[OAK-9875] Prefix query on a Long, analyzed field fails when executed over Elastic

• Type: Bug
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Component/s: elastic-search, indexing
• Assignee: Unassigned
• Reporter: Nuno Santos
• Created: 29/Jul/22 12:48
• Updated: 30/Aug/22 16:50

Description

Index definition:

"testIndex": {
 "jcr:primaryType": "nt:unstructured",
 "analyzed": true,
 "type": "Long",
 "name": "propa"
}

Data:

/test/a    [propa = 1234 ]
/test/b    [propa = "1234"]
/test/c    [propa = "1234a"]

The following query executed with Elastic:

/jcr:root//*[jcr:contains(@propa, '123*')]

fails with:

14:28:51.713 ERROR [I/O dispatcher 1] ElasticResultRowAsyncIterator.java:326 Error retrieving data from Elastic: closing scanner, notifying listeners
co.elastic.clients.elasticsearch._types.ElasticsearchException: [es/search] failed: [search_phase_execution_exception] all shards failed
	at co.elastic.clients.transport.rest_client.RestClientTransport.getHighLevelResponse(RestClientTransport.java:281)
	at co.elastic.clients.transport.rest_client.RestClientTransport.access$200(RestClientTransport.java:63)

The detailed error is not shown in the logs, but can be found by repeating the query directly to Elastic:

"Can only use prefix queries on keyword, text and wildcard fields - not on [propa] which is of type [long] 

The same query executes correctly with Lucene.

Analysis

The root cause of the problem is that the Elastic plugin does not create a field for full text search on typed properties:

https://github.com/apache/jackrabbit-oak/blob/e13d6c7dc5df3087955a85cb28f5bfb8b6171977/oak-search-elastic/src/main/java/org/apache/jackrabbit/oak/plugins/index/elastic/index/ElasticIndexHelper.java#L197

Therefore, it is not possible to run full text queries on this index.

Issue Links

• is related to: OAK-9910 Full-text queries fail in Boolean analyzed fields when executed over Elastic
• links to: GitHub Pull Request #649
• links to: GitHub Pull Request #670