land: storage-cross-volume

[aft-alfonso]

Accepted agent delivery; branch-keyed search descriptor included.

---

Summary by cubic

Makes storage operations work when storage folders live on a separate volume, falling back to copies when a rename would cross devices instead of failing.

Cross-volume moves

• New cross_volume::move_path tries a plain rename first and, on a cross-device error, copies the tree, publishes it under the destination name, and removes the source; a failed copy leaves the source untouched.
• Legacy folder repairs (root backups, bash tasks, filter folders, state files) now use this fallback and keep the source in place with a warning when the move still fails.
• Callgraph orphan sweeping now reaches another key's build temporaries when the callgraph folder is a mount point, instead of skipping the whole folder.
• A simulated-volume test seam makes CI reproduce EXDEV errors without a second filesystem.

Retention and migration preflight

• Retention retires a mount-point cache folder by emptying it on its own volume instead of renaming it into the domain trash; the emptied mount point counts as gone on later passes.
• Storage migration preflight now checks free space on each destination volume (the target root plus any cache folder on another volume), charging each volume only the bytes headed for it.

^{Written for commit a7bb59f. Summary will update on new commits.}

Confidence Score: 4/5

Fix mounted-cache cleanup recovery before merging so interrupted deletes do not leave files permanently stranded.

Findings

1.  Interrupted cleanup strands cache files ▶
2.  Slow deletes block storage work ▶

Summary

Adds cross-volume moves for legacy storage repair and checks migration space on each destination volume.

• Legacy storage repairs can move data across volumes.
• Storage migration checks free space on every destination volume.
• Retention empties mounted cache folders without moving the mount.
• The callgraph sweep scans keys on a separately mounted volume.
• The search benchmark records this change as engine-unwired.

Diagram

%%{init: {'theme': 'neutral'}}%%
flowchart TD
  A[Legacy storage repair] --> B[Try rename]
  B -->|Success| C[Use destination]
  B -->|Cross-volume error| D[Copy into destination staging path]
  D --> E[Sync and publish copy]
  E --> F[Remove source]
  F --> C
  D -->|Copy fails| G[Keep source and report failure]

Loading

_{Reviews (1) · Last reviewed commit: "release notes and search-quality descrip..." · Reviewed by Greptile}

[greptile-apps]

Interrupted cleanup strands cache files

If the daemon exits after moving every entry into the mount-local .retention-trash, or deleting that trash fails, emptied_mount_point still returns true. Later passes skip the cache, and reap_retired only checks trash under the domain folder. The leftover data is never reclaimed.

Retry or reap mount-local trash before treating the mount as empty, and add an interrupted-cleanup test.

[greptile-apps]

Slow deletes block storage work

retire_mount_point_contents deletes the full tree before returning. Its caller still holds retention/sweep.lock and the guard from filesystem_guard, so a slow mounted volume can block new binds, pins, and SQLite file work throughout deletion. The entry limit does not bound how long disk operations take.

Leave the slow delete to a background worker, as the normal retirement path does.

Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!