Glossary

Kubernetes Operator

Operational knowledge about one application, written as software that runs inside the cluster and keeps the real deployment matching the declared one. Not the same thing as a Helm chart.

General definition

Kubernetes Operator is the name for a pattern, introduced by CoreOS in 2016, in which the know-how of running one application is encoded as a controller inside the cluster. An operator adds a Custom Resource Definition (CRD), for example MongoDBCommunity or Cluster for Postgres, and a controller that watches those resources and keeps the live cluster matching the declared spec: creating pods and volumes, initialising replication, rotating credentials, taking scheduled backups, failing over when a node dies and performing version upgrades in the right order. The reconcile loop never stops, so drift is corrected as well as created.

Operators are usually graded by capability level: basic install, seamless upgrades, full lifecycle (backup, restore, failover), deep insights and finally auto-pilot. Most well-known ones manage stateful infrastructure: databases (the MongoDB Community and Enterprise operators, CloudNativePG, the Percona operators), message brokers, monitoring stacks (Prometheus Operator) and certificates (cert-manager is itself an operator). They are built with frameworks such as Kubebuilder or the Operator SDK and distributed through OperatorHub and Operator Lifecycle Manager, or simply as a Helm chart that installs the controller.

A Helm chart solves a different problem. Helm is a package manager: it templates Kubernetes manifests from a values file, installs them as a release and can upgrade or roll back that release. It runs at install and upgrade time and knows nothing about the application in between. The practical split:

  • Helm chart: day-one install and configuration, and ongoing upgrades, of stateless services and of stateful ones whose day-two work is done elsewhere
  • Operator: day-two automation for stateful components that need application-aware failover, backup and upgrade logic
  • Both: a chart that deploys the application and points it at databases run by operators or by the cloud provider, the common pattern for business software

The cost of an operator is the operator. It is code with its own release cycle, broad RBAC permissions over the cluster and its own failure modes; a buggy reconcile loop can delete what it manages. For an application whose stateful parts are standard databases, reusing mature database operators or managed services is almost always better than a vendor writing a bespoke one. Canary and rolling upgrades, horizontal scaling of stateless pods and high availability all work with charts alone; what an operator adds is automation of the stateful edge cases.

In the Ethora ecosystem

Ethora ships a Helm chart, not an operator. helm install ethora oci://docker.io/dappros/ethora-core with a root domain and an admin e-mail deploys the API and job workers, the ejabberd XMPP server, the web app, MinIO file storage and the databases, behind one Ingress per host with cert-manager TLS. The chart renders the same configuration as the Docker Compose bundle, so an install is identical on a laptop, a single server or a cluster. First-boot steps (base app, admin account) run as an idempotent Job after every install and upgrade, and helm test runs an end-to-end check through the public URLs.

For the stateful parts the chart offers two routes rather than its own controller. The default runs MongoDB (as a single-node replica set), MySQL, Redis and MinIO as StatefulSets with persistent volumes, which suits evaluation and small deployments. For production, a values file switches those off and points the API and ejabberd at external services: a managed MongoDB replica set, a managed MySQL 8, Redis and S3-compatible storage, or the same databases run by the operators your platform team already trusts. Ethora TechOps can also run the whole stack for you as a dedicated server. The current list of channels and their status is on the Get Ethora page.

Get started

Chat and AI on infrastructure you control

Run Ethora self-hosted or dedicated, with full control of data, keys and residency. Talk to our team.

Start Free
Free tier available Enterprise SLA No vendor lock-in