Async engine creation from bindings

[gorandp]

Hi,

Using create_engine_from_binding() creates a sync engine which when running queries blocks the event loop for async functions because it is not awaitable. Is it possible to make a create_async_engine_from_binding()? This way, it isn't necessary to use the D1 API to create an async engine. Or am I missing something and it isn't blocking the event loop?

Thanks!
Goran

[CollierKing]

Hi @gorandp, great question — we investigated this and can confirm that create_engine_from_binding() does not block the event loop.

What we found

SyncWorkerConnection uses pyodide.ffi.run_sync() internally, but run_sync() drives the JS event loop rather than blocking it. This means asyncio.gather() genuinely interleaves concurrent D1 calls — both create_engine_from_binding() and WorkerConnection show real speedup when queries are fired in parallel.

Proof

We added timing-based integration tests in PR #21 that fire 10 unique parameterized queries sequentially then via asyncio.gather():

• Sequential: 10 queries one after another (~1.2s total)
• Parallel: all 10 in flight at once (~0.15s total) — ~6-7x speedup

Both create_engine_from_binding() and WorkerConnection show the same concurrency behaviour. You can see the tests here:

• TestParallelQueries::test_engine_queries_are_concurrent
• TestParallelQueries::test_worker_connection_queries_are_concurrent

Bottom line

create_engine_from_binding() is safe to use inside async def fetch() handlers. A create_async_engine_from_binding() equivalent is not possible anyway — SQLAlchemy's async engine requires greenlet, which cannot run in Pyodide/Workers.

If you want raw async access without the SQLAlchemy engine layer, WorkerConnection is available:

from sqlalchemy_cloudflare_d1 import WorkerConnection

class MyWorker(WorkerEntrypoint):
    async def fetch(self, request):
        conn = WorkerConnection(self.env.DB)
        cursor = conn.cursor()
        await cursor.execute_async("SELECT * FROM users WHERE id = ?", [1])
        rows = cursor.fetchall()

Hope that clears it up!

[gorandp]

Hi @CollierKing, perfect! Thanks for checking this and clearing this up!