The default for max_ongoing_requests was changed to 5 in a recent release, which is the most likely issue. Try tuning it upwards like:

@serve.deployment(max_ongoing_requests=100)
@serve.ingress(app)
class FastAPIDeployment:
    def __init__(self):
        self.connections = {}

    @app.websocket("/ws")
    async def websocket_endpoint(self, websocket: WebSocket):
        await websocket.accept()
        # Connection handling logic

Reference docs: https://docs.ray.io/en/latest/serve/configure-serve-deployment.html#configurable-parameters