Managed Meilisearch migration note: use full reindex after switching

[demigodmode]

While testing the move from the old two-container setup to the new managed Meilisearch mode, I ran into one migration gotcha worth calling out.

The compose change itself is pretty simple: remove the separate meilisearch service, remove MEILI_URL=http://meilisearch:7700, add ONESEARCH_MANAGED_MEILI=true, keep onesearch_data:/app/data, and add the new managed index volume at onesearch_index:/app/meili_data.

The important bit: after the cutover, don’t rely on a normal incremental reindex.

OneSearch keeps source config and indexed-file metadata in SQLite under /app/data. Managed Meili uses a fresh /app/meili_data index. So after migration SQLite can still “remember” files as indexed, while the new Meili index is empty. A normal reindex may skip unchanged files.

Use full reindex for each source instead:

onesearch source list
onesearch source reindex <source-id> --full

Or from the UI in v0.14.0+, go to Admin → Sources and use Full reindex.

Before running it, make sure the source paths still exist inside the container. If a path changed or a mount is missing, full reindex now fails before clearing existing metadata, but it’s still worth checking mounts first.

Useful check:

docker exec onesearch-app sh -lc '
  curl -fsS \
    -H "Authorization: Bearer $MEILI_MASTER_KEY" \
    http://127.0.0.1:7700/stats
'

If indexes.documents.numberOfDocuments is 0 or way lower than expected after migration, run full reindex.

I’d also keep the old external Meili volume around until managed mode works, search results look right, and the app survives a restart.