Searching over multiple indices

[sylbru]

Hi!

I’m just discovering Meilisearch and being quite amazed honestly, but here’s a question: how do you folks handle searching over several types of entities? I think there’s no way of searching over multiple indices, so I currently do several searches (one for each index) and merge the search results then filter order them by ranking score. This seems to work okay, but I’m wondering if there’s a better solution? Indexing everything in a single index? Something else?

Cheers,
Sylvain

[sanders41]

Have you tried multi-search? https://www.meilisearch.com/docs/reference/api/multi_search#perform-a-multi-search

[sylbru]

:o No I haven’t! Will try right now if my laptop’s battery allows it!

[sylbru]

@sanders41 Trying that right now, and it looks to me as if the results are partitioned by index. What I’m looking for is actually to do a global search with the results from multiple indices merged into one list of search results. It works by doing separate searches and merging the results, probably the same with this multi-search feature, but I’m wondering if there’s a better way, i.e. one that would give more "relevant" ranking scores. But maybe that’s not needed and just merging the results from searches and sorting by ranking score is good.

[curquiza]

Hello @sylbru

I'm sorry but what you are looking for is not in Meilisearch yet. You are not the first one needing this feature. It's not trivial to implement, technically, but also regarding the API UX.
You can participate in this product discussion if you want to help us by defining your needs: meilisearch/product#489

[sylbru]

@curquiza Ah thanks for pointing me to that discussion! I'll continue testing and see if my workaround is good enough or if I should change my search UI (basically present separate searches and let the user pick where to look).

[sylbru]

So for reference, to summarize what I’ve said in the other discussion, my current workaround is to:

• use the multi-search API to perform multiple search queries over the indexes
• each of them has no offset and a limit of page_number×page_size (that’s required for proper pagination)
• each of them has the showRankingScore parameter set as true
• then merge the results and order them by descending ranking score
• finally slice the resulting results array to get the page that you want.

[curquiza]

Thank you @sylbru for your feedback

Pinging @macraig and @dureuill who had previous discussion about federated search

[macraig]

Thanks @curquiza , we've continued the conversation in the other discussion :)