rank_window_size instead of window_size is failing

[mariaapostigo]

When using AsyncDenseVectorStrategy(hybrid=True, rrf={"rank_constant": 40, "window_size": 30}) from elasticsearch/helpers/vectorstore/_async/strategies.py I am getting the following error:

BadRequestError(400, 'x_content_parse_exception', '[1:21990] [rrf] unknown field [rank_window_size] did you mean [window_size]?')

I have a docker with elasticsearch 8.14 and my elasticsearch-py is 8.17.1

What is going on with the compatibilities?

[pquentin]

Hello, as noted in the README's compatibility section, language clients are forward-compatible, meaning that elasticsearch-py 8.17.1 can talk to Elasticsearch 8.17.0 or greater. We can't guarantee backward compatibility, especially for technical preview features, which is the case of RRF in Elasticsearch 8.14.

window_size was renamed to rank_window_size in Elasticearch 8.15 and RRF became generally available in Elasticsearch 8.17. To accommodate this, the choice we made in elasticsearch-py 8.16 and above is to automatically rename window_size to rank_window_size, see #2666 for more details. Additionally, starting with elasticsearch-py 8.16, we emit Python warnings for usage of tech preview APIs.

I would suggest upgrading Elasticsearch to 8.17. Downgrading elasticsearch-py to 8.14 is possible too, but you would miss on tons of improvements made to Elasticsearch regarding vector search in the past year, and this could cause issues with dependencies like LangChain if you use those.