Picks an existing cache node and sends it off to cache on the farm. The nodes to cache are always selected manually from a UI list (a multiparm with node picker and enable toggle). Optionally a dependency graph also captures upstream cache nodes: collected via BFS, each only once, topologically sorted (those without dependencies first) and wired with a dependency on their direct parents — so the farm submitter knows the correct order.

Picks a solver and caches it itself — including wedging across parameter combinations. Each wedge is written out as a JSON job (solver path, overrides, paths, frames, cache type); a generic hython script reads the JSON, applies the overrides and caches. This quickly produces many versions and wedges — which is what makes the read-from-farm node necessary.

The counterpart to Solver Sweep: reads the generated caches back in (path from hip / solver / version / wedge) and shows in the UI which parameter held which value in this wedge — read from the accompanying JSON.

Both run automatically on the farm and only start once all wedges have finished (via job dependencies). The result: a Nuke file with every wedge in one grid plus a metadata overlay for perfect comparison — and an mp4 you can also compare with.

Because Solver Sweep quickly produces many caches you don't need, the Cache Manager offers a pleasant UI to quickly flag caches as useful — and through that also very simply delete caches to free up disk space.