layers_clearUnload rocket_launch

Park the full source graph once you have Carved the slice you need — so the big graph is not holding the working set while Reason runs over the slice.

Roadmap — not shipped in v0.6.14

Unload is C4 — the park-graph lifecycle (v0.6.18) on the Roadmap: a park_graph / unpark_graph pair that offloads a graph's working set and brings it back on demand. The one-call verbs are not yet shipped. The manual approaches below work on v0.6.14 today.

What it will do

After you carve a right-sized slice, the full source graph is dead weight for the reasoning step — it occupies buffer cache and disk you would rather give the slice. Unload parks it: park_graph releases the graph's working set (its indexes, its share of cache) without dropping the data; unpark_graph restores it when you need the full graph again.

The manual approximation today (v0.6.14)

• Same instance — DETACH PARTITION the big graph and drop its hexastore indexes, so only the carved slice's indexes occupy cache. Re-attach when you need it back.
• Separate instance — keep the full source graph in its own PostgreSQL instance and simply stop it while you reason on the slice elsewhere.

Where it will sit in a chain

After Carve (and Sealing the slice), before Reason. Unload is the step that frees the box for single-threaded reasoning over the slice.

See also

• Roadmap — C4, the park-graph lifecycle (v0.6.18).
• Per-graph LIST partitions — the DETACH PARTITION primitive behind the manual approach.
• Pattern: Ingest → Carve → Reason — where Unload fits.