- `search_string` field is removed
- text search vector for `content` and `torrent_contents` is generated in code with improvements in normalisation and tokenisation
- the slow `LIKE` query is removed from text search
- a query DSL is implemented which translates the input search string to a Postgres tsquery
- a CLI command (`reindex`) is provided for updating the tsv for pre-existing records
- the unused `tsv` and `search_string` fields are removed from the `torrents` table
See https://github.com/bitmagnet-io/bitmagnet/issues/89
mgdigital