Backup and restore

Octane stores state in three different places, and they restore differently. Knowing which is which is the difference between a restore that works and one that produces an index nothing can open.

What	Where it lives	In a snapshot?
Accelerated index data	The index's own segments	Yes, with the index
Graph edges, vocabularies	Ordinary fields in ordinary indices	Yes, with the index
The license and the trial record	Cluster state	Yes
Named compliance policies	Cluster state	Yes
Controller policies, drains, leases, restart plans	.octane-* indices	Yes, when matched
Ingest and search pipelines	Cluster state	Yes
The graph traversal scratch directory	Node-local disk, outside any commit	No, and it does not need to be

Install the plugins before you restore

An accelerated index cannot be opened without octane-accelerator. The codec is recorded in every segment, so a restore onto a cluster that lacks the plugin produces shards that will not open — the same one-way door described under removing the plugin, reached from the other direction.

The same applies to cluster state. octane-license registers the license custom and octane-content registers the compliance-policy custom; restoring a snapshot that carries them onto a cluster without those plugins means the state cannot be deserialized.

So the order is: install the plugins, start the cluster, then restore. In the install order described in Install — octane-license first, because the others declare it as an extended plugin and a node without it does not start.

The license travels with the snapshot

The license lives in cluster state, so it is written into snapshots alongside it, and the trial record travels with it. Restoring a snapshot onto a different cluster carries the record that a trial was started — which is intended, and is what stops a trial being renewed indefinitely by snapshotting and restoring somewhere new. See A trial travels with a snapshot.

If the restored cluster is a genuinely separate deployment, it needs its own license. It will not get a second trial.

Named compliance policies travel the same way. A restored cluster redacts according to the policies the source cluster had, which is usually what you want and is worth checking when it is not.

The controller's indices

The five .octane-* indices are hidden, so a wildcard that does not include hidden indices will skip them. Decide deliberately:

• Include them and a restore brings back autoscale policies — useful when rebuilding the same cluster.
• Exclude them when restoring into a different cluster. Drain records and leases describe nodes that do not exist there, and a restored drain record refers to an allocation exclusion that was never written.

If you restore drains or leases by accident, the safe move is to delete those two indices and let the controller recreate them; it treats a missing index as no work in progress. Policies are the only one of the five worth keeping across clusters.

Snapshotting while the controller is running

A snapshot in progress makes the safety service return Wait rather than Unsafe — the controller pauses drains until the snapshot finishes rather than fighting it. No coordination is required, but a drain will not start while a snapshot runs, and a long snapshot delays scaling.

Stopping the controller before a snapshot is not necessary. If you do stop it mid-drain, the drain record survives in .octane-drains and the controller resumes from it — see Troubleshooting.

A restore checklist

1. Install octane-license, then the plugins the source cluster had.
2. Confirm the accelerator version matches what the segments expect — an accelerated index restored onto a node with an incompatible accelerator will not open. See Compatibility.
3. Restore, excluding .octane-drains and .octane-drain-leases if this is a different cluster.
4. Check the license: GET /_lucenia/license. A restored trial record may mean this cluster needs a license immediately.
5. Start the controller last, so it observes a settled cluster.