Koios Automation Platform 2026.11.4 — release notes

Published 2026-11-02 in the private registry ghcr.io/koiosplatform. Catalog: ghcr.io/koiosplatform/koios-operator-catalog@sha256:c756f34972f5ca9e766a47330ce14174bb44f55694f94c954ecedda04d7ec8a0 (see olm-install.md).

Only the gateway operator changes. Every other image is the same as in 2026.11.3, with the same digest.

Reissued on 2026-11-02 with the same version number. The first 2026.11.4 build (catalog 5f164cc7…) could fail reconciling a platform that had previously run Redis in cluster mode (status.redisCluster changed type). Do not use that catalog: point the CatalogSource at the digest above. If a cluster already installed the first build, delete the koios-operator.v2026.11.4[-cluster-scoped] CSV so that OLM installs the reissue; if OLM does not reinstall it, recreate the Subscription (platforms and their data are not affected).

Upgrading from 2026.11.x

Point the koios-operators CatalogSource at the new catalog digest. OLM upgrades the operators in the stable-2026.11 and stable-2026.11-cluster-scoped channels (koios-operator.v2026.11.4 replaces v2026.11.3). Existing platforms keep the standalone Redis; nothing changes until you set redis_mode: cluster.

Console: instance details

The KoiosPlatform details page in the OpenShift console now shows the Redis mode and the Redis cluster state. As with the AnsibleAutomationPlatform resource, the creation form only offers the shared external database (database.database_secret); per-component database Secrets (controller, eda, eda.event_stream and hub .database_secret) are set in YAML.

New: Redis in cluster mode

spec.redis_mode: cluster, as in the AnsibleAutomationPlatform resource, deploys the platform Redis used by the gateway as a Valkey cluster of six nodes (three masters and three replicas). Controller, EDA and automation hub keep their own Redis.

  • Stable node identity. Nodes are announced by host name through a headless Service, and each node keeps its identity in a small volume (1Gi; the cache itself is not persisted). The cluster recovers on its own when all nodes restart with new IP addresses.

  • Default user disabled. Replication uses an internal user. No password appears in command lines or ConfigMaps.

  • Switch modes on an existing platform, in both directions (the Redis data is only a cache). idle_aap scales the cluster down and back up without recreating it.

  • Status: status.redisMode and status.redisCluster (state, knownNodes, masters, slotsAssigned). A cluster that does not reach ok is reported with its cause.

  • External Redis in cluster mode: set redis_mode: cluster (and optionally redis_hosts, host:port pairs separated by commas) in the Secret of redis.redis_secret.

Notes:

  • redis.replicas must stay 1 in both modes.

  • A Redis cannot be shared by two platforms: the gateway key prefix is fixed.

  • Not included: automatic repair of a node that lost its volume (the status message gives the valkey-cli --cluster steps), scaling the number of nodes, and TLS inside the cluster.