Improve the inference of the primary_key

[curquiza]

Currently, to infer the primary_key, the core engine chooses the first attribute found in the first document that contains id as the primary key.

Problem

A document with region_id first and then id in the same document could not be inferred correctly without specifying the primary key. Since id is the unique identifier of a DB in most of the datasets, we might want to find a solution for this kind of dataset.

Solution

An alternative could be to

• first, check if an attribute named exactly id exists and take it as primary_key
• if not, do the inference as we do currently

[gmourier]

Hello @curqui!

Indeed I think it's a good idea to have a dedicated field for id inference without taking into account the position of the field in the sent payloads.

Typically id or _id.

We could also generate this unique id when it is not filled by the user so that he can still use the search engine in optimal conditions.

Selecting the first field containing id within the first document to make it the primary key is not the best solution and may prevent the indexing of all documents. (I think of the possible collision on region_id which is not unique among all documents. If this one is chosen by the engine to be the primary key by the inference system some documents may not be indexed).

[gmourier]

Locked in favor of #206