How severe does this issue affect your experience of using Ray?
- High: It blocks me to complete my task.
We wanted to use ray serve to replace our existing serving infra to serve XGboost model, our model is relatively simple, an xgboost model (the model.xgb is around 8.8MB). When we use fast api to serve this model, the memory consumption is about 780MB, however when we use ray serve, it consumes more than 5.8GB memory. There are a lot of ray::IDLE processes, each consumes around 70MB memory. Any advice here? Many thanks!