Skip to main content
Version: 4.0

Khepri FAQ

I see the khepri_db feature flag is marked as experimental in the RabbitMQ code. Is it supported?​

Yes, starting from RabbitMQ 4.0, the RabbitMQ team will support nodes using Khepri. Likewise, an upgrade path will be provided to move to future versions.

Khepri in RabbitMQ 3.13.x is unsupported however. It is not possible to upgrade from 3.13.x to 4.x and beyond if Khepri was enabled on the 3.13.x node(s). You will have to use a blue-green deployment to move everything to another node or cluster.

I see the khepri_db feature flag is marked as experimental in the RabbitMQ code. Will it be possible to upgrade to future releases?​

Yes; see the previous question and answer.

Why do I see "Starting Khepri-based RabbitMQ metadata store" in logs even though I did not enable khepri_db?​

The Khepri metadata store is always initialized even if RabbitMQ does not use it and instead continue to use Mnesia as its backend.

The reason is that the khepri_db feature flag can be enabled from any node. That node must be able to communicate with Khepri on all nodes in the cluster. Therefore Khepri is started everywhere to be ready when it is enabled in the future.

In the meantime, it will not use resources or interfere with the operations of the node or the cluster.

Why do I see "Khepri-based RabbitMQ metadata store ready" in logs even though I did not enable khepri_db?​

See the previous question and answer.

Is it possible to disable Khepri?​

No, once the migration is performed, going back to Mnesia is unsupported.

An alternative is to use a blue-green deployment to move everything to another node or cluster and decommission the Khepri-enabled nodes.

Is it possible to take definitions from a Mnesia-based node and import them in a Khepri-based one?​

Yes. The definitions are metadata-store-agnostic and can be imported in a RabbitMQ node regarless of the metadata store backend used by the node used to export the definitions of the one where they are imported.