Graveler Lab

A lab for the versioning engine inside lakeFS. Describe an experiment in YAML (writes, commits, branches, merges, diffs and reads) and it replays against a model of Graveler, drawing every commit as a meta-range of SSTable ranges laid over the sorted keyspace. You see which ranges a commit rewrites, which it shares by reference, and what a diff, a merge or a lookup actually opens.

• Runs as you type: the YAML editor reruns the experiment on every change; step through it, play it at three speeds, or scrub across the step meter.
• Two views: the lattice shows every commit as a row of ranges; the tree shows one commit as commit, meta-range, ranges and records, with ids, identities and addresses.
• Numbers for every step: new ranges, reuse, objects under _lakefs/, deduplication, SSTable size, diff work and lookup hops.
• Inspector and log: select a range or a commit to see where it is used, or list every object in the store with its size and reference count.
• Split rules and identity: compare content-defined and fixed-size range boundaries, and identity by entry or by content.
• An agent that sees the step you are on, explains it, and writes or edits experiments for you (it needs the hosted runtime; everything else works without it).

Seventeen experiments come with it, from One change, one range (the example from the lakeFS docs) through Zero-copy branching, Diff skips shared ranges, Three-way merge, Merge conflict, Deletes and tombstones and Bulk import, to an empty Sandbox. A reference for the experiment language is on the page. It is a teaching model, not lakeFS: ranges are sized in entries, and a fast 53-bit hash stands in for SHA-256.