-
Notifications
You must be signed in to change notification settings - Fork 0
Milestones
List view
Exit criterion: 3-master × 3-replica cluster on monoio survives node kill (auto-failover < node_timeout×2), live slot migration under load with zero lost acknowledged writes, 72h soak, jepsen-lite green. Plan: .planning/rfcs/v0-9-milestone-plan.md (moon-docs@65a6f1a).
No due date•0/10 issues closed**Theme:** Production GA — extend the moat, drop "experimental," land first managed/paid deployment. **Q3 (moat extension):** - **Vector quantization R&D extension** (R4 defense) — PQ + binary (1-bit) + learned quantization; publish recall × QPS × memory frontier vs valkey-search, Qdrant, Weaviate, Milvus on SIFT1M/GIST1M/DEEP1B/MS-MARCO - **WASM Module Sandbox MVP** (RFC-gated) — Wasmtime host, fuel-bounded, memory-bounded, no syscall surface; `MODULE.LOAD`/`MODULE.CALL` - **Agentic primitives v2** — session memory, `FT.JOIN`, bi-temporal hybrid (`AS_OF + VALID_AT`) - **GPU vector kernel productionization** (conditional on design partner) — cuBLAS L2/IP/cosine ≥10× CPU SIMD on H100/A100 - **License/governance review** — Apache-2.0 stay vs CNCF Sandbox vs LF AI & Data **Q4 (GA push):** - **v0.3.0 GA release** — remove experimental warning, publish `docs/SUPPORT.md` - **Independent benchmark publication** — Q1's third-party results published with full reproduction - **Cluster-scale demonstration** — 64-node, 16-shards/node, 100M+ RPS mixed workload vs Valkey 9.x - **Managed-service pilot** — one MSP/cloud partner (DigitalOcean / Aiven / Render) - **Design partners (3-5)** — anchor users in AI startups / latency-sensitive ad-tech / vector-search SaaS - **Conference talks** — KubeCon / QCon / CMU DB / NeurIPS **Exit criteria:** - [ ] v0.3.0 GA published - [ ] Three customer case studies live - [ ] One managed-service pilot in private beta - [ ] README badge: `production` - [ ] Independent benchmark published - [ ] 64-node cluster bench published **Kill-criteria:** - WASM sandbox RFC reveals fundamental escape risk OR >5× hot-path overhead → defer to v0.4 - No GPU design partner by end of Q3 → drop GPU productionization - Active-active CRDT → defer to v0.4 by default **Anchor doc:** [.planning/milestones/v0.3.0-ROADMAP.md](../tree/main/.planning/milestones/v0.3.0-ROADMAP.md) §5 (v0.3.0)
Due by May 25, 2027**Theme:** Distribution & Ops — make Moon easy to deploy and consume from any stack. **P1 deliverables:** - **Java SDK** (`moondb-java` on Maven Central) — sync + reactive, cluster-aware, vector/graph/text helpers - **Go SDK** (`moondb-go`) — go-redis-compatible interface for drop-in import - **Atomic slot migration** — replace key-by-key MIGRATE with batch slot transfer (`SLOT_TRANSFER_BEGIN / DATA / COMMIT` bus messages, reuses snapshot segment-COW) - **Kubernetes Operator** (`kube-rs`) — `MoonCluster` / `MoonReplica` CRDs, StatefulSet management, rolling upgrades with PSYNC continuity, PVC management - **Helm chart** — published to `charts.moon.io` and ArtifactHub - **OpenTelemetry server-side exporter** — wire reserved `otel` feature: spans per command, RED metrics per shard, exemplars on slow queries - **OPERATOR-GUIDE.md hardening** — capacity planning, failover runbook, upgrade matrix, DR procedures **Exit criteria:** - [ ] Java + Go + Python + Rust SDKs all on semver-stable channels - [ ] Helm install → working 3-node cluster in 5 minutes - [ ] Atomic slot migration ships with measured stall-time improvement - [ ] OTel exporter shipping spans + metrics by default **Anchor doc:** [.planning/milestones/v0.3.0-ROADMAP.md](../tree/main/.planning/milestones/v0.3.0-ROADMAP.md) §5 (v0.2.5)
Due by November 25, 2026**Theme:** Credibility Foundation — make the project safe to depend on, patch visible parity gaps, ship the proof. **P0 deliverables:** - Storage format v1 commitment (`docs/STORAGE-FORMAT-V1.md` + `--storage-format v1` flag) - Hash-field TTL family (11 cmds): HEXPIRE / HPEXPIRE / HEXPIREAT / HPEXPIREAT / HEXPIRETIME / HPEXPIRETIME / HPERSIST / HTTL / HPTTL / HGETDEL / HGETEX - CLUSTERSCAN (cluster-wide SCAN with `{node_seq, intra_node_cursor}`) - Multi-node cluster benchmark harness (3 / 8 / 16 / 32 / 64 nodes vs Valkey 9.1) - Third-party independent benchmark — contracted (publishes in v0.3.0) - Chaos test suite expansion (50+ scenarios in `tests/chaos/`) - LTS policy + `SECURITY.md` + 90-day CVE disclosure **Exit criteria:** - [ ] Hash-field TTL + CLUSTERSCAN merged - [ ] Cluster bench published with ≥16-node run - [ ] Storage format v1 doc committed - [ ] Third-party benchmark contracted - [ ] LTS + security disclosure policy published - [ ] README badge: `experimental` → `production-ready (RC)` **Anchor doc:** [.planning/milestones/v0.3.0-ROADMAP.md](../tree/main/.planning/milestones/v0.3.0-ROADMAP.md) §5 (v0.2.0) **Competitive context:** [docs/comparison-valkey.md](../tree/main/docs/comparison-valkey.md)
Due by August 25, 2026•11/17 issues closed