Elasticsearch default configuration

Fielddata and Nuxeo

The current Nuxeo ES mapping uses fieldata on the dc:title field.

=> see default-doc-mapping.json

This setting seems to be there for NXP-22808 and the idea is to be able to do partial (like) match on the title.

However, as pointed in the ES doc, fielddata has an impact on memory usage at the ES level.

Fielddata and ES Circuit-breaker

During the benchmark, usage of fielddata triggered a Circuit-Breaker Exception because it required too much memory.

see SUPINT-1744

 org.elasticsearch.ElasticsearchStatusException: 
  Elasticsearch exception [type=circuit_breaking_exception, 
   reason=[parent] Data too large, data for [<http_request>] would be [23679924882/22gb], which is larger than the limit of [22845023846/21.2gb], 
   usages [request=0/0b, fielddata=22159489269/20.6gb, in_flight_requests=2502/2.4kb, accounting=1520433111/1.4gb]]
	at org.elasticsearch.rest.BytesRestResponse.errorFromXContent(BytesRestResponse.java:177)

The solution to continue without increasing again the number of ES nodes was to re-index with a different mapping:

See this mapping:https://github.com/nuxeo-sandbox/nuxeo-jit-blobstore/commit/59bbeb76555d5f64a672fd4425cccff9391a5cb2#diff-ebf15b720fb42a01d48d5ce4b23fdf5e

Nuxeo default-configuration

In its current state, the ES mapping does not allow us to scale, so we basically have 2 options:

• we remove usage of fielddaete from the default mapping
• we create a "large repository" dedicated mapping

In both cases we need to:

• document
• define migration path