search-indexer image (uid 1000) silently fails to write PROVENANCE_FILE to a root-owned dir, disabling skip-unchanged

[ddeboer]

Summary

The @lde/search-indexer image runs as the non-root node user (uid 1000), but writes PROVENANCE_FILE without ensuring the target directory is writable by that user. With a root-owned mount (the default for a fresh named volume or a bind-mounted host dir), the provenance file is silently never written, so skip-unchanged never engages and every run reprocesses the full dataset.

Symptom

Configured per the docs with a persisted provenance location:

volumes:
  - provenance:/provenance
environment:
  PROVENANCE_FILE: /provenance/provenance.json

The run completes successfully (exit 0) and indexes correctly, but a second, identical run does not skip – it re-imports and re-projects everything. Inspecting the volume afterwards:

$ docker run --rm -v search-indexer_provenance:/p alpine ls -la /p
drwxr-xr-x root root  .        # <- root-owned, mode 755
                                # (no provenance.json)

The directory is empty: node (uid 1000) cannot create provenance.json in a root:root 755 dir, and the failure is not surfaced.

Expected

Either:

• the pipeline/image ensures the PROVENANCE_FILE parent directory is writable by the runtime user (e.g. chown/mkdir in the entrypoint, the way DATA_DIR is handled), or
• a failed provenance write is reported as an error rather than swallowed, so a misconfigured mount is caught instead of silently disabling skip-unchanged.

At minimum, document that PROVENANCE_FILE’s directory must be writable by uid 1000.

Impact

Skip-unchanged is a headline feature for cron-driven reindexing (avoid rebuilding QLever + re-projecting an unchanged dump every tick). As shipped it is silently inert on the most natural configuration. Indexing correctness is unaffected – re-runs are idempotent and produce identical counts – but the optimization never kicks in.

Notes

Runtime user confirmed:

$ docker run --rm --entrypoint sh ghcr.io/ldelements/search-indexer:latest -c 'id'
uid=1000(node) gid=1000(node) groups=1000(node)

Related: #652 (the indexer image), #634 (the file-backed ProvenanceStore this consumes).