---
title: "2026-09-17 vxData NFS to Ceph mapping retrieval"
description: "Recovered the local mapping CSV and September 14 confirmation; proposed SDK-only metadata updates."
image: "https://docs.virdx.dev/img/virdx-social-card.png"
---

> Documentation Index
> Fetch the complete documentation index at: https://docs.virdx.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# 2026-09-17 vxData NFS to Ceph mapping retrieval

## Work Done

- Retrieved `~/work/playground/nfs_to_ceph_mapping.csv`: 222,494 rows, 217,353 unique source paths, no conflicting mappings or distinct-source target collisions.
- Confirmed September 14 Prime session `01a08fc6-7d73-76f1-9f3e-37f4454acdcb`, messages at 13:49–13:52 UTC, explicitly uses `s3://vxdata/nfs-copy/`, not `nfs-sync`.
- Prefixes `/mnt/storage/data/`, `/mnt/artifacts/data/`, and `/mnt/artifacts/` map to that destination, preserving the remaining suffix.
- Original extraction approach survives in `~/work/playground/fix_essen02_crop_status.py`: `histo_scans.url`, `histo_maps.url`, `diffsim_scans.url`, `generic_files.s3_url`; export mapping, transfer bytes separately, then replace metadata references.
- Proposed exact CSV lookups and typed SDK partial updates, dry-run by default, with unknown local paths rejected before writes. No production reads or writes were performed.

## Pitfalls

- Retrieved a later confirmation conversation, not the original mapping-generation conversation.
- Original extraction explicitly leaves nested `DiffsimScan.other_data` references unresolved. Four fields do not prove deployment-wide coverage.
- Mapping existence is not proof that copies completed or that consumers support S3. Verify both before applying.
- SDK update batches are separate transactions; the complete migration is not atomic.

## Improvements

- Reuse the exact mapping CSV rather than regenerating destinations from a broad regex.
- Preserve an identifier/field/old/new manifest before applying; pause concurrent writers during the switch.

## Unmapped paths intentionally skipped

- User clarified that new /mnt data may have arrived since the NFS copy. The flip script now leaves unmapped paths untouched and counts these separately from missing-copy exclusions.
- Added a dry-run/apply regression test for copied, missing, and unmapped records. Formatting, lint, and scoped type checking pass. Direct pytest initially failed on a missing fsspec import, but the normal Nx/Pixi test command passed all three tests, including both new cases. No production operations were run.

## Script saved

- Saved the proposed job in `mono/apps/vxdata-jobs/src/vxdata/jobs/f_20260917_nfs_to_ceph/main.py`, dry-run by default with explicit `--apply`.
- Added exact exclusions from `~/Downloads/vxdata-missing.txt`; all 611 entries resolve after normalizing `/mnt/data/storage/artifacts/` to `/mnt/artifacts/`.
- Scoped formatting, lint, typecheck, CLI help, and offline loading of the real mapping/exclusions passed. No production queries or writes were run.

Source: https://docs.virdx.dev/knowledge/inbox/2026-09-17-infrastructure-vxdata-path-mapping-retrieval/index.mdx