Skip to content

Roadmap

CQLite is developing a Cassandra-compatible storage engine for analytics, reusable materialization, bulk ingestion, and compaction. The CLI, language bindings, and query interfaces remain supported product surfaces.

The Product Requirements define the scope, guarantees, and exit criteria. The research assessment explains the rationale. GitHub tracks execution; the changelog records shipped behavior.

Last reviewed: 2026-09-06.

These milestones are qualification pending. Existing code and research provide a foundation; completion requires the PRD’s workload and operational evidence.

MilestoneOutcome
SE1 — Engine contractExplicit capabilities, authority, supported workloads, and measurable acceptance criteria
SE2 — Shared offline executionSafe compaction and reusable analytics materialization, including qualified compressed output
SE3 — Analytics offload valueMeasured analytics benefit within foreground Cassandra latency, resource, and freshness budgets
SE4 — Bulk writing and policy toolsBounded bulk generation through native Cassandra import, plus compaction planning and simulation
SE5 — Qualified freshnessProven memtable-tail behavior and cost for opted-in analytical reads
SE6 — Bounded Cassandra integrationOne reversible native-format or coordinated-compaction pilot
SE7 — Native lifecycle qualificationSupported integrated behavior backed by recovery, lifecycle, and operator evidence

Compaction planning can proceed alongside the engine contract. Bulk writing shares output qualification with offline compaction. Live analytics measurement can continue while reusable materialization is developed.

External-engine 1.0 requires SE1–SE4, independent operator evidence, stable public contracts, and the release checks specified in the PRD. WASM, Vortex adoption, a continuous Iceberg daemon, and native Cassandra integration are not prerequisites. Freshness and native integrations retain separate experimental status until their applicable gates pass.

The initial authority remains Cassandra-readable SSTables. New analytical representations are derived and rebuildable. The proposed Mustang architecture and Vortex representation are candidates to validate, not promises of shipped support.

M1–M5 delivered core reading, the CLI, output writers, Python/Node bindings, and writing with STCS compaction. Those accomplishments remain the foundation.

The old M6 (WASM) and M7 (performance/size/1.0) sequence is replaced by the qualification program above. The original PRD preserves that history. Historical completion does not imply universal compatibility or qualify every new offload workflow.

Open an issue describing the workload, Cassandra version, data shape, freshness requirement, current cost or latency, and the improvement that would make deployment worthwhile. Operator workloads and reproducible evidence guide qualification priorities.