gDown becomes a systemd service (user gluonify-gdown, data in /var/lib/gluonify-gdown). A first node starts a new cluster of one; the generated administrator password is kept in a file readable by root only.
Run the first command on the first machine. Its cluster token is in /etc/gluonify-gdown/gdown.env (GLUONIFY_GDOWN_AUTH_TOKEN): copy it, then run the last command on each other machine, with --seed pointing to the first node. The token goes through the environment, never as an argument. The new member configures itself from the seed and joins. Three members keep working when one is lost.
Options
- --node-id: name of the member (default: the host name).
- --advertise-host: the address the other members use to reach this one (default: the host name).
- --seed: URL of a member of an existing cluster; leave it out on the first node.
- --data-dir: data directory (default /var/lib/gluonify-gdown).
- --version: release to install (default: the latest).
- --key-sha256: fingerprint of the release key; the script refuses any other key.
Ports
8080 (HTTP: API, web console), 7687 (Bolt) and 9000 (Raft, between members only). Open 9000 between the machines of a cluster, and nothing more than 8080 and 7687 for your clients.
Docker
The public image gluonify/gdown runs the native executable on Red Hat UBI 9 micro (since 0.6.1), for linux/amd64 and linux/arm64: non-root user (uid 10001, or any uid in group 0, as OpenShift assigns), data in the /data volume, read-only root filesystem, healthcheck by the executable itself (no curl in the image). It runs unchanged under OpenShift’s restricted-v2 policy; a 3-member StatefulSet is on this page. It has no default password and refuses to start without the initial administrator password (preferably in a file, a Docker secret, or in the environment) or an OpenID Connect provider. The executable is built by Mandrel 25 (GraalVM Native Image, community edition); the runtime parts it contains are under the GPL version 2 with the Classpath Exception, whose text ships in the image under /licenses.
Windows (PowerShell): the same single member (Docker Desktop; curl.exe and PowerShell come with Windows; the password comes from .NET's random generator, and there is no chmod on Windows)
Under WSL2 (Ubuntu, with Docker Desktop’s WSL integration, working in the Linux home directory, not /mnt/c) the Linux commands above work as they are.
Three members with mutual TLS (Docker Compose)
The Compose example starts three members with mutual TLS on Raft between them, HTTPS for the console and the API, and Bolt over TLS. Two scripts make test certificates and random secrets; use your own certificate authority in production.
Windows (PowerShell): the same three members; gen-certs.ps1 needs PowerShell 7 (winget install Microsoft.PowerShell) for the PKCS#8 keys, gen-secrets.ps1 runs in Windows PowerShell 5.1 or 7
- Secrets: the initial administrator password may be given directly in GLUONIFY_GDOWN_AUTH_LOCAL_INITIAL_ADMIN_PASSWORD, but GLUONIFY_GDOWN_AUTH_LOCAL_INITIAL_ADMIN_PASSWORD_FILE (a file, such as a Docker secret) is recommended, because the environment shows in docker inspect; likewise the service token through GLUONIFY_GDOWN_AUTH_TOKEN_FILE.
- Without TLS between members the cluster still works, but Raft travels in clear text: gDown logs a warning at startup and on each join. Mutual TLS is the recommended setup.
- Volumes: /data (databases and Raft log, keep it) and /var/lib/gdown/native, where the TLS library is unpacked (it must allow execution, unlike a noexec /tmp).
- The images are signed with cosign using the Gluonify key and carry an SBOM and a provenance attestation; upgrades replace one member at a time.
OpenShift and Kubernetes
The image runs under OpenShift’s restricted-v2 policy unchanged: random uid in group 0, no capability, read-only root filesystem. Leave runAsUser unset (OpenShift picks it, and its fsGroup makes the volume writable): no anyuid needed. Create the secret with the administrator password and the service token, then apply the manifest: a headless Service and a 3-member StatefulSet in which gdown-0 bootstraps the cluster and the others join it.
Raft travels in clear text between the members: add the TLS variables of the Compose cluster example to encrypt it. HTTP goes through a Route, Bolt through a LoadBalancer Service (a Route carries HTTP only); never expose port 9000. Verified under Docker with random uids in group 0, not yet on an OpenShift cluster.
Sign in with Microsoft Entra ID
gDown accepts Entra ID access tokens over HTTP, WebSocket and Bolt, and its console signs people in with Entra ID (authorization code with PKCE). Rights come from app roles, in the roles claim.
- Register the API (single tenant) and set accessTokenAcceptedVersion to 2 in its manifest, so that tokens are v2.
- Expose an API: Application ID URI api://<api-client-id> and a scope gdown.access, which clients ask for to get a token meant for gDown.
- Create the app roles gDown.Reader, gDown.Writer and gDown.Admin; assign them to users or groups, and to daemons as application permissions.
- Register the console as a single-page application (PKCE, no secret) with the console URL as redirect URI, and give it the delegated permission gdown.access.
- Mount the configuration below in the image at /opt/gdown/config/application.properties.
What the token must carry
iss = https://login.microsoftonline.com/<tenant-id>/v2.0 (your tenant, v2 token); aud = the client id of the API (a token for another API is refused); roles = the app roles, mapped to reader, writer or admin; tid = your tenant (checked through iss, or against allowed-tenants for several tenants); exp and the Microsoft signature. The scp claim is informative: rights do not depend on it.
Use app roles rather than the groups claim: past 200 groups Entra drops it (overage), and gDown does not call Microsoft Graph. A token stays valid until it expires.
License
Copyright (c) 2026 Gluonify. gDown is released under the Business Source License 1.1: you may copy, modify and redistribute it and use it in production, except to provide a hosted or managed database service to third parties. Four years after each release, that version converts to the Apache License 2.0. For any other use, a commercial license is available. The source code will be published from version 1.0.0, due shortly after the official release of Quarkus 4. Third-party components remain under their own licenses, listed in the NOTICE file shipped with the image (/licenses/NOTICE).
Neo4j, Bolt and Cypher are trademarks of Neo4j, Inc. openCypher is a project of the openCypher Implementers Group (its TCK is under the Apache License 2.0). Oracle, Java and GraalVM are trademarks of Oracle and/or its affiliates. Red Hat, UBI and Mandrel are trademarks of Red Hat, Inc. Docker is a trademark of Docker, Inc. gDown and Gluonify are not affiliated with, sponsored by or endorsed by any of them; "compatible with" describes protocol and language compatibility only.
Community
Questions, ideas or a problem to share: the gDown channel is on the Gluonify.io Discord. Security vulnerabilities are never reported there, but privately to beacon@gluonify.io. Join #gdown on Discord