Reindex with shared index

[jeremylynch]

The project describes how to share a single index across multiple models. However, when calling reindex! as described, the index is cleared and replaced with the objects from only one of the two models.

Algoliasearch-rails has an option for atomical indexing, is it possible to achieve this in meiliesearch-rails also?

[brunoocasali]

Hey @jeremylynch how are you?

Unfortunately, it does not :/ But I agree that is a very good addition to the library. PRs are welcome!

[ellnix]

The project describes how to share a single index across multiple models. However, when calling reindex! as described, the index is cleared and replaced with the objects from only one of the two models.

I cannot reproduce this issue. Nowhere in the source does reindex! actually clear the index (it updates/deletes on a per document basis). Are you sure your primary keys are distinct? It could be that nothing is actually being cleared but simply written over.

Edit:
I also do not see how atomical indexing would fix this issue, I can only see how it would cause it. Please let me know if I'm wrong or misunderstood something, I am new to this project.

[brunoocasali]

I cannot reproduce this issue. Nowhere in the source does reindex! actually clear the index (it updates/deletes on a per document basis). Are you sure your primary keys are distinct? It could be that nothing is actually being cleared but simply written over.

How did you try to reproduce @ellnix? Can you share some code 😇?

Talking more about the way we do the reindex:

I do like the idea of using a delete index and then starting to index the documents again. However, doing so will imply a service disruption for the users right away. So, I believe the initial implementation tried to create a behavior similar to swap_index.

[ellnix]

How did you try to reproduce @ellnix? Can you share some code 😇?

Unfortunately I don't think I can find the code since I just kept editing it. I create two models Book and BlogPost, added an Author index to both of them and no matter what I did, and no matter the order I did it in, #reindex! on Book would not erase BlogPost documents and vice versa.

The only time when calling #reindex on BlogPost removed Book documents from the index was when their primary keys were not distinct (blog post with id 0 would override book with id 0, etc...). However, this is not a supported use case and the README specifically mentions having distinct primary keys.

At the end of it I resorted to going through the relevant parts of the source code and trying the meilisearch-ruby methods myself on an irb session.

I do like the idea of using a delete index and then starting to index the documents again. However, doing so will imply a service disruption for the users right away. So, I believe the initial implementation tried to create a behavior similar to swap_index.

If I understand it correctly, it would run into the same problems that index-swap runs into: #267 (comment).