For binaries: Fail to disable database fulltext search while keeping it in elasticsearch

The way to disable fulltext index in the database while still using it in elasticsearch is to set, in nuxeo.conf, nuxeo.vcs.fulltext.disabled to true.

This does not work for binaries: The full text index then does not work at all for the content of ingested files (still works on metadata).

To reproduce (tested initially with MongoDB, reproduced with postgreSQL, here is the postgreSQL test):

1. Have a fresh Nuxeo 9.10 with HF02 (most recent HF at the time this ticket is created)

2. Have a fresh PostgreSQL database (here, named "aaatest")

3. Setup nuxeo.conf to use PG and disable fulltext in the db

# This is nuxeo.conf
nuxeo.templates=postgresql,nuxeo-vision,drive
nuxeo.vcs.fulltext.disabled=true
nuxeo.db.host=localhost
nuxeo.db.name=aaatest
nuxeo.db.password=_THE_PWD_
nuxeo.db.port=5432
nuxeo.db.user=nuxeo

4. Start nuxeo

5. In the domain, create a workspace. In this workspace, drop a Microsoft Office document (or any file whose content should be full text indexed)

6. In dc:description, set some words including a word that you know does not exist in the file you uploaded (for more testing)

7. Display (WebUI) the search

8. Search for a word that exists only in the file => not found, no error in server.log nor in the UI nor in the WebInspector.

9. Search for a word that exists only in the metadata => found

More details:

• In {server}/nxserver/config, the default-repository-config.xml file contains the correct values:

  . . .
  <indexing>
     <fulltext disabled="true" searchDisabled="false"
     . . .
  

• In the pg database, there is no fulltext table

• We know the search in WebUI does use a page provider that uses elasticsearch (and anyway, in the UI, the aggregates work very well. Only the fulltext fails, with no errors)

• Perform a search against the database (SELECT * FROM Document WHERE ecm:fulltext = 'nuxeo') => Fails with an error [...] Fulltext search disabled by configuration