feat(supabase): add async support to SupabaseGroongaDocumentStore and SupabaseGroongaBM25Retriever

[Aftabbs]

Summary

• Adds warm_up_async() and a full set of *_async() methods to SupabaseGroongaDocumentStore using supabase-py's AsyncClient and acreate_client
• Fixes SupabaseGroongaBM25Retriever.run_async(), which was a stub that delegated to the synchronous run() instead of using real async I/O
• Adds 14 new unit tests for the async path across both classes

Closes #3379

Root Cause

supabase-py ships two parallel clients: create_client (sync) and acreate_client (async). The original implementation only wired up the sync client. run_async() on the retriever simply called self.run(), blocking the event loop on every invocation — making AsyncPipeline usage effectively synchronous.

Changes

File	What changed
groonga_document_store.py	Added _async_client field; warm_up_async(), _setup_table_async(), and async counterparts for all public methods (count_documents_async, filter_documents_async, write_documents_async, delete_documents_async, delete_all_documents_async, delete_by_filter_async, update_by_filter_async, _groonga_retrieval_async)
groonga_bm25_retriever.py	Fixed run_async() to call _groonga_retrieval_async(); removed stale docstring note saying async is unsupported
test_groonga_document_store.py	Added TestDocumentStoreAsync class with 12 async unit tests using AsyncMock for the async client
test_groonga_retriever.py	Updated test_run_async to verify _groonga_retrieval_async is actually awaited (not the sync path)

Testing

• Existing sync tests pass (no sync code was changed)
• New async tests: TestDocumentStoreAsync — 12 tests covering warm_up, count, filter, write (OVERWRITE/SKIP/FAIL policies), delete, delete_all, retrieval, and uninitialized-client guard
• Retriever test: test_run_async_uses_async_retrieval — asserts _groonga_retrieval_async is awaited
• ruff check passes with no errors

Impact

Users running SupabaseGroongaBM25Retriever inside an AsyncPipeline now get true non-blocking I/O instead of a blocking call that defeats the purpose of async execution.

[davidsbatista]

@Aftabbs thanks for your contribution - can you please make the CI green so that I can start the review

[davidsbatista]

Regarding the tests, since we are testing against a local Docker instance no need to mock up; please convert them to integration tests

[github-actions]

Coverage report (supabase)

Click to see where and how coverage changed

File	Statements	Missing	Coverage	Coverage (new stmts)	Lines missing
integrations/supabase/src/haystack_integrations/components/retrievers/supabase
groonga_bm25_retriever.py
integrations/supabase/src/haystack_integrations/document_stores/supabase
groonga_document_store.py					136-137, 140, 168-169, 211-212, 222-229, 461, 476-477, 479-481, 486-515, 527-528, 537-541, 570, 580-593, 601-602, 608-609, 623, 633-634, 682-694, 731, 743, 761
Project Total

This report was generated by python-coverage-comment-action

[davidsbatista]

Hey @Aftabbs, thanks for the fix to run_async()

The new warm_up_async() method is inconsistent with how every other integration in this repo handles async initialization.

Looking at others, none of them expose a warm_up_async(). Instead, they use lazy initialization: each async method calls a private _initialize_async_client() at the top before doing any work, creating the client on first use. For example, from the Qdrant store:

  async def _initialize_async_client(self) -> None:
      if self._async_client is None:
          self._async_client = qdrant_client.AsyncQdrantClient(...)
          await self._set_up_collection_async()

  async def count_documents_async(self) -> int:
      await self._initialize_async_client()

[davidsbatista]

I have handled the remaining issues. Thanks for your contribution