We're using OGM 5.1 Beta 1, with Neo4j at the moment.

I'm seeing that the labelling of embedded fields is different depending on whether the field is a single object or some sort of collection. The 'EMBEDDED' label is always created. And in the case of a collection a label is created as <entity_name>_<field_name>. When it's not a collection there is no additional label. We've annotated with @ElementCollection in the one case, and @Basic in the other.

This isn't causing us any grief at this point though, just an observation.

The other thing that could be causing us trouble is the query generation when a field within an embedded is constrained in the where clause. The cypher query always uses the EMBEDDED label despite there being a more specific label available (we're querying on a collection of embedded in this case).

This potentially causes trouble in two ways:
- the search within Neo4j is broader, as there are naturally many more rows labelled as EMBEDDED than with the more specific label
- it means we have to create an index on EMBEDDED.<field> rather than use the more specific label in order to ensure an index search - not sure, but potentially this creates more indexing work in Neo4j

Would it be possible to create and use more specific labels throughout?

Hi,
thanks for letting us know.

I'll have a look. I just have a couple a question:



I think I understand what you mean but could you give me an example?



What would you suggest?

Davide

Thanks.

The query is:

MATCH (ms:`FM_ALARM`) -[:sources]-> (`_ms1`:EMBEDDED), (ms:`FM_ALARM`) -[:sourceSystem]-> (`_ms2`:EMBEDDED) WHERE ((`_ms2`.scheme = {ssyssch}) AND (`_ms2`.namespace = {ssysns})) AND (((`_ms1`.scheme = {srcsch1}) AND (`_ms1`.namespace = {srcns1})) AND (`_ms1`.fdn = {srcfdn1})) RETURN ms

The alias _ms1 is for an embedded collection, which has it's own label - :fm_Alarm_sources (:<entity name>_<field name>). The field _ms1.fdn is the has an index. Ideally we'd only index :fm_Alarm_sources.fdn but since the query is referencing the :EMBEDDED label we're forced to create the index on :EMBEDDED.fdn, which is much wider in scope.

So the suggestion to use more specific labels amounts to a) adding specific labels to all embedded nodes (e.g. the alias _ms2 is for a non-collection embedded, and it doesn't seem to get a label on creation, so maybe use :fm_Alarm_sourceSystem), and b) using these labels in queries instead of :EMBEDDED to narrow the search scope, as well as narrowing the scope of any indexes one might need to create.

Thanks I've created an issue to keep track of this proposal: https://hibernate.atlassian.net/browse/OGM-1231