New storage in Keycloak

July 27 2022 by Hynek Mlnařík

The current store in Keycloak has some known limitations. For small deployments, it takes too long to initialize the database and start Keycloak. There is no native support for cloud-native deployment. Upgrading to a new Keycloak version often means a requirement for stop-the-world updates with little chance for no downtime upon upgrade.

Keycloak 19 brings in an early preview of the future Keycloak store supporting no-downtime upgrades, per-realm storage, and cloud-nativity from its very inception. This so called map storage stays focused at delivering optimal experience and thus limits its support to Postgres and CockroachDB databases, and Infinispan datastore. The early preview is available in the Quarkus distribution.

This early preview lacks several features necessary for optimal performance and is thus not yet production ready; still we have a plan to deliver those:

Note that the store used in previous versions (now called legacy store) does not go away soon! Since the new store is not yet production ready, it remains the default store in Keycloak 19. Also other databases (MySQL, MariaDB, MS SQL Server, Oracle) are not supported by the new store, these are only supported by the legacy. We do consider second-level support for these databases, based on results of a community survey.

For development and testing purposes, we offer a fast in-memory store called chm. This store is intentionally not capable of zero-downtime upgrade as that is not needed for development. As such, it also offers a quick Keycloak-without-database deployment.

We will be glad to hear your feedback in GitHub Discussions.

Try it out

The following section contains examples of how to configure Keycloak 19 with the new store.

The experimental command line options starting with --storage might change without prior warning, based on the feedback from the community and project needs.

See what Keycloak can do without a database

After downloading, you can try Keycloak in a single-node deployment as simply as running

bin/ start-dev --storage=chm

This starts a Keycloak instance with a simple testing in-memory store which is saved to the local file system across Keycloak restarts but has no support for zero-downtime upgrades or clustering.

Next step: Persist data in a relational database

Requirements: Postgres 14 or CockroachDB 22.1 (or newer)

To keep the data in a database, run the following command:

bin/ start-dev --storage=jpa --db-url=<jdbc-url> --db-username=<username> --db-password=<password>

Reference to Q&A live document.

This feature has known issues in CockroachDB, see this tracker for list.

Alternative next step: Persist data in external Infinispan

Requirements: Infinispan 12.1

If you have a Infinispan up and running, you can use it for storing the data as well. You can achieve it by running the following command:

bin/ start-dev --storage=hotrod --storage-hotrod-host=<host> --storage-hotrod-port=11222 --storage-hotrod-username=<username> --storage-hotrod-password=<password>

Keycloak creates the needed caches in Infinispan upon first start if the caches do not exist already. You are free to create and configure the caches yourself if you prefer to fine-tune those or e.g. want to set up persistence. Refer to Infinispan documentation for available options for cache configuration.

Reference to Q&A live document.

Next step: Store data in separate storages

Eventually it turns out that some of the data should be stored in files or a database, other (e.g. session) data should be stored in Infinispan. Can this be achieved?

Yes! Keycloak storage is divided into the following storage areas: Realms, clients, users, groups, roles, client scopes, authorization services, events, authentication sessions, user/client sessions, login failures, action tokens, and single-use tokens (last two areas to be merged in Keycloak 20).

For more details on this division, please see architecture specification.

Each area maintains its own storage for storing the data, and each area is independent of others. In other words, realms can be served by a database and users by Infinispan.

To have all the session data stored in an external Infinispan, and realm / client / user / group / role / …​ data stored in a relational database, you can issue the following command:

bin/ start-dev --storage=jpa \
  --db-url=<jdbc-url> --db-username=<username> --db-password=<password> \
  --storage-hotrod-host=<host> --storage-hotrod-port=<port> \
  --storage-hotrod-username=<username> --storage-hotrod-password=<password> \
  --storage-area-action-token=hotrod \
  --storage-area-auth-session=hotrod \
  --storage-area-single-use-object=hotrod \

This starts a Keycloak server with all areas being handled by the relational database (--storage=jpa) but those listed in specific --storage-area-…​ options which are set to use external Infinispan protocol. This setup is similar to the legacy store where all the session data are stored in Infinispan, with the exception that Infinispan is not embedded within Keycloak.


We will be happy for you to try the new store and share your feedback!