Operate

Operating Koios Automation Platform means maintaining a reconciled, observable platform and checking that its services can perform useful work. The gateway operator reconciles KoiosPlatform and its shared services; the controller, EDA and hub operators reconcile their respective child resources.

Change supported settings in KoiosPlatform, or use the product UI/API for application objects. Do not edit generated Deployments, container configuration files or service registrations by hand: reconciliation can replace them. See Configure the platform and the KoiosPlatform reference.

Components and workloads

The following map uses the default names observed on the test platform on 2026-11-18. <p> is the KoiosPlatform name. The controller, EDA and hub resource names default to <p>-controller, <p>-eda and <p>-hub; spec.controller.name, spec.eda.name and spec.hub.name can select other names, including adopted resources. Read their actual names from status.controller, status.eda and status.hub before constructing workload names.

Component

Default workloads

Responsibility / checks

Platform gateway and proxy

Deployment <p>-gateway

Shared UI/API entry point, authentication and routing; api and proxy containers.

Automation controller

Deployments <p>-controller-web, <p>-controller-task

Controller API/UI backend, scheduling and control tasks. Job execution can use separate container group pods or external execution nodes.

Event-Driven Ansible

Deployments <p>-eda-api, <p>-eda-default-worker, <p>-eda-activation-worker, <p>-eda-event-stream

API, background work, activation management and event streams. Rulebook activations have their own runtime pods/history.

Automation hub

Deployments <p>-hub-api, <p>-hub-content, <p>-hub-worker, <p>-hub-web, <p>-hub-redis

Metadata/API, artifact delivery, imports/background work, web service and the hub’s own Redis/Valkey.

Managed PostgreSQL

StatefulSet <p>-postgres-15 in the verified setup

Shared database service, with component-specific databases/users. The major-version suffix can change; use status.postgresStatefulSet.

Gateway Redis/Valkey

StatefulSet <p>-redis in standalone mode

Gateway cache/messaging. Cluster mode has a different topology; inspect status.redisMode and the actual resources. This is separate from controller sidecars and hub Redis.

Disabled optional components and external databases/Redis change this map. A ready gateway does not establish the health of controller jobs, activations or content imports.

Operator deployments

The installer and OLM bundles name the operator Deployments:

  • koios-gateway-operator-controller-manager

  • koios-controller-operator-controller-manager

  • koios-eda-operator-controller-manager

  • koios-hub-operator-controller-manager

These names come from the installation package’s component prefixes, not from <p>. For example, naming a platform production does not rename the gateway operator to production-gateway-operator-controller-manager.

Discover the operator namespace for your installation: normally openshift-operators for cluster-scoped OpenShift OLM, operators for cluster-scoped Kubernetes OLM, or the platform namespace for a namespaced install. The documented Kustomize package deploys them in koios. One operator deployment can manage several platforms; filter diagnostics for the affected name and namespace. See installation methods.

Status

Use Check platform health for bounded kubectl commands and service endpoints. The operator reports the last reconciliation and component resource names in KoiosPlatform.status.

Condition

Interpretation

Successful

True after a successful reconciliation; in idle mode the platform can still report success while workloads are intentionally scaled down.

Failure

True indicates a reconciliation failure; inspect its reason and message. Wait for Failure=False as well as Successful=True after a change.

Running

A reconciliation lifecycle condition, not a standalone readiness check. It was True with reason Successful on the healthy test platform; do not wait for it to become False.

DeprecatedFields

Warns that deprecated CR fields are in use; migrate them to the fields named in the message.

Status also includes version, URL, gatewayImage, component names, databaseVersions and redisMode. adminUser and adminPasswordSecret identify the administrator and its Secret; they do not contain the password. Inspect only needed fields rather than dumping configuration.

Upgrades

The canonical installation and upgrade guidance is under Install and upgrade and Upgrading and reissued releases. Back up before changing release and read its release notes.

  • With OLM, apply the new release’s catalog and Subscription for the chosen scope/channel from an authorized installer checkout. Check the new operator CSVs and follow any manual approval policy configured for that installation.

  • With Kustomize, apply the new release’s operator package from its installer tag.

  • Verify the expected operator/platform version, Successful=True, Failure=False, ready workloads and service checks. Version alone is not a functional acceptance test.

A release reissued with the same number needs the extra procedure in the reissued-release guide; a same-number reinstall cannot be inferred from a version string. This page does not duplicate its catalog/CSV recovery commands.

Backup and restore

Use Backup and restore for the canonical procedure and confidentiality requirements. After a restore, follow the restored-platform runbook before enabling integrations, activations and schedules that could affect the original platform.

Daily operation

Need

Canonical guide

Platform/resource health

Check platform health

Workload/operator output

Logs and diagnostics

External controller log aggregation

Logging and Aggregation, Controller logfiles

Metrics and scraping

Metrics

Job history/activity retention

Management Jobs

Capacity, resources and scaling

Performance, Instances, Topology Viewer

Controller diagnosis

Troubleshooting Automation Controller

Activation diagnosis

Troubleshooting Automation Decisions

Import/download diagnosis

Troubleshooting Automation Content

Idle mode

Idle mode intentionally scales platform components down while preserving their configuration/data. Account for it when interpreting successful reconciliation, missing running pods and unavailable HTTP endpoints. Resume through the CR as described there, not by manually scaling Deployments.

Deleting a platform

Deletion is different from a restart, idle mode or upgrade. Check backups and retained data first. Use the intended namespace and context for the CR deletion:

kubectl --context '<context>' -n '<platform-namespace>' delete koiosplatform '<platform-name>'

The operator removes the platform and its components. It retains, without a platform owner, the managed PostgreSQL and hub content volumes and the component Secrets needed to reuse the data, including encryption keys and database connections. Creating a platform with the same name in the same namespace can reuse those retained resources; preserve the matching database and keys together. With external databases, the operator does not delete your databases or supplied connection Secrets.

Removing the platform completely is a separate data-destruction operation. Inventory the retained volumes/Secrets and verify backups before planning it; do not use namespace or PVC deletion as a routine troubleshooting step. See Backup and restore.