Hi Robert, Thanks for the proposal and for the illustrative PR, it really helps to see the whole picture before the split. It's great to see a familiar database being used in Polaris :-)
I went through #6024 and left comments there; the overall approach looks good to me, and it follows the same patterns as the MongoDB backend. The proposed commit split also makes sense, I'll review the pieces as they come in (starting with #6088). On the questions you raised: - Consistency: I understand LOCAL_SERIAL / LOCAL_QUORUM as defaults, assuming that multi-DC setups would be active-passive deployments, which sounds reasonable. But I think the levels should be configurable (I don't think they are right now), and the assumption documented. - Maintenance: setupSchema() also runs on every server startup, not only in the Admin Tool. Given that concurrent CREATE TABLE is unsafe before Cassandra 6.0, we may want servers to only validate the schema instead. Tests & packaging look mostly OK, I left a few minor comments in the PR. Thanks, Alex On Mon, Nov 9, 2026 at 3:13 PM Robert Stupp <[email protected]> wrote: > > Hi everyone, > > I'd like to propose Apache Cassandra as another backend for Polaris' > existing NoSQL persistence implementation. > > I've opened a draft PR [6024] with the implementation so there is something > concrete to review alongside this discussion. > > This would give operators who already run Cassandra another option for > storing Polaris metadata using infrastructure and operational experience > they already have. > Following the recent DynamoDB integration, it also extends the choice of > databases available through the existing NoSQL framework. > > The implementation uses the existing NoSQL backend interface and > catalog-state model. > Cassandra stores the metadata objects and references, with lightweight > transactions implementing conditional reference updates. > The catalog-reference update remains the publication point for a catalog > change set. > > The draft includes: > - The Cassandra adapter and Java-driver configuration. > - Server and Admin Tool integration, including authentication, TLS, metrics, > and readiness checks. > - Backend tests, wiring into the shared persistence and correctness suites, > and a server mTLS smoke test. > - Configuration and bootstrap documentation. > > Cassandra is selected explicitly through the NoSQL backend configuration. > The draft adds its driver dependencies to the server and Admin Tool > distributions. > It keeps Cassandra-specific implementation and configuration in the backend > module. > > The proposed baseline for new deployments is Cassandra 5.0 or later. > Operators create the keyspace with the replication settings appropriate for > their deployment; Polaris initializes its tables during bootstrap. > > I'd welcome feedback on adding this backend and on the implementation, > particularly: > - Whether the consistency settings and conditional operations satisfy the > existing NoSQL persistence contract. > - Gaps in test coverage or operational documentation. > - Dependency, packaging, or maintenance concerns with the proposed > integration. > > The PR is a draft to support that discussion, and I can split it into > smaller reviewable changes as needed. > > Thanks, > Robert > > [6024] https://github.com/apache/polaris/pull/6024
