Term filter vs multiple indices

[Eldad_Moneta]

Hi,

I have millions of metrics for ~1 hundred sources (the number of sources will raise in the future).
A query is always for a single source.
I'm already splitting the indices per day.

I'm trying to decide between multiple indices - one per source vs a single index with the source as a term in the document (in this case, all queries will use a term filter with the source id).

Is there a performance difference between the 2 above? during index? during query?

The cluster is write heavy- thousands of index requests (that translate to tens of bulk requests) per second.

Thanks.

[dadoonet (David Pilato)]

If whatever the source all docs have same fields, I'd try with one single index.

If not, I'd try to avoid sparse fields and would split into separate indices.

[Eldad_Moneta]

Thank you @dadoonet.
All docs are of the same type and have the same fields so I guess the answer is single index.
Does this answer based on better maintainability, or there are performance considerations as well?

[dadoonet (David Pilato)]

Sharing the same Lucene instances will consume less resources. That's why I'd probably mix that.