Add time_series_dimension property into _field_caps API method

[kuzaxak]

Description

time_series_dimension determine whether field will be used by Opensearch Dashboards to build histogram or not.

Fixing issue: #5003

How was this PR tested?

Added test to cover my change.

[CLAassistant]

All committers have signed the CLA.

[fulmicoton]

Suggested change

	let indexes_metadata = resolve_index_patterns(&index_id_patterns, &mut metastore).await?;
	let indexes_metadata = resolve_index_patterns(&index_id_patterns, &mut metastore).await?;

I'm really not fond of doubling the calls to resolve_index_patterns.
We happen to do one in root_list_fields too.

It seems like we pre-build /cache the ListFieldsEntryResponse in the packager (and indexing time) so that root_list_fields(..) is a light operation.

It would make sense to add the bool property there, in the ListFieldsEntryResponse.

message ListFieldsEntryResponse {
  ...
  // True means that the field is the timestamp field of all of the requested indexes.
  bool is_timestamp = 8;
}

We then need to know how and when we populate it.

Solution 1:

The logical place would be the packager.
Pros: that's where we build the rest.
Cons: Right now the packager does not have access to the docmapper and does not know what the timestamp field is. We would have to pass that information. It is reasonable however.
Cons2: the data is not available for existing data.

Solution 2:

Keep it set to false in the packager, and actually compute it in the root_list_fields function.
At this point we have access to the index_metadatas already.
Pros: easy
Cons: ugly.

[fulmicoton]

See comment inline. I don;'t think we want an extra call to the metastore here.

[kuzaxak]

See comment inline. I don;'t think we want an extra call to the metastore here.

Clear, will have a look on weekend into the architecture of the service more deeply. Thanks for suggested solutions, will try to implement the nice one.