Unomi 3 · Platform
Unomi 3 platform foundation
The 3.0 runtime floor — Java 17, Karaf 4.4, Elasticsearch 9, and clustering without Cellar or Hazelcast.
Before multi-tenancy and the Unomi 3.1 product leap could stick, the project needed a boring, modern runtime. That is what the 3.0 line delivered: fewer legacies, clearer installers, and a cluster model that matches how Unomi already stores data.
Baseline upgrades
- Java 17+ — Java 11 is retired for Unomi 3 installers and CI
- Apache Karaf 4.4 — continued into 3.1 with current patch lines (for example 4.4.11)
- Elasticsearch 9 — Java API client; HLRC retired; OpenSearch as a peer distribution (see Elasticsearch 9 and OpenSearch)
- Log / dependency pins — platform bump PRs keep Log4j and client stacks on supported lines
Persistence-based clustering (UNOMI-877)
Cellar and Hazelcast are gone. Nodes register ClusterNode documents in Elasticsearch or OpenSearch, publish heartbeats and stats, and discover peers through the shared index.
The same mechanism underpins executor selection and locking for the
cluster-aware task scheduler.
What operators should do
Treat 3.0 as the platform cutover: confirm Java 17 hosts, pick an ES9 or OpenSearch distribution, and read the clustering and health-check chapters before stacking tenants and tasks. The migrating from 2.x post covers the REST and auth cutover; the manual holds the day-two runbooks.