Since T275268: Address "image" table capacity problems by storing pdf/djvu text outside file metadata part of file metadata can be stored in text table, and in Wikimedia what is stored in text table is just a pointer to external storage. After T362566: Stop growth of text table by storing ES addresses in content table img_metadata can refers to external storage directly, so no new text table row will be added.

By querying img_metadata for djvu files in Commons, we will find a number of rows still refer to text table. They should refer to external storage directly, then the text table rows can be deleted.

Note after this there may be still rows for file metadata in text table: when file metadata is refreshed (rare for existing files), we add new blobs but not delete old ones. Therefore some rows in text table may be orphaned, but there are no obvious way to find them.

See also: T183490: MCR schema migration stage 4: Migrate External Store URLs (wmf production)