Parent-Child support

[jma]

Add the possibility to specify the parent id and parent document type during indexing.

[lnielsen]

This is how it should be specified for Elasticsearch:

• Direct indexing:

  client.index(...,parent='<id of parent>')

• Bulk indexing:

  {
    '_index': '...',
    '_parent': '<id of parent>',
    ...
  }

Now in terms of Invenio-Indexer API I see two options:

Option 1) Explicit specify parent

RecordIndexer.index(child_record, parent=parent_record.id)
RecordIndexer.index_by_id(child_record_id, parent=parent_record_id)
RecordIndexer.bulk_index([(child_id, parent_id), ...])

Option 2) Programmatically extract parent from record

The Invenio Indexer API would stay unchanged:

RecordIndexer.index(child_record)
RecordIndexer.index_by_id(child_record_id)
RecordIndexer.bulk_index([child_id, ...])

Parent should then either be stored or programmatically deteremined from some other record field:

2a) Stored in RecordMetadata

RecordMetadata:
  id = ..
  json = ..
  version_id = ...
  created = ...
  updated = ...
  parent = ForeignKey(RecordMetadata.id)

With this FK constraints can be enforced by the database.

2b) Stored in the record:

Record({'_parent': 'parent record id'})

2c) Stored or calculated from another field:

record = Record({'parent': {'$ref': 'http://..../asdsd'}})
parent_record_id = parent_for_record(record)

Personally I think I'm mostly leaning towards option 2a, since it allow to enforce the foreign key constraints.

---

In addition to Invenio-Indexer, parent-child support is probably needed in Records-REST/UI (specifying endpoints for sub records, parent-child aggregations etc).

---

Use cases

Some few uses cases for parent-child relationships include:

Record<->Annotation
Record<->Files (e.g. CERN OpenData 15k files for in a record)

[jirikuncar]

IMHO the parent-child relationship would be better expressed by PID relation (see also inveniosoftware/invenio-records-rest#185).