Configuration

The controller reads one YAML file, resolved in this order:

the first command-line argument
the OCTANE_CONFIG environment variable
/etc/octane/autoscale.yml

A complete file

octane:
  mode: custom_resource          # recommend | kubernetes | custom_resource
  interval_millis: 30000

  clusters:
    prod:
      endpoint: https://opensearch.prod.svc:9200
      authorization: "ApiKey <redacted>"

  kubernetes:
    namespace: search
    custom_resource: clusters.lucenia.io/v1/luceniaclusters/prod
    replica_paths:
      hot:  spec.nodePools.hot.replicas
      warm: spec.nodePools.warm.replicas

  license_path: /etc/octane/license.json
  license_issued_to: acme-corp

Reference

Key	Default	Meaning
`octane.mode`	`recommend`	`recommend`, `kubernetes`, `custom_resource`.
`octane.interval_millis`	`30000`	How often a cycle runs.
`octane.clusters.<id>.endpoint`	—	Required. REST endpoint. At least one cluster.
`octane.clusters.<id>.authorization`	—	Verbatim `Authorization` header value.
`octane.kubernetes.namespace`	—	Required for `kubernetes` and `custom_resource`.
`octane.kubernetes.statefulsets.<tier>`	—	(kubernetes mode) StatefulSet name for a tier.
`octane.kubernetes.custom_resource`	—	(custom_resource mode) The resource to patch.
`octane.kubernetes.replica_paths.<tier>`	—	(custom_resource mode) Dotted path to that tier's replica count.
`octane.license_path`	—	License file. Absent ⇒ recommendation mode only.
`octane.license_issued_to`	`self-serve`	Name the license is checked against.

Multiple clusters

octane.clusters is a map, and one controller can drive several:

octane:
  clusters:
    prod:    { endpoint: "https://prod:9200",    authorization: "ApiKey ..." }
    staging: { endpoint: "https://staging:9200", authorization: "ApiKey ..." }

At least one is required; a file with none is refused at startup.

It refuses to start on a configuration it cannot act on

Every one of these is checked when the file is read, not at the first scale:

Condition	Error
no clusters	`no clusters configured under octane.clusters`
a cluster with no endpoint	`cluster [id] has no endpoint`
`custom_resource` with no `custom_resource` set	`octane.mode is custom_resource but octane.kubernetes.custom_resource is not set`
`custom_resource` with no `replica_paths`	`octane.mode is custom_resource but no octane.kubernetes.replica_paths are configured`
`kubernetes` or `custom_resource` with no `namespace`	`octane.mode is <mode> but octane.kubernetes.namespace is not set`

This is deliberate, and the reason is worth stating: a controller that starts happily and then cannot act at the one moment it is needed has hidden the mistake until it costs something. Failing at startup means the mistake surfaces during deployment, when someone is watching.

The replica_paths check has a second reason. Without a path there is nowhere to write the node count, and guessing one could set a field that means something else entirely.

Secrets

authorization is a header value and belongs in a mounted secret, not in a ConfigMap. The same is true of license_path. Mount both read-only.

Choosing an interval

interval_millis is the cycle period, not a scaling rate. A cycle that finds nothing to do is cheap — a few REST calls — so 30 seconds is a reasonable default. A drain in progress is resumed across cycles rather than restarted, so a short interval does not mean repeating work.

A complete file​

Reference​

Multiple clusters​

It refuses to start on a configuration it cannot act on​

Secrets​

Choosing an interval​