kubernetes-sigs / etcdadm

etcdadm

etcdadm is a command-line tool for operating an etcd cluster. It makes it easy to create a new cluster, add a member to, or remove a member from an existing cluster. Its user experience is inspired by kubeadm.

For major project goals, see the roadmap

Table of Contents

• Getting Started
  • Building
  • Creating a new cluster
  • Adding a member
  • Removing a member
• Advanced Usage
  • Creating a new cluster from a snapshot
• Caveats & Limitations
• Design
• Questions

Getting Started

Building

1. Clone the git repository.
2. Build on the host:

   make etcdadm
   

3. Build in a container, using docker:

   make container-build
   

   If you are in a country/region where Internet access is restricted and you cannot download Go dependencies successfully, make with the GOPROXY environment variable set to a third party you trust, for example:

   GOPROXY=https://goproxy.io make container-build
   

   goproxy.io is not endorsed specifically by the project.

Creating a new cluster

1. Copy etcdadm to each machine that will become a member.
2. Choose one machine and run

etcdadm init

Adding a member

1. Copy the CA certificate and key from any machine in the cluster to the machine being added.

rsync -avR /etc/etcd/pki/ca.* <Member IP address>:/

2. Choose a cluster endpoint (i.e. client URL of some member) and run

etcdadm join <endpoint>

Removing a member

On the machine being removed, run

etcdadm reset

Advanced Usage

Creating a new cluster from a snapshot

If you have an existing etcd snapshot, you can use it to create a new cluster:

etcdadm init --snapshot /path/to/etcd.snapshot

Caveats and Limitations

1. Must run as root. (This is because etcdadm creates a systemd service)
2. Does not support etcd v2.
3. Currently tested on Container Linux, with plans for other platforms.

Design

The goal of etcdadm is to make it easy to operate an etcd cluster. It downloads a specific etcd release, installs the binary, configures a systemd service, generates certificates, calls the etcd API to add (or remove) a member, and verifies that the new member is healthy.

Etcdadm must be run on the machine that is being added or removed. As a consequence, if a member permanently fails, and the operator cannot invoke etcdadm reset on that machine, the operator must use the etcd API to delete the failed member from the list of members.

On its own, etcdadm does not automate cluster operation, but a cluster orchestrator can delegate all the above tasks to etcdadm.

Sequence Diagrams

See docs/diagrams for sequence diagrams of init, join, and reset.

Questions?

For more information reach out to etcdadm slack channel