We would need to test it first in a real network environment (i.e. not with the Hibernate Search application and Elasticsearch node on the same machine). If this does improve performance, it may be worth it.
The impact on performance will always be depending on various factors, and we can't reproduce them all. I'd rather accept that it will be worth it for some users and therefore implement it. No need to performance test this, other than verifying we don't introduce regressions, but that's not a specific requirement for this issue. |