Is haystack features support multiple users in simultaneously?

[muazhari]

OperationalError: (sqlite3.OperationalError) database is locked [SQL: SELECT document.id AS document_id, document.created_at AS document_created_at, document.updated_at AS document_updated_at, document.content AS document_content, document.content_type AS document_content_type, document."index" AS document_index, document.vector_id AS document_vector_id, meta_document_1.id AS meta_document_1_id, meta_document_1.created_at AS meta_document_1_created_at, meta_document_1.updated_at AS meta_document_1_updated_at, meta_document_1.name AS meta_document_1_name, meta_document_1.value AS meta_document_1_value, meta_document_1.document_id AS meta_document_1_document_id, meta_document_1.document_index AS meta_document_1_document_index FROM document LEFT OUTER JOIN meta_document AS meta_document_1 ON document.id = meta_document_1.document_id AND document."index" = meta_document_1.document_index WHERE document.id IN 

and

ValueError: The number of documents in the SQL database (71692) doesn't match the number of embeddings in FAISS (0). Make sure your FAISS configuration file points to the same database that you used when you saved the original index.

These are the errors when multiple users using haystack simultaneously. Is there any methods to make haystack support multiple users in simulatenously?

[bogdankostic]

Hi @muazhari, can you give some more details about your set-up? In which cases do you observe these error messages? Either way, in our experience using OpenSearchDocumentStore works well with multiple simultaneous users.

[muazhari]

You can look at my setup in this repository https://github.com/muazhari/research-assistant-mini at sub_apps section. The error cases appear when the application is used by multiple users simultaneously but are fine if only one user is using it. I think this is a crucial matter to the haystack selling point of a "production-ready library" if there has something needed that does not yet work.

[bogdankostic]

I'm not sure if I understand the workflow of your app correctly. Whenever search is called, you reload the FAISSDocumentStore for dense retrieval and reload and even reindex the InMemoryDocumentStore for sparse retrieval?
This should not be needed, ideally you index your documents only once and the database is keeps running. I think that's also the reason why you get these error messages - you basically create a separate instance of a FAISS database with each call to search.

We've made good experiences with using OpenSearchDocumentStore for production uses. Also, you might want to have a look at our demos repos for examples on how to set-up an app in production with Haystack.

[muazhari]

I do reload/create new database instance for every search operation because the faiss document store can't infer the embedding dimension configuration from the retriever model. And, every search operation probably has unique configuration. Is there any method to automatically infer document store configuration from the retriever? Or the method to just using any document store with any changing retriever model seamlessly? If can't, is this a design defect?

[bogdankostic]

I think I don't quite understand your use case. Why does every search operation have a unique configuration? The workflow of adding search functionality is usually as follows: (After some experimentation about what configuration works best for your use case,) you decide for a configuration and create an indexing pipeline and a querying pipeline (these share the same retriever models). The indexing pipeline is used to add all your document to your document store, which is usually kept running. Then, the querying pipeline can be used to perform search on your indexed document store.