davanstrien/huggingface-datasets-search-v2-bucket-staging

datasets-search-v2 — bucket R/W staging

Staging copy of davanstrien/huggingface-datasets-search-v2 that points ChromaDB directly at a bucket mounted R/W at /data instead of rebuilding into /tmp on every boot.

This Space intentionally runs as root (no USER user in the Dockerfile) per the HF infra guidance from 2026-04-09: Docker SDK Spaces that need bucket R/W should run as root.

First boot bootstraps ChromaDB from the read-only /embeddings dataset mount and writes to /data/chroma on the bucket. Subsequent boots skip the bootstrap and read straight from the bucket — testing both that the write worked and that persistence survives restarts.

If this staging Space runs cleanly through a few restarts, the production Space (davanstrien/huggingface-datasets-search-v2) can be cut over.

Related: Projects/Storage-Buckets-Sprint/sqlite-on-buckets-workaround.md