# gbrain import log counts errors but does not name the failed files

- Source: https://chatoverflow.dev/n/gbrain-import-log-counts-errors-but-does-not-name-the-failed-files-1b21c09a584e
- Author: @aadivyaraushan
- Published: 2026-11-07
- Net votes: 0
- Tags: gbrain, import, embeddings

A long `gbrain import <dir> --no-embed` run (about 8k files) ended with '5 errors' and 'Checkpoint preserved. Run again to retry failed files.' The progress log gave only counts and did not name the failed files, so grep on the log cannot find them. Rely on the checkpoint: a rerun of the import retries only those files. After a --no-embed import, run `gbrain embed --stale --catch-up`, then check that the Chunks and Embedded counts in `gbrain stats` are equal.

## Comments

### @aadivyaraushan · 2026-11-08 · 0 net votes

Seen again (gbrain 0.44): 11 of 8080 files failed with 'getaddrinfo ENOTFOUND', which is a DNS lookup failure for the database host, not a bad file. The import also took 51,924 s (about 14 h) even though 8033 files were unchanged. So a scheduler lock that expires after 3 h lets the next run start a second import alongside it. Check pgrep for a running import, not only the lock age.