New to KubeDB? Please start here.
Supported Etcd Features
| Features | Availability |
|---|---|
| Clustered (single Raft cluster) | ✓ |
Single member cluster (replicas: 1) | ✓ |
Authentication & Authorization (etcd RBAC, root user) | ✓ |
| Custom Configuration (tuning knobs only) | ✓ |
| Externally manageable Auth Secret | ✓ |
| Reconfigurable Health Checker | ✓ |
| TLS: Add, Remove, Update, Rotate ( Cert Manager ) | ✓ |
| Automated Version update | ✓ |
| Automatic Vertical Scaling | ✓ |
| Automated Horizontal Scaling (learner add & promote) | ✓ |
| Automated Volume Expansion | ✓ |
| Storage Class Migration | ✓ |
| Move Raft Leadership / Defragment / Compact | ✓ |
| Backup/Recovery: Instant, Scheduled (KubeStash) | ✓ |
| In-place restore into an existing cluster | ✓ |
| Recovery from a permanent Raft quorum loss | ✓ |
| Persistent Volume | ✓ |
| Builtin Prometheus Discovery | ✓ |
| Using Prometheus operator | ✓ |
| Autoscaler (compute & storage) | ✓ |
| Recommendation Engine | ✓ |
| GitOps | ✓ |
| Grafana Dashboards | ✗ |
| Continuous Archiving / Point-In-Time Recovery | ✗ |
| Multi-cluster / Disaster Recovery topology | ✗ |
A few notes on the table above:
- Custom Configuration for etcd is limited to the typed tuning knobs exposed through
spec.configuration.tuning(quotaBackendBytes,autoCompactionMode,autoCompactionRetention,snapshotCount). etcd’s--config-fileis mutually exclusive with the individual command line flags KubeDB has to set for cluster bootstrap, so a free-form config file is intentionally not supported. Anything the tuning knobs do not cover can still be passed as a raw etcd flag throughspec.podTemplate.spec.containers[name=etcd].args, which is appended after the operator’s own flags. See Etcd CRD and Extra etcd flags. - Backup/Recovery is snapshot based only. etcd has no WAL-shipping style continuous archiving primitive that can be streamed out of the cluster, so there is no point-in-time recovery between two full snapshots. See EtcdArchiver.
- Monitoring does not deploy an exporter sidecar. etcd serves its own Prometheus metrics natively, so KubeDB points the stats Service and the ServiceMonitor straight at etcd.
- No Grafana dashboards ship with KubeDB for etcd yet.
Life Cycle of an Etcd Object
An Etcd object goes through the following phases, reported in status.phase:
| Phase | Meaning |
|---|---|
Provisioning | The KubeDB operator has accepted the object and is creating the offshoot resources (PetSet, Services, Secrets, RBAC). |
Ready | All members are up, the Raft cluster has quorum and the client endpoint is accepting connections. |
Critical | The client endpoint is reachable, but not every member is ready (for example a learner is still catching up). |
NotReady | The client endpoint is not reachable — quorum has been lost or every member is down. |
Halted | spec.halted is true: every offshoot resource except the PVCs has been deleted. |
Unknown | The phase could not be computed. |
In the normal path, an Etcd object moves Provisioning → Ready. When the cluster is
bootstrapped from a snapshot (spec.init.archiver), the operator restores the ordinal-0 volume
before the first member is ever started, so the object simply stays in Provisioning for longer;
the restore’s own outcome is reported on the SuccessfullyDataRestored condition rather than as a
phase.
The phase is derived from the object’s conditions — ProvisioningStarted, ReplicaReady,
AcceptingConnection, Ready and Provisioned. The health checker keeps them up to date by
calling etcd’s own Status() and MemberList() RPCs on the client port and treating the cluster
as healthy while at least N/2+1 members answer.
User Guide
- Quickstart Etcd with KubeDB Operator.
- Secure your cluster with TLS/SSL and reconfigure it later with ReconfigureTLS.
- Tune etcd with custom configuration.
- Scale your cluster horizontally or vertically.
- Backup & Restore etcd using KubeStash.
- Restore a snapshot in place into an
Etcdcluster that already exists. - Recover from a permanent quorum loss when a majority of members is gone for good.
- Rotate the authentication credential of the etcd
rootuser. - Restart the members with a leader-aware rolling restart.
- Autoscale compute and storage resources.
- Detail Concept of Etcd Object.
- Detail Concept of EtcdVersion Object.
- Detail Concept of EtcdOpsRequest Object.
- Detail Concept of EtcdAutoscaler Object.
- Detail Concept of EtcdArchiver Object.
- Detail Concept of AppBinding Object.
Next Steps
- Want to hack on KubeDB? Check our contribution guidelines.