Maintained by a former CoreOS engineer and inspired from the etcd-operator designed for Kubernetes, the etcd-cloud-operator automatically bootstraps, monitors, snapshots and recovers etcd clusters on cloud providers.
Used in place of the etcd binary and with minimal configuration, the operator handles the configuration and lifecycle of etcd, based on data gathered from the cloud provider and the status of the etcd cluster itself.
In other words, the operator operator is meant to help human operators sleep at night, while their mysterious etcd data store keeps running safely, even in the event of process, instance, network, or even availability-zone wide failures.
-
Resize: By abstracting cluster management, resizing the cluster becomes straightforward as the underlying auto-scaling group can simply be scaled as desired.
-
Snapshots: Periodically, snapshots of the entire key-value space are captured, from each of the etcd members and uploaded to an encrypted external storage, allowing the etcd (or human) operator to restore the store at a later time, in any etcd cluster or instance.
-
Failure recovery: Upon failure of a minority of the etcd members, the managed members automatically restarts and rejoins the cluster without breaking quorum or causing visible downtime - First by simply trying to rejoin with their existing data set, otherwise trying to join as a new member with a clean state, or by replacing the entire instance if necessary.
-
Disaster recovery: In the event of a quorum loss, consequence of the simultaneous failure of a majority of the members, the operator coordinates to snapshot any live members and cleanly stop then, before seeding a new cluster from the latest data revision available once the expected amount of instances are ready to start again.
The operator and etcd cluster can be easily configured using a YAML file. The configuration notably includes clients/peers TLS encryption/authentication, with the ability to automatically generate self-signed certificates if encryption is desired but authentication is not.
Running a managed etcd cluster using the operator is simply a matter of running the operator binary in a supported auto-scaling group (as of today, AWS only).
- AWS: You will need to provide IAM credentials with the following capabilities in the container's environment, scoped to the appropriate instances: "ec2:DescribeInstances" "autoscaling:DescribeAutoScalingGroups" "autoscaling:DescribeAutoScalingInstances"
A Terraform module is available to easily bring up production-grade etcd clusters managed by the the operator out, and integrate them into your infrastructure.