bug(server): PyMilvus synchronous RPC commands freeze the ASGI event loop

Problem

In both server/app.py and server-https/app.py, the milvus_search() function executes PyMilvus gRPC calls (connections.connect(), collection.load(), collection.search()). Because these are synchronous, I/O-bound methods and they are called directly within an async def context without being wrapped in asyncio.to_thread or a thread pool, they block the entire ASGI event loop.

This causes the entire server to freeze under concurrent connections.

• Note: This is distinct from Issue #28 (connection pooling), which addresses how many connections are opened. This issue addresses what thread those connections run on to prevent event loop starvation.

Proposed Solution

Refactor milvus_search() to wrap the PyMilvus executions in await asyncio.to_thread(<blocking_call>, *args).

Related

• Distinct from #28 (Connection pooling)
• Distinct from #75 (httpx client lifespan)
• GSoC Pre-work: Part of codebase cleanup.
  • Tracker: #72

Start with milvus_search() in server/app.py and server-https/app.py, focusing on the synchronous PyMilvus calls named in the issue: connections.connect(), collection.load(), and collection.search(). Confirm that these executions are moved off the ASGI event loop, then verify that concurrent connections no longer freeze the server.