Google announced the general availability of the AlloyDB Omni Red Hat RPM orchestrator, a production-ready deployment mode for running PostgreSQL workloads on virtual machines or bare-metal servers with AlloyDB’s performance and AI features. This GA release accompanies AlloyDB Omni version 18.3.0 and adds the RPM-based orchestrator as a fourth deployment option, joining standalone containers (Debian/UBI), Kubernetes-based HA containers, and a standalone RPM.

The RPM orchestrator targets organizations that need cloud-like database automation but prefer local, non-containerized infrastructure. The stated use cases include workload modernization (claimed to be more than 2x faster for transactional workloads and up to 100x faster for analytical queries than standard PostgreSQL), regulated environments requiring strict data residency and security, edge or on-premises deployments with low-latency or disconnected operation, and AI-ready infrastructure for vector search and generative AI integrations via AlloyDB AI capabilities.

The orchestrator supports flexible reference architectures. Administrators specify a reference architecture by adjusting instance parameters, node configurations, and networking. The documented HA architecture shows client applications connecting to a load-balancing tier (using Virtual IP, Keepalived for VIP failover, PgBouncer for connection pooling, and HAProxy for routing). Read-write traffic goes to an active primary node, while read-only traffic routes to replica nodes. Synchronous replication across zones provides HA for the primary cluster; separate read pools can be added asynchronously for scaling read-heavy workloads. A control plane with redundant cluster managers and a 3-node etcd distributed configuration store manages cluster state and health, with Node Managers on each database node. For a single cluster deployment, control and data plane components can run on the same nodes.

High availability and read scalability are emphasized. The orchestrator automatically handles failures across data and control path components, including readable standby nodes and network disruptions. Read pools can be created or added dynamically to a cluster, with dedicated read endpoints configured for standby nodes and read pools.

Data protection features include automated backup and restore to GCS, S3-compatible storage, or local storage, with configurable schedules. Point-in-time recovery and fully automated in-place PIT restore are natively supported. SELinux enforcement is available at or after bootstrap to satisfy enterprise compliance and mandatory access control requirements.

Operational overhead is reduced through a fully automated Low Downtime Maintenance (LDTM) feature for minor version upgrades and CPU/memory adjustments with minimized downtime and automatic rollback. Dynamic configuration allows modifying GUCs at or after bootstrap. Cluster footprint management supports adding or removing database nodes as workloads shift.

Observability features include data and control path log direction to a log disk, plus custom metrics that can export business-level events (e.g., user registrations, tenant sessions, order volumes) to external monitoring platforms. The orchestrator supports all AlloyDB Omni AI features, such as natural language queries, AI-powered search, and AI functions, as well as a list of extensions.

The announcement positions the RPM orchestrator as a way to bring AlloyDB performance, AI capabilities, and automation to on-premises or VM environments. Google points to documentation and a codelab for deploying an HA cluster with the RPM orchestrator. Future capabilities are mentioned as upcoming, including advanced enterprise features for business continuity and cross-region resiliency.