Why do I see empty files in my blob storage export?
The blob storage integration exports one time window of data per run, writing a file per table (for example observations_v2/ and scores/). If you find empty files in your bucket, this is expected behavior — here's why, and how to handle it.
Empty files are expected and normal; they are not an error or a sign of data loss. A given time window may contain data for some tables but not others (for example, observations but no scores in that window). When a table has no data for a window, the exporter may still write a file for it, and that file will be empty.
What an empty file looks like depends on the configured file format:
| Format | Empty file looks like |
|---|---|
| Parquet | A valid Parquet file with schema metadata but zero rows (not a zero-byte file) |
| JSON | A file containing [] |
| JSONL | An empty file |
| CSV | A header row only (or empty) |
Build downstream pipelines to tolerate empty files:
- Treat an empty file as "no data in this window for this table," not as a failure, and don't alert on its presence.
- Don't assume every window contains a file for every table.
- Process the objects listed in the completed run's manifest instead of deriving filenames or listing table directories.
See Process exports for the bucket layout and manifest workflow, or the export field reference for schemas.
Last edited
