Vector DB Snapshot Operator

An opinionated Kubernetes Operator to orchestrate consistent, automated snapshots for multiple vector databases (Qdrant, Weaviate, Milvus, OpenSearch k‑NN/Vespa, and generic engines) and ship them to object storage with retention policies.

Goals

• Engine‑agnostic CRDs with per‑engine adapters.
• Cluster‑aware discovery: label/annotation selectors; shard‑wise coordination.
• Pluggable backends: S3, GCS, Azure Blob, NFS/CSI.
• Integrity: checksums + manifest; optional quiesce hooks.
• Observability: conditions, events, Prometheus metrics.
• Safety: idempotent reconcile, finalizers, concurrency limits, per‑engine backoff.

Local testing

0) Ensure code-gen & CRDs exist

make manifests

1) Spin up kind & install CRDs

make dev-setup

2) In a second terminal: run the controller locally

make dev-run

3) Back in the first terminal: apply sample resources

make dev-apply

4) Trigger a manual snapshot and watch

make dev-snapshot make dev-watch