DAG memory usage

How severe does this issue affect your experience of using Ray?

  • Low: It annoys or frustrates me for a moment.
    Hi Community, I deployed a DAG with Ray Serve to the Ray cluster and then send 1000 requests to it, I found the RAM usage of DAG driver becomes very large.
    My questions are: Why the RAM becomes so large? How can I check which element consumes these RAM? How to reduce it?

Thank you in advance!

hi @Rui , thanks for trying out deployment graph ! Do you mind sending me a minimal reproducible code for your issue (like swapping out your model with dummy class / image with tensor of similar size, etc.) ?

Meanwhile we’re actively doing user POCs for deployment graph beta readiness and looks like you already started using it, we would like to schedule a call with you to hear your feedbacks to make it better :slight_smile: