License

etcd is under the Apache 2.0 license. See the LICENSE file for details.