New

The executive guide to generative AI

Read more

Relation names inference

edit

Relation names

edit

Prior to Elasticsearch 6.x you could have multiple types per index. They acted as a discrimatory column but were often confused with tables. The fact that the mapping API’s treated them as seperate entities did not help.

The general guideline has always been to use a single type per index. Starting from 6.x this is also enforced. Some features still need to store multiple types in a single index such as Parent/Child join relations.

Both Parent and Child will need to have resolve to the same typename to be indexed into the same index.

Therefore in 6.x we need a different type that translates a CLR type to a join relation. This can be configured seperately using .RelationName()

var settings = new ConnectionSettings()
    .DefaultMappingFor<CommitActivity>(m => m
        .IndexName("projects-and-commits")
        .RelationName("commits")
    )
    .DefaultMappingFor<Project>(m => m
        .IndexName("projects-and-commits")
        .RelationName("projects")
    );

var resolver = new RelationNameResolver(settings);
var relation = resolver.Resolve<Project>();
relation.Should().Be("projects");

relation = resolver.Resolve<CommitActivity>();
relation.Should().Be("commits");

RelationName uses the DefaultTypeNameInferrer to translate CLR types to a string representation.

Explicit TypeName configuration does not affect how the default relation for the CLR type is represented though

var settings = new ConnectionSettings()
    .DefaultMappingFor<Project>(m => m
        .IndexName("projects-and-commits")
    );

var resolver = new RelationNameResolver(settings);
var relation = resolver.Resolve<Project>();
relation.Should().Be("project");
Was this helpful?
Feedback