Troubleshooting
Download is slow or stalls
- Tune S3 transfer concurrency:
--max-concurrent-requests,--max-queue-size,--multipart-chunksize(applied via a temporary AWS config, so your global~/.aws/configis never modified). - Re-running
downloadresumes incrementally — only missing/changed files are transferred.
verify_download reports a size or row-count mismatch
The local file no longer matches the published manifest.json. Re-run download to re-fetch
the affected file(s), then re-run verify_download. Use --full for a row-scan that also
catches corrupt data pages, or --quick to check presence + size only.
index or extract can't find works
extract routes W IDs to parquet/works/ (the enriched corpus). If you only downloaded the
raw snapshot with --no-enrich, run enrich first (or re-run download without --no-enrich).
index --dataset all deliberately skips the raw *_aws staging dirs.
Out of memory
The pipeline processes one parquet file at a time, so memory use is modest. If you still hit
limits, constrain workers/memory: --workers <N> and --max-memory-mb <N>.
Metadata layout
Current canonical location is openalex-snapshot_metadata/:
openalex-snapshot.lock— present while a command runsreports/— a JSON report per command, written only when failures occurdownload/manifest.json— the fetched OpenAlex manifest (audit copy)<dataset>/index/,<dataset>/index-verify/— index logs