Reble
Git-like branching for Apache Iceberg data warehouses, so you can test SQL transformations without copying data
🔗 Visit Reble
Description
Testing a risky change to a data pipeline usually means either running it against production data (scary) or copying a warehouse's worth of data into a staging environment first (slow and expensive). Reble gives data teams a third option, borrowed from how Git branches work for code: create an instant, free branch of your data, test against it, then merge or discard.
Reble is an open-source SQL transformation engine that adds Git-like branching to Apache Iceberg lakehouses. It leverages Iceberg's metadata-only branches to create zero-copy references with sub-10ms creation time and no storage overhead, so you never maintain a separate warehouse copy just to test something. It automatically detects lineage via SQL parsing, guarantees atomic and idempotent commits, and works across multiple catalog systems (Glue, Polaris, Nessie, Hive, REST) with execution via DuckDB or Spark. It also includes optional CI/CD integration and MCP protocol support so AI agents can drive transformations directly.
💬 Our review
The short version: zero-copy, sub-10ms data branching is a genuinely compelling idea for data engineers tired of maintaining shadow copies of warehouses just to test transformations safely — the question is whether the project is mature enough to trust yet.
Compared to LakeFS, which brings similar Git-like versioning but across multiple storage protocols, or Nessie, which does something similar specifically for Iceberg and Delta, Reble's angle is being catalog-agnostic (Glue, Polaris, Nessie, Hive, REST all supported) plus native DuckDB and Spark execution. With zero GitHub stars and a pre-1.0 version number (v0.6.x), it's genuinely early, and setup requires real familiarity with Iceberg catalogs and S3. Free and open source; worth piloting if you already run an Iceberg lakehouse and want safer testing; Nessie is the more established choice if you need production stability today.
📊 Global score
🤖 AI-enriched data
Gratuit, licence Apache 2.0. Nécessite Python 3.10-3.13, un bucket S3 et un catalogue Iceberg configuré.
Pros
Branches zéro-copie créées en moins de 10ms, sans surcoût de stockage
Détection automatique du lineage par analyse SQL (AST via SQLGlot)
Agnostique au catalogue : Glue, Polaris, Nessie, Hive, REST
Commits atomiques et idempotents par table, adaptés à l'orchestration CI/CD et Airflow
Cons
Logiciel encore précoce (v0.6.x, pas de version 1.0 stable)
Aucune étoile ni fork sur GitHub, adoption/visibilité minimale
Nécessite Python 3.10-3.13, un bucket S3 et une bonne connaissance des catalogues Iceberg