// the find
delta-io/delta-sharing
An open protocol for secure data sharing
Delta Sharing is an open REST protocol for sharing large datasets across organizations without requiring consumers to run the same compute platform as the provider. Data lives in the provider's cloud storage; the server generates pre-signed URLs and consumers download directly. It's for data platform teams at companies that need to share Delta Lake or Parquet tables externally without copying data or negotiating vendor lock-in.
The protocol itself is clean and well-specified — PROTOCOL.md is a proper spec, not just a README with curl examples. Pre-signed URL delegation is the right call: the server never proxies data, so it doesn't become a bottleneck or cost center. CDF (Change Data Feed) support means consumers can do incremental syncs rather than re-reading full snapshots, which matters at scale. The Rust-backed delta-kernel-rust-sharing-wrapper in the Python connector is a meaningful performance investment, replacing pure-Python Parquet parsing for Delta format reads.
The reference server is genuinely a reference implementation, not something you'd run in production — it uses static bearer tokens for auth with an explicit recommendation to stick NGINX in front of it, which is a lot to ask of a first-time deployer. The ecosystem connector table is mostly 'released but read-only': most community connectors only implement QueryTableLatestSnapshot, so anything involving streaming or CDF is Spark-or-Python only. The delta-kernel-rust-sharing-wrapper installation story is a mess — glibc version pins, Rust toolchain as a fallback build dep, and a version escape hatch to 1.0.5 that silently drops the performance benefit. At 951 stars this is low adoption for a 4-year-old project, which is telling: in practice most teams use Databricks-managed sharing and never touch the OSS server.